[2026-11-16] eagle: work/zoom-transcripts/2026-10-20 Apple Weekly Team Meeting - Zoom ID 2026-10-20 20 01(GMT+6 00) gBBa-iiUQLycw55WRV1tnw/transcript.json work/zoom-transcripts/2026-10-20 Apple Weekly Team Meeting - Zoom ID 2026-10-20 20 01(GMT+6 00) gBBa-iiUQLycw55WRV1tnw/transcript.md work/zoom-transcripts/index.json work/zoom-transcripts/sync-status.json work/zoom-transcripts/undated All-Company Meeting bPprGg5DTye9H-RQXkwMuw/transcript.json work/zoom-transcripts/undated All-Company Meeting bPprGg5DTye9H-RQXkwMuw/transcript.md
This commit is contained in:
1 parent
0f6ec6ccbc
commit
e67448ffca
6 files changed
+5
-1091
No files matched your search
-670
@@ -23,676 +23,6 @@
|
||||
"time": "20:05:36",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Mm, got it."
|
||||
},
|
||||
{
|
||||
"time": "20:05:37",
|
||||
"speaker": "Tom Strba",
|
||||
"text": "And I think I'm next week. Because I almost got a heart attack when you said that I am…"
|
||||
},
|
||||
{
|
||||
"time": "20:05:40",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yes, this week is Diego. And sorry for not closing my task."
|
||||
},
|
||||
{
|
||||
"time": "20:05:41",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Okay."
|
||||
},
|
||||
{
|
||||
"time": "20:05:45",
|
||||
"speaker": "Tom Strba",
|
||||
"text": "that I am this week."
|
||||
},
|
||||
{
|
||||
"time": "20:05:45",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "This week… Why is Diego… I don't see a task… this is a macOS release we're talking about, right?"
|
||||
},
|
||||
{
|
||||
"time": "20:05:52",
|
||||
"speaker": "Tom Strba",
|
||||
"text": "Yeah, I think Diego maybe marked it as complete in the beginning of the week, because there's just, like, triage…"
|
||||
},
|
||||
{
|
||||
"time": "20:05:57",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Yes, you're right, Tom, your block is the 26th to the 30th, because that task is missing, I didn't see it, I wasn't looking at the dates, I was just reading the list."
|
||||
},
|
||||
{
|
||||
"time": "20:06:04",
|
||||
"speaker": "Tom Strba",
|
||||
"text": "That's okay. Just wanted to be sure."
|
||||
},
|
||||
{
|
||||
"time": "20:06:07",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Okay. Diego is responsible for the release this week. It sounds like he's actually doing that. Is that right?"
|
||||
},
|
||||
{
|
||||
"time": "20:06:08",
|
||||
"speaker": "Tom Strba",
|
||||
"text": "Perfect."
|
||||
},
|
||||
{
|
||||
"time": "20:06:14",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "He knows. Okay, cool. Thank you."
|
||||
},
|
||||
{
|
||||
"time": "20:06:14",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "That's right."
|
||||
},
|
||||
{
|
||||
"time": "20:06:18",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Um, no problem. Okay. Anyway, think about starting your self reviews this week. If you feel like you're gonna need more than a day to knock it out. Bartek, over to you."
|
||||
},
|
||||
{
|
||||
"time": "20:06:31",
|
||||
"speaker": "Bartek Waresiak",
|
||||
"text": "Yep, I will be stepping down as IS app DRI. Jacek will take the primary role, and Brindy will take on secondary role. So, yeah, if there is IS drama, if there are IS issues, if there are any urgent things. Go to them! And thank you very much."
|
||||
},
|
||||
{
|
||||
"time": "20:06:57",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "All right. Thank you. An, you're next on the agenda."
|
||||
},
|
||||
{
|
||||
"time": "20:07:08",
|
||||
"speaker": "Anh Do",
|
||||
"text": "Oh, that was quick. Yeah, so in OK, there have been a few projects that involve native changes required as part of the front-end-led projects. Um, so the background is that they, uh, want to contribute some PR, and, uh, um… AI assisted Pi, of course, and then what have been working for us lately is like depends on the size of the task like I would be the one to like either. Dri the change that they are they want to make, or like they prepare the Pi, and I do a thorough review of that. So and now we want to kind of do this more regularly. And so there was a need for some kind of like guidelines that we can give to contributor from outside the apple team. So I was just starting like a small task to basically collect ideas from people. And if you have anything that we already have somewhere, maybe I could get put in. And so. it becomes something that we can point other people to when they want to contribute to the Apple codebase to ensure our standard and make sure that, like, everything is being done the way that we want. So that's about it. Nice."
|
||||
},
|
||||
{
|
||||
"time": "20:08:19",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Cool, thank you. I just took a quick peek. I see you do mention the engineering expectations. I think that's a good source. We might also want to link… to the onboarding process. Not that they would need to follow all of it, but some of… that's basically our best source of, like, how to get set up with the codebase. So there's probably a little work to do to figure out how to have a slimmed down version of that that folks could follow. And I'm sure other people have other ideas as well. So thank you. Um… Luke, we have a non-Apple person lurking in our midst. Welcome. What can we do for you?"
|
||||
},
|
||||
{
|
||||
"time": "20:08:54",
|
||||
"speaker": "Luke Andrew",
|
||||
"text": "Hello. Yes, thanks for having me. Luke from SRE, for people that I haven't met before. Um, this is just a plea for generic Apple help. I don't want to, sort of, get all of your attention to focus on this task. We basically have… a looping client that seems to be sending, uh, requests once a second to us, uh, that is Apple, but not DuckDuckGo Apple, based on the user agent, and doesn't seem to be Safari either. Um… And it's cumulatively across lots of different devices, lots of different queries, sending in the region of 3 million searches a day. Which is substantial, not sort of catastrophic, but substantial enough searches that it's something that we want to deal with. Umm. And I don't really know where to go next to try and understand where it might be coming from or how we might try and reproduce it. So I don't want to solve that in this call, but I just wanted to bring your attention to the task. And if there's anyone who thinks they might be able to help, I'd really appreciate. a second pair of eyes, and pointed out that it's not even worth, with what we have at the moment, that we can, like, escalate this to Apple, because we just don't have enough, uh, information to get any help from them. So yeah, that's it. If anyone has any questions immediately or async, just pop them on the task. Yeah, that's all I have."
|
||||
},
|
||||
{
|
||||
"time": "20:10:18",
|
||||
"speaker": "BA",
|
||||
"text": "Quick one, Luke."
|
||||
},
|
||||
{
|
||||
"time": "20:10:18",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "This is not from a single client, but from a particular app version, potentially, is that right?"
|
||||
},
|
||||
{
|
||||
"time": "20:10:24",
|
||||
"speaker": "Luke Andrew",
|
||||
"text": "Yeah. Yeah."
|
||||
},
|
||||
{
|
||||
"time": "20:10:24",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Okay. You know, be scared that it was one client going crazy. Okay, um…"
|
||||
},
|
||||
{
|
||||
"time": "20:10:30",
|
||||
"speaker": "BA",
|
||||
"text": "Yeah, I was going to ask, what's the — we eliminated bots from the equation. Are we happy that it's not bots?"
|
||||
},
|
||||
{
|
||||
"time": "20:10:36",
|
||||
"speaker": "Luke Andrew",
|
||||
"text": "It doesn't look like bots. The queries all look human. The rate is really low compared with bots. The moment that I put in a block for it and it immediately stopped, like redirected away."
|
||||
},
|
||||
{
|
||||
"time": "20:10:39",
|
||||
"speaker": "BA",
|
||||
"text": "Okay. Right."
|
||||
},
|
||||
{
|
||||
"time": "20:10:52",
|
||||
"speaker": "Luke Andrew",
|
||||
"text": "It got, like, a 302 from deep, and then just immediately stopped, so that doesn't feel like… bossy behavior either. They normally persist a little bit and then give up. Whereas this was just immediately, it broke out of whatever loop it was in and stopped."
|
||||
},
|
||||
{
|
||||
"time": "20:11:09",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Interesting."
|
||||
},
|
||||
{
|
||||
"time": "20:11:10",
|
||||
"speaker": "Luke Andrew",
|
||||
"text": "Uh, so don't all take a look, because that would be a waste of time, but if you think you can help, um, I'd appreciate a pair of eyes. I'm gonna drop off. Thank you all for having me."
|
||||
},
|
||||
{
|
||||
"time": "20:11:19",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Yeah, one last quick question. It sounds like you're treating this as something that definitely needs attention, but not an incident at this point. Is that correct?"
|
||||
},
|
||||
{
|
||||
"time": "20:11:25",
|
||||
"speaker": "Luke Andrew",
|
||||
"text": "Yeah, that's correct. It's been going on for… Based on… um, how we categorize queries for, um, like, ads perspectives. These get classified as a separate category based on user agents, so we don't send them to Bing and say, these are definitely good queries. Um, it's been happening for about a month. So it's, at the moment, not urgent. I'm gonna raise it with SRE, maybe, and see if we think it should be an incident that we just let run, but at the moment, that's not how we'."
|
||||
},
|
||||
{
|
||||
"time": "20:11:55",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Okay. Great. Thank you for the clarity there. Uh, and yeah, if anybody thinks they can help Luke out and figure out what's going on there, feel free to hop into the task. Thanks for coming. Moving on. Alex, you're first in the regular topics list."
|
||||
},
|
||||
{
|
||||
"time": "20:12:16",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah, hey. I have several topics for today, including the VS Code integration postmortem to run later. And the first one goes with a proposal. The proposal is pretty short. And kind of related to the, uh, recent incident with Mac OS 12 crash. that we are not. Testing for macOS 12 and 13, I guess, because we are missing the infrastructure on CI runners. And the suggestion is pretty simple to extract the macOS UI tests target from the main macOS Xcode project, leaving it as a separate macOS project inside the. uh, Xcode workspace. So it will be available the same way it is now, and, uh, runnable. So only, uh, the only issue I can see for those who use, uh, the actual macOS Xcode project instead of the workspace. they would have to switch to another project to run the tests. So, I just wanted to hear if there is any concerns or objectives against it."
|
||||
},
|
||||
{
|
||||
"time": "20:13:39",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I think some time ago, we… someone asked… who uses or doesn't use Workspace, and people who didn't use Workspace were in a… In a minority, anyway, like, very few of them, so… I guess… I could move to the workspace."
|
||||
},
|
||||
{
|
||||
"time": "20:14:06",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "I think it'd be good to ask that question in that proposal, at least more explicitly, and just give people a chance to chime in if they feel like that would be a hurdle. Um… Other question for me is, is there… should we think about the iOS side of this as well, and to match… structure, or is this really a macOS only question?"
|
||||
},
|
||||
{
|
||||
"time": "20:14:26",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Okay. Maybe, I… Didn't think about it, actually. And the main concerns here is the compatibility. So we can reach macOS 12 compatibility, at least for some local runners that I'm going to present further. And the build times, so we have package fetching and building for every macOS runner that runs the UI tests and any of the dependencies. None of the dependencies are used during the runs, so it's a 10-second step for each runner that can be easily omitted, and so, yeah. And yeah, it's not related to the main browser targets, so there's no risks. Uh, and they're both, uh, U.S. Probably to think about it in the ER."
|
||||
},
|
||||
{
|
||||
"time": "20:15:25",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I think iOS is not…"
|
||||
},
|
||||
{
|
||||
"time": "20:15:25",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "I mean, I don't have a strong… sorry, I don't have a strong feeling there, I just felt like if we're gonna make this change, like, if it's gonna end up with different structures depending on the app, we should think about whether we really want that to be the case, or if we should do the other."
|
||||
},
|
||||
{
|
||||
"time": "20:15:37",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "No, it actually shouldn't change any structure, any file structure, just the project placement."
|
||||
},
|
||||
{
|
||||
"time": "20:15:38",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Screwed up. Yeah, yeah, yeah. Dominic, I interrupted you. I'm sorry."
|
||||
},
|
||||
{
|
||||
"time": "20:15:44",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Ah, no worries, I just wanted to say that iOS is not affected by this at all, because… iOS doesn't need testing on various macOS versions."
|
||||
},
|
||||
{
|
||||
"time": "20:15:51",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Alright. So the limita… this is about dealing with a limitation that comes from Mac OS 12 build, particularly."
|
||||
},
|
||||
{
|
||||
"time": "20:15:59",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yes."
|
||||
},
|
||||
{
|
||||
"time": "20:16:01",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "And bill times."
|
||||
},
|
||||
{
|
||||
"time": "20:16:01",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I'm Bill Times."
|
||||
},
|
||||
{
|
||||
"time": "20:16:02",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Got it. Okay. thanks for pointing it out. Uh, like I said, I think you might want to give people a little bit more time to respond to async if they have… particularly if they feel like the… Workspace thing is gonna be a hurdle, but my guess is it's not gonna be a big issue. Uh…"
|
||||
},
|
||||
{
|
||||
"time": "20:16:22",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Okay."
|
||||
},
|
||||
{
|
||||
"time": "20:16:24",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Next topic."
|
||||
},
|
||||
{
|
||||
"time": "20:16:24",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "We can…"
|
||||
},
|
||||
{
|
||||
"time": "20:16:26",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "From you, also."
|
||||
},
|
||||
{
|
||||
"time": "20:16:26",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Perfect. To the next topic then. Yeah, so, uh, this is another iteration of, uh, the Verfield project. I've been live coding for a while already. Uh, so this is the current state of things. It's changed a little bit, some of you may remember the first, uh, first demo. Uh, so the change is it's, uh, migrated to Go language, and, uh, now it runs on TART system that's used on, uh… We tried runners, I guess. And so the system is here to show you the available images, available runners and locations. So here you see I can have it. stored on the external SSD, because I have my internal 2TB ETH. up to completely, and I have the old MacBook Pro also managed from the same environment. And, uh, you can add other machines as well. Uh, so this is, uh, a wrapper around the TART system that manages, uh, Xcode installs, uh, SIP protection, and, uh, for virtual machine and real machines. Uh, if you use older hardware and, uh. It's, uh… basically providing MCP server for AI clients to. to run these UI tests. Mostly. on the older virtual machines. It can be not older virtual machines, but actual virtual machines. New refreshable machines as well, sir. Uh, for example, I don't update to macOS 27, because, uh… I expect some issues, but I use, uh, testing on macOS 27 using this, uh… a virtual machine system. And so basically the… Again, I have a question to the team in… In the light of a recent discussion of testing on order. macOS versions, for example, whether should I try to push it forward to, like, team-wide? So the idea… the idea is to be able, uh… again, we'll… And in the post-mortem. for the VS Code integration projects. So here you can see VS Code. Basically, it's running the Google browser. It can… run iOS browser as well, in simulator, and in addition, it can also run remote debugging, unlike Xcode, which had supported remote debugging. in a long time ago, but currently lacks the ability, so… Integrating these systems together allows, like, click one button. choose the virtual machine you want to run to test on, and it builds, deploys, and debugs in real time without any lags for library symbols transfer and other stuff. So it just works like you would do on your host machine, but in the virtual machine. So I wanted to hear opinions whether should be. try to adopt this team wide."
|
||||
},
|
||||
{
|
||||
"time": "20:20:06",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I can offer an opinion."
|
||||
},
|
||||
{
|
||||
"time": "20:20:09",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Nobody wants to."
|
||||
},
|
||||
{
|
||||
"time": "20:20:09",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "uh… No, well, I mean… No one to be the only one commenting, but anyway. This will be the topic of, uh, of, uh, of the upcoming postmortem, uh, as well, that… The Apple engineering expectation of testing on older macOS version is essentially a dead low. We are supposed to test on older macOS versions, but we rarely do it. Perhaps for new features on macOS? I'm not sure. But certainly not for… Small… tasks. I don't think… well, I don't know, but I don't think we bother to test on all the ROS versions, every small version. Change to the UI, especially. And, uh… Because of that, if we had a way, and also we can't, we are unable to test in CI. So, if… I had a way to… Ask an agent to… run UI tests on a virtual machine. on an arbitrary OS version, that would be great. I would like to see the demo of that, actually. That would be awesome. And that is why we might be interested in adopting it. Perhaps."
|
||||
},
|
||||
{
|
||||
"time": "20:21:41",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah, again, about the UI tests, correct point. So sometimes you want to run an agent reproducing an issue, for example, and it's not always convenient to upload it to CI, wait for the UI tests to be run. Look for the results, while if you run it locally on your machine, it will be blocked for 5 minutes, and do the stuff while you take a coffee. So this thing allows you to just run directly from the agent on the same machine in background and provide you screen recording with all this and the results."
|
||||
},
|
||||
{
|
||||
"time": "20:22:22",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Yeah, I mean, I think… You know, your point that our expectations are that we'd be testing on all versions that we support, but in practicality, I think almost nobody actually does that. And particularly with something like as old as Mac OS 12 where we don't have the hardware, we don't have like, most people don't have access to test it to your point, L. Um… I think I'd like to understand a little bit more about how the system would work, what kind of third-party dependencies there are, what it would look like to really set up and what it would cost. I think the idea of we would be able to test. On versions that we say we actually support is pretty appealing."
|
||||
},
|
||||
{
|
||||
"time": "20:23:00",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Okay, so probably the resolution here would be, um… So, first of all, I do use it as my self-made tooling, and uh… if… I can reach it to the point, I can make it to the point when you click a button, install the system, click a button, and it just runs. I think it would be worth. adopting at least as a selected auditory who can. have, uh, the means to testing, like, enough of RAM or some older hardware, uh, like, choose the early adopters and, uh, see how would we… expand it. And, uh, yeah, correct question by, uh, Federico about, uh, older ores. It's another topic that I wanted to raise during the post-mortem. During the incident, actually. Both for the incident, uh, that, uh, having… Maybe one dedicated machine, older machine, maybe somewhere that would run the UI test suite at least once per week for each release would be a great addition, even if it's like a physical machine somewhere. someone's office or home space."
|
||||
},
|
||||
{
|
||||
"time": "20:24:28",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Yeah, that's an interesting idea, too. I think… I guess I think that a proposal here that sort of lays out what a strategy for old testing would be that used this would be maybe a useful next step. Um, because then that would be a place that people could rally around, sort of solving that higher order problem of how do we actually make good on our promise of testing on the systems that we support. And it sounds like this would be one, but maybe not the only tool to consider, or maybe even as a combination. Does that make sense?"
|
||||
},
|
||||
{
|
||||
"time": "20:25:00",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yep."
|
||||
},
|
||||
{
|
||||
"time": "20:25:01",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Cool. Okay, thank you. Interesting stuff. Sam, you're up next."
|
||||
},
|
||||
{
|
||||
"time": "20:25:10",
|
||||
"speaker": "SS",
|
||||
"text": "Yeah, recently I have been chatting with some folks about being able to put Claude workflows in our CI. I talked about this with OT and they told us that they don't want Anthropic tokens being available in public repos. I've created a private repo to host the agents, uh, which is… Pretty bare bones right now. It lives here in the Apple Team Agents repo, it has CI set up, it has the Claude, uh, action set up. I have a workflow first for testing it, uh, just so we can make sure you can actually talk to Claude and back. Um, the only thing I'm waiting for is whether or not we should get a dedicated API key, because I've been told. Two different things from OT. So I'm just waiting to —. Get that sorted out. I think we should get a dedicated one. once we've done that, yeah, we can start doing things like an automation to clean up feature flags and such, like Fernando's, uh, proposed. So, uh, I'll post to Apple Diary once that's ready, but I think you'd be… if you do want to start building automations in this now, like, feel free, uh. As long as it's, you know, not super expensive. Pete, done."
|
||||
},
|
||||
{
|
||||
"time": "20:26:19",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "Yeah, thanks, Sam. This, this looks really cool. Just because I've seen, um, like, there's a few different conversations. There's a conversation with this, and there's the Bitrise, you know, running, um. uh… Agentic stuff on Bitrise as well. How does it all fit together? Like, I think I understand it, but, like, say I wanna… say we wanted to, you know, uh, build our feature overnight while we're in bed, um… How could we use this to do that? Or can we?"
|
||||
},
|
||||
{
|
||||
"time": "20:26:43",
|
||||
"speaker": "SS",
|
||||
"text": "This is not that. This is definitely not that. This is not like the development environment thing. Yeah, these are not like the development machines. I really just expect these to be like GitHub Actions workflows that just do helpful things with Claude. Uh, and we just need a place to put those. Like, if it weren't for the API key limitation, this would live in the monorepo."
|
||||
},
|
||||
{
|
||||
"time": "20:26:56",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "Yep. Yeah, yeah, I guess, I guess what I'm wondering is, so, like, yeah, exactly, we can, we can hit Claude and, um, but, you know, Claude can't, well, build for us, so, like, well, build and check that it's actually working, so where would… say, say even, like, removing feature flag, where would that. On the PR, at the PR station."
|
||||
},
|
||||
{
|
||||
"time": "20:27:17",
|
||||
"speaker": "SS",
|
||||
"text": "Yeah, this is something that Fernando and I was talking about, about doing some kind of coordination between these repos where the monorepo would talk back and forth, do the things that it needs to. This can create a branch. The monorepo can then do the actual testing. Yeah. Right now, I'm just using these for, like, dependability analysis kind of problems, or planning on anyway, which I have a demo for coming up soon."
|
||||
},
|
||||
{
|
||||
"time": "20:27:33",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "Okay. Yep. Cool. Yeah, cool. Thanks."
|
||||
},
|
||||
{
|
||||
"time": "20:27:44",
|
||||
"speaker": "SS",
|
||||
"text": "Dominic."
|
||||
},
|
||||
{
|
||||
"time": "20:27:46",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Mm-hmm, uh… one of the limitations here is that GitHub Actions workflows are limited to 6 hours of runtime. So…"
|
||||
},
|
||||
{
|
||||
"time": "20:27:53",
|
||||
"speaker": "SS",
|
||||
"text": "Also that, yeah."
|
||||
},
|
||||
{
|
||||
"time": "20:27:54",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yeah. And also, it's, of course, non-interactive. Um… Which is why we're currently trying to figure out if Bitrise… or Bitrise or not, but anyway, looking for some infrastructure provider for, uh… macOS machines that could run Agentic workflows and be… and offer. interaction with them, right? So, uh… Bitrise RDE remote. development environments is one of the options there. Um… That is in progress. For now. They also… it's also quite a fresh feature on their end, and they have some bugs there that we're trying to resolve right now. And we have a CloudOps team member on that as well. the side of… Connecting them to ClickHouse or Sentry. Yeah, so… We'll see in some time."
|
||||
},
|
||||
{
|
||||
"time": "20:28:58",
|
||||
"speaker": "SS",
|
||||
"text": "Yeah, yeah, to Dominic's point, like, that problem is much bigger than what this repo is. This really is just, I want to be able to run Claude on an automation, like I currently do on my own machine, and I'm just sick of doing it on my machine. I just want to push that up to GitHub Actions. Have it run on its own, and… Show up with PRs and Asana tasks occasionally, so. Okay."
|
||||
},
|
||||
{
|
||||
"time": "20:29:19",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "So the limitations here are that it can… It can inspect code, it can run as an agent, but it can't build and run, so it would… prospectively submit a pull request, and then CIA would tell you whether it did the thing or not."
|
||||
},
|
||||
{
|
||||
"time": "20:29:32",
|
||||
"speaker": "SS",
|
||||
"text": "Yeah, the idea that I'm thinking here is that, yeah, like PRs would go over back to the monorepo. This thing would open PRs against the monorepo. And then CI would do its thing over there. Yeah."
|
||||
},
|
||||
{
|
||||
"time": "20:29:38",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Right. Yeah, and if we needed to inspect it from there, we could pull a branch and do whatever else we needed."
|
||||
},
|
||||
{
|
||||
"time": "20:29:46",
|
||||
"speaker": "SS",
|
||||
"text": "Yeah."
|
||||
},
|
||||
{
|
||||
"time": "20:29:47",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Okay, cool. Definitely sounds like a good step, even if you're not trying to solve everything all at once, which is reasonable. You're the next topic as well, Sam."
|
||||
},
|
||||
{
|
||||
"time": "20:30:00",
|
||||
"speaker": "SS",
|
||||
"text": "Oh, is this the demo? Probably should have just done this all at once. Yeah. So one of the things that I."
|
||||
},
|
||||
{
|
||||
"time": "20:30:04",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Oh, before you… sorry, if you're doing demos, before we get into demos real quick… Uh, we had a note here that we were gonna actually record demos. Um, but we haven't ever done that. Mostly people have opted out of wanting to be recorded, so I wanted to just check in since we have 3 demos here. do we want to record these? Should we rely on the… just the transcript summaries at this point, and whatever tasks people have written? I'm kind of okay either way. Um… I don't know. I don't know that we've ever recorded a demo."
|
||||
},
|
||||
{
|
||||
"time": "20:30:35",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "Alessandro. Alessandro is waiting for his Wednesday morning. Won't be the same if we don't record our demo."
|
||||
},
|
||||
{
|
||||
"time": "20:30:39",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Right. That's the question, right? I know, yeah, it's like this is sort of a hard way to do this. So, Sam, would you like me to record or not?"
|
||||
},
|
||||
{
|
||||
"time": "20:30:51",
|
||||
"speaker": "SS",
|
||||
"text": "I mean, I've already shown this to Alessandro, so I don't think he cares. Yeah. Up to you. I'm good either way. I remember. Don't worry. This is a super quick one anyway, just to show the initial problem that I'm trying to solve with… with a repo like this. Let me share my screen. Um, so I have, for a while, wanted the other half of the flaky tests project. Right now, we have a flaky test project, which keeps track of, uh, you know, tests that are failing. Um… I have traditionally been more interested in the problem of what are the workflows that are failing, uh, and why, because these things were just, like… this stuff comes up a lot, like, during Xcode upgrades in particular, so… This is something that is being used by Claude to pull the monorepo, get the actions, go and look at the code, figure out what's causing a problem, and report back and give a little write-up about what's going on. Um, the idea being that eventually you could then move something into this ready-to-fix section, and the agent would then go and actually open the PR later. Um, you know, this is just moving a… moving an automation that I've had on my machine is like a daily, like, 10 AM workflow for a while. and just putting it remotely so I don't have to think about it anymore. Which is the ultimate goal. So yeah, this is the kind of thing that I'm trying to do with this repo. Um… As an aside, OT is kind of working on… something along this as well, which I learned about as I was setting up this repo. They are setting up this template, which lives on Dub, for what it's worth, for these kind of analyzing report workflows. Umm. So that might be interesting if you are needing like access from things within the system already. Uh, like, I think Lucas is currently looking at getting that thing to have sentry access and all this kind of stuff, so… Yeah. Go check that project out if you're interested. That's it."
|
||||
},
|
||||
{
|
||||
"time": "20:32:48",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Awesome. Thank you. Pete, you're next."
|
||||
},
|
||||
{
|
||||
"time": "20:32:57",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "Yeah, thanks. Yeah, this is in the demo section. It's not really a demo, I guess, but just wanted to share. I'll speak a little bit about, um. ASTAC, so I put a link in… a link into the chat there, so I'm sure people have heard… of… PSTAC, or maybe people have heard of PSTAC. I'm going to just share my screen. So if you haven't, um… Pstack is a plugin created by an engineer at Cursor or SpaceX. It's basically just a set of skills that this engineer found were very good. at, um, fixing bugs, implementing features. So they created the PSTACK plugin. What ASTACK is, is basically just a port of PSTACK. So PSTACK was. built for cursor. So it has a lot of built in you know, instructions in terms of cursor models and how they work so forth. So I just. Ported PSTAC, uh, over… To, uh, ASTAC, um… Basically replacing any of the cursor references with Cloud Code and Codex references. So what AStack is, is a plugin that you can use in Codex or Cloud Code. It's like it sounds very abstract. It's not. It's really just a. A set of skills, a set of playbooks, some defined principles. And they all work together to what I found to just to produce good output. um… So I'll just, I'll talk a little bit in a moment about how ASTAG works. So how it's set up at the moment, you can see on the left, I just have it there under my public, under my own public repo. But if anybody wants to try it out, I would encourage it. You can see like I didn't do a lot of work here. I really want to undersell this. asked an agent to port this over. So take PStack, port it, make sure it works with Cloud Code and Codex. The most work I did was creating that logo you see on the left of the screen, so that's where all my creativity went into. Um… But yeah, you can go here, you can grab the plugin, you can install it in Cloud Code. And you can try it out. I'd recommend trying it out for… for bug fixes, even like small features that are well defined. So, you know, the evidence, like any evidence I have is just anecdotal. I've just found these particular set of skills, these set of playbooks and the principles that are defined just produce very good output. Again, anecdotal. I haven't measured this. I also think it seems to do it at lower token usage. But again, I haven't measured it. So how it works, how ASTAC or PSTAC work is, you know, you install the plugin, it sets up a hook, so at the start of your sessions, the agents are aware that ASTAC is there. And if you say, hey, build this feature or fix this bug or whatever, you can explicitly say. Enter potato mode. But that's actually part of the hook. The agent will know, oh, if you give it a feature or a bug, it will enter potato mode. All that is like a top level skill, which is basically a router. And depending on the task that you give it, like fix this bug or implement this feature. You will go and find the right playbook and follow that playbook and in the playbook. That would reference what skills to use, et cetera. So. again, you know, there's nothing, like, there's nothing magical here. I mean, there's been other plugins that have been very popular. I was talking to Fernando earlier, and we talked about… was it the Superpowers one? I think was another popular plugin. So things like this will probably be, as I was speaking to Fernando about, they'll probably be, can, you know, eventually end up in the harness. Probably won't need plugins like this at some point. But at the moment, this is something I'm using daily. And I find it very beneficial. It just results in really good output. So I would encourage people to try it out. Like, what what, you know, we've seen probably saw the post from Sebastian, from OM. He's creating, like, auto, which is their version of pstack. He just took pstack. And ported that and kind of made it very specific to Duck AI. That's like the the OM version of Pete's group. So if ASTAC, if people find it useful, we could basically turn that into like the Apple version of PSTAC, Apple team version of PSTAC. and… that's something we could do. So yeah, I would just encourage people to try it out, if you're interested. There's an Asana task there that I posted, and you can give any feedback, and if people want to. People find it useful and want to contribute and make it better. We can do that. It's under AI enablement AOR. I spoke for a long time, if anybody has. Sorry, I shared that artifact. Sorry, key. I didn't go through all this artifact."
|
||||
},
|
||||
{
|
||||
"time": "20:37:59",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "It's linked in… The Apple Weekly Task, so you can go and look at that artifact, it explains at a very high level how it works. Again, it's not magic, it's just a skill, a ruler, lots of other skills. Sorry, my dog is barking. I don't know if you can hear my dog, but. He's over there barking at me. That's very low. That's it. If anybody has any questions, you can ask me questions, or you can ask me questions later. Jane, sorry, your hand up."
|
||||
},
|
||||
{
|
||||
"time": "20:38:35",
|
||||
"speaker": "JF",
|
||||
"text": "Yeah, sorry if you said already and I missed it. What changes have you made over the original piece, Doug?"
|
||||
},
|
||||
{
|
||||
"time": "20:38:42",
|
||||
"speaker": "Pete Smith",
|
||||
"text": "Yeah, so… Oh, I'm sorry, it looks like. Okay, dog went dead. Um, yeah, that's why I wanted to undersell it. Like, I didn't, I didn't, I didn't do anything. So, uh, I created that logo. What, what, what I did, or what the agent did, was just port it over, support Pstack over, um, for Cloud Code and Codex. Like, the original Pstack has references to Cursor, and how Cursor. Works. So I just ported it over, so it works in… Um, codecs and cloud code. I didn't add anything additional yet. Um… But the idea will be we can add things if we find them useful. We can add Apple-specific things. If we find them useful. So yeah, no changes really. Sorry, one more thing, James. Thanks for reminding me. I'm going to reshare. Sorry, everybody. Very annoying. I'm going to reshare. So a good question. Like, PStack is being continuously updated. So I have a scheduled task that runs every Monday, looks at the latest PStack, and consumes the latest changes, and, like, ports over the latest changes. So I keep it up to date. So I look at the upstream and keep. a stack up to date. But that's other than that, I just I just port it. I don't make any other changes. Like, you know, of course, there's a chance it's not ported well. It might not work properly, but the intention is it's just a direct port."
|
||||
},
|
||||
{
|
||||
"time": "20:40:09",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Cool. Thank you, Pete. Paul, I'm protesting."
|
||||
},
|
||||
{
|
||||
"time": "20:40:16",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "Yes, thank you. I'm going to share my screen. This is pretty much. a demo that I owe you of the snapshot testing. So, quick reminder, this is already available on iOS, not yet on macOS, because we are waiting for macOS 27. runners on big drives on the CI. on, uh… Yeah, pretty much. It's a package that has a very nice README explaining of it, what I will show today. So you can read it or tell your favorite agent to read it. But it has like requirements, how to use it. A bunch of examples for both iOS and macOS, even though, as I said, macOS. is disabled. But let's do a bit of an example. if I find the Xcode that just crashed, apparently. Ah, it's here, yeah. So, uh, how does it work? We start with the bill. Maybe, or maybe not, you will have previews. That is the new, like, hashtag… preview macro but for a snapshot so I created this one you don't need to use previews but it's just if you use them it becomes handy so there is this new preview provider that's the system. protocol, then if you conform it like this, so pretty much. You configure the different states that you want to have in the view. In this example, it's this segmented control. So there are two previews, one for each tab, let's say. You configure it here. So that's the model or the state of the view. And then you configure it here. So it will render one of the previews for each tab in this case. That basically gives you the previews here, but what is nice about it is that then you have the snapshot test. becomes basically a one liner or yeah one four liner uh but yeah just one call because it reuses the same previews that you refine it and then uh this is just a unit test like. Any other that we have. that runs the test. And this. Basically, you can configure it in different ways, for example. Here, if you want, um… to have a preview, but you don't want to snapshot it, because it's not needed. You have different. uh parameters here in this um. Let me show you. In these configurations, so there is… a scope where you can say, I want just for previews, just for snapshots, or for all of them. And then, um… in the test, you can snapshot in a different set of, um, configurations. So, in these asset snapshots. But everything is here in this package. We have different things. We have the different strategies. If you want to snapshot, for example, iPhones, iPads, all the combinations, just a single screen or a custom. Uh, size? And then you have also. the different sizes of components that you can have in an app, right? Majority of things are the screen, but you can also test some intrinsic content size, bottom sheets, or just any big size that you want. So all of that has, like, default parameters. In this example, for example, it's just, you know, normal screen, and I want to judge a snapshot in that mode, because that screen doesn't have light mode. So, it would work like this. And this is a unit test, it runs super fast, and uh… It passes in less than a second. how it works when you change something so let's say remember we have this previous let's say now i change this uh. this view. No, there is a design change, so we have this title here. If I run the tests, now they will fail. I will show you how it looks like in Xcode. So pretty much, issue recording, blah blah blah, does not match reference. And then, you can go here in your tests. We see the test that failed, and if you double-click, there is this gallery. That basically is showing us the reference, the failure, and the difference. So you can preview it like this. That's the original one. And then the new one that got recorded is this one with hello."
|
||||
},
|
||||
{
|
||||
"time": "20:45:21",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "So you can see a diff that it's generated. Basically, you see everything is moving down. Um… So here it can happen two things. It's a real failure and you need to fix your UI, or it's a design change that is… expected, and you want to re-record the test flight, because this is now the new correct design. To do it, you have different ways. Here, there is a record Boolean. You can pass it to true. But at the same time, instead of changing code. If you go to the folder of the test. In in in Git you can let me show you. in a… in another screen. You can just delete the reference image in the file system. So that's the submodule that we have. In the, um… With all the snapshots. So if you go there and delete the file, it will recreate it on top of it. Um, so yeah, basically you have these two options. If there is no original file in the file system, it will just create it right away. If there is an existing one, it will do the comparison and so on. So the easiest usually is, um, to just remove. the image and re-record it. But let's say that… This works, and you upload a PR, and it fails in the CI. So how does it look like? Pretty much the current PR checks that we have for iOS, we'll report them here with the faillet, and it generates an HTML report, which is this one. that, uh… creates the same thing that we see in Xcode, but in the CI. So if you see a PR failing, you don't need to download the artifacts and so on. You just have to open this, and it has, like, these drop-downs. with the same concept, like the reference, the failure, and the diff. Um, so that's… A bit what I wanted to show. Then, um… So the snapshots live in a submodule, as you probably know. So it's a different repo. It's these DuckDuckGo Apple browser snapshots. And it just contains the PNG files. So that is this. Orchestration, if you want to run when you create snapshots, you need to open, let's say, two PRs. But there is a script that handles everything. So what you would do is here when you record your new snapshots. Maybe I can do it really quick. Just a second. I will. Show you exactly what I mean. So we have the submodule. Snapshot references. I'm gonna remove… The, um… Tests for this one. which is the scan QR code test. So I remove it from the file system. Okay, from this folder. And then when I record it or when I run the test, it will record. So first time it fails, but now it says no reference found on disk, which is fine. And then second time it runs, it already has the references and it passes. Okay, so now we want to open the PR, uh, and we can see that the snapshot reference module has the changes. Again, old and new with a new title. So there is a script in the scripts folder. around doo doo doo scripts. called Open Snapshot Zoom Module PR. You just run this over there, runs it, and it will open the PR, so commit, push, open PR, and so on. Um, and then once in your main branch or main PR, there is a new label that you can tag. Uh, let me show you here. Merge the snapshots. So this will merge the snapshots PR and update the pointer and so on. So, it's these two extra steps. You can do it manually, but, uh, the easiest is once you record it, and you have the changes here in Git, you just run the OpenPR, uh… script, and then once your main PR is approved, you can, uh, just use this label to merge the… the snapshots PR. So you…"
|
||||
},
|
||||
{
|
||||
"time": "20:50:21",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "It's a bit of extra work, but I try to minimize it and automate as much as possible, so… That's how we currently have it set up with these failures and so on. Any questions so far?"
|
||||
},
|
||||
{
|
||||
"time": "20:50:37",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yes, this is amazing, and do we have a documentation for that in the Apple documentation project?"
|
||||
},
|
||||
{
|
||||
"time": "20:50:44",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "For the CAI part, you mean?"
|
||||
},
|
||||
{
|
||||
"time": "20:50:46",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yeah, for…"
|
||||
},
|
||||
{
|
||||
"time": "20:50:47",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "Or in general."
|
||||
},
|
||||
{
|
||||
"time": "20:50:48",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "In general. You've mentioned that there's a script that simplifies the work, and that would be great to have it as a task in the Apple Documentations project, if it's not there yet."
|
||||
},
|
||||
{
|
||||
"time": "20:51:00",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "Okay, I will add it, yeah. Yeah, so the biggest documentation right now is the README, but I understand that it's not, uh…"
|
||||
},
|
||||
{
|
||||
"time": "20:51:02",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Awesome. Thank you so much."
|
||||
},
|
||||
{
|
||||
"time": "20:51:09",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "in the best place. I mean, it's useful, but…"
|
||||
},
|
||||
{
|
||||
"time": "20:51:10",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Mm-hmm."
|
||||
},
|
||||
{
|
||||
"time": "20:51:12",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "Yeah, I will let it also in Asana."
|
||||
},
|
||||
{
|
||||
"time": "20:51:16",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Thanks so much."
|
||||
},
|
||||
{
|
||||
"time": "20:51:16",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Very cool, thank you, Pao."
|
||||
},
|
||||
{
|
||||
"time": "20:51:20",
|
||||
"speaker": "Pau Ballart",
|
||||
"text": "I have one more thing to share, which is… So two things. Why is this useful, right? Like, because, okay, we can add a snapshot test and so on. There are three things that I see that are important. First one is PR reviews. Since now the two PRs are linked, it makes it easier to review changes, right? You read UI code of SwiftUI, for example, and you see paddings, frames, blah, blah, blah. Seeing a screenshot helps a lot, so now it would be easy to compare side by side, because you will have here, um… For example, this one. The… the file changes. Directly in the… well, it's not the… This one needs a rename. And you will have the file changes on GitHub to review side by side like this. So PR reviews are better than. I think it was already mentioned we can. think about our UI tests, and if we have UI tests that are asserting. Just how things look like. It's a good candidate to replace with. with the snapshots. Um, so… I will open a proposal at some point, but the idea is… Since UI tests are flaky and snapshots and units. are less flaky and run quite fast. We might look at opportunities to move some of these tests and with a good combination of unit and snapshots. Yes, we are still missing the navigation part of I tap this button and this other screen opens. But yeah, it's up for discussion, but I think you have already a good coverage of a feature, right? And the last thing is ship reviews, because here in this other repo, it's becoming basically like an S3 bucket of data. PNG files. So when you prepare your ship review, if you have used this, you don't need, uh… to regenerate the snapshots and so on. You would just even link it from here or pick them from here. And that's just a starting point. We can improve it with future ideas, like we were talking about different languages. So how we could snapshot test different languages, different dynamic sizes for accessibility, like a large text size and so on. Even, I don't know, when the snapshots PR is open. with some automation, check the original Figma design, and compare it, and there are, there are, um… ways to improve this. So this is just a starting point. And last thing I want to mention, it was part of the tech design to track adoption of this. So far, I'm the only one contributing, which is why I'm doing this demo as well. But I created with Claude an automation that every week checks the repo. It's very easy, because we can just go to the insides of this, um… of this repo and see contributors, but also just check, uh, the asset and snapshot, uh, usages. So this is a dashboard that I shared, uh, within DuckDuckGo. It's updated every week, and basically we can see how. how it grows over time. Um… So yeah, if you fancy, please use it. And if there are any issues, let me know. I'm like the unofficial or official DRI of this. Happy to be on top of it if it's causing issues as well."
|
||||
},
|
||||
{
|
||||
"time": "20:54:59",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "It's really cool, Pat. Thank you for sharing. Yes."
|
||||
},
|
||||
{
|
||||
"time": "20:54:59",
|
||||
"speaker": "DR",
|
||||
"text": "Good work, pal."
|
||||
},
|
||||
{
|
||||
"time": "20:55:04",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Very cool. We are now at 5 minutes to the top of the hour. Dominic and Alex, you both have postmortems on the list. Do you want to… run those now, understanding we'll go past time, and anybody who can stay is welcome to stay or not? Or would you rather find another time for those?"
|
||||
},
|
||||
{
|
||||
"time": "20:55:24",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I'm fine running it. It should also be short, but not sure about Alexa's one."
|
||||
},
|
||||
{
|
||||
"time": "20:55:33",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Probably can wait till the next Apple Weekly."
|
||||
},
|
||||
{
|
||||
"time": "20:55:37",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Okay."
|
||||
},
|
||||
{
|
||||
"time": "20:55:37",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "I think it was waiting for a year and it didn't wait another week."
|
||||
},
|
||||
{
|
||||
"time": "20:55:41",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "All right."
|
||||
},
|
||||
{
|
||||
"time": "20:55:42",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Thank you."
|
||||
},
|
||||
{
|
||||
"time": "20:55:43",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Thank you, Alex. Um, then, Dominic, why don't you go ahead and get started? Anybody who's not interested in Dominic's incident post-mortem, obviously, is welcome to drop. If we run a little bit late and you need to be somewhere else, that's also okay to drop. And Alex, I'll keep yours on the agenda for next week. Thank you."
|
||||
},
|
||||
{
|
||||
"time": "20:56:01",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Thanks. I've dropped the link in the agenda as well for the incident postmortem. What happened last Monday is we released the app that was crashing on macOS 12. And, uh, the app was out… well, uh, so the app was out for, like. 3 hours before we, uh, realized that there is a crashing scenario on macOS 12, and then that was handled real quick, uh, with app store rollout stopped. And, uh, Sparkle rollout stopped, uh, as well at some point, um, and the fix was released, actually… well, it… It. Well, almost immediately on DMG, and the next day on App Store, because of the review time. Um, but, uh, how it happened? So, it happened, uh, because we had a change to, um, the save credentials pop-up, which shows up every time you're offered to… you're being offered to save a password. Um… On macOS 12, um… And that change was due to dropping a storyboard, replacing a storyboard with a code-initialized view. So what didn't go as planned? This, uh, change was not tested on old OS versions before merging, and it was strictly macOS 12 issue. Uh, like I mentioned before in this call, we had that engineering expectation in place, but we're not really enforcing it for every change. There's no way of enforcing it, uh, and there's no way of. testing, um… all those versions in CI, so… that's why it happened. Really. Uh, the other thing is, uh, we didn't know about the crash, because, uh, I overlooked it in Sentry. I was the release DRA and I overlooked it in Sentry. There was a single crash on macOS 12 there. Uh, that I treated as noise, and I went ahead and released the app. The sentry summaries weren't generated on that day, because we had another… Whoopsie. Um, the week before, when we updated, uh. Well, some updated the model version and I reviewed it and approved it. and we merged that change, and we updated the model version, but we didn't update the SDK version, and the old SDK didn't… wasn't recognizing the new model, and it failed to… collaborate, so we didn't generate summaries. Um… So that's the other… the other thing. Uh, when the… but! When the summary was generated, uh, it clearly pointed out that there is a crashing scenario on Mac OS 12, which started the incident. A week of internal testing didn't catch a relatively easy-to-reproduce crash. It was a 100% crashing scenario, which wasn't caught. Because we don't have testers on macOS 12. Well, apparently we have one, because we got this one crash at some point. I don't know who it is. I'm…"
|
||||
},
|
||||
{
|
||||
"time": "20:59:33",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah, and by the way, there was another kind of related issue about the crash on Intel machines. It's not directly related, but yeah, it was close."
|
||||
},
|
||||
{
|
||||
"time": "20:59:41",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yes, another one, exactly. We also don't have testers, or many testers, on x86. architecture. The App Store review took 15 hours, which is rather long for an expedited review, but also there's nothing we can do about it. At least it got in the queue… I mean, it got in the review pretty quickly after requesting a completed review. What went well is that the sentry summary that was eventually generated, um… Clearly identified the crash, and suggested addressing it, and the fix was simple, quick to prepare and test, it was just assigning an initial frame to a view. And the hotfix release automation workflow less at this time. It's always important to point this out, because… We only talk about CI when it's not working correctly. Despite the long review time, we only had 37 crashes in total and only 9 affected users. That's all. for action items and DRIs, we actually, uh… there was an action item to update agent's instruction prefect… essentially update prefect to ensure that Claude agent SDK version is… compatible with used models. Uh, Sam has handled it, and, uh, well, there is a code command right now to, to update the SDK version whenever you update the model version. Um… The other topic that's not actually mentioned here is what to do about testing on older OS versions, but this is something we will be handling separately through Alex's work, mentioned earlier on the call. And… That's it. Finally, I created an incident task through the form. Asana forum, and marked the incident as high, because it's a crashing scenario, we need a hotfix, and an SRE person joined the call, and we talked a little bit, and he, um… Explain to me that if it's a one-person rollout. It's not a high severity incident, and we dropped it to medium, and they dropped the call. And that's it. And, um… That's something to keep in mind for later, for next incidents, when they are only starting rolling out, we may not need SRA help there. This is all I have."
|
||||
},
|
||||
{
|
||||
"time": "21:02:09",
|
||||
"speaker": "SS",
|
||||
"text": "Regarding the action item, are you happy enough with just having the code comment now that reminds about the Claude SDK?"
|
||||
},
|
||||
{
|
||||
"time": "21:02:15",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I think so."
|
||||
},
|
||||
{
|
||||
"time": "21:02:16",
|
||||
"speaker": "SS",
|
||||
"text": "you know, it's… it's not bulletproof, like, what I… what I wanted to do was to just ask the Claude SDK, what models do you support, and check that it's in the list, but it only includes… yeah, it doesn't include all the actual supported models."
|
||||
},
|
||||
{
|
||||
"time": "21:02:18",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I think, yeah. I saw that, yeah."
|
||||
},
|
||||
{
|
||||
"time": "21:02:28",
|
||||
"speaker": "SS",
|
||||
"text": "Eh, maybe the comment's okay for now. I tried to make it very obvious what to do and how to confirm this."
|
||||
},
|
||||
{
|
||||
"time": "21:02:32",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yeah, I mean, that's okay. Also, that comment would show up, because it's just very close to the line that is changing, it will show up in the GitHub UI when reviewing the code, so it should be fine."
|
||||
},
|
||||
{
|
||||
"time": "21:02:37",
|
||||
"speaker": "SS",
|
||||
"text": "It's right there. Yeah, the only way I was considering extending this was to just have a GitHub Actions check that looks for model changes and adds a comment, GitHub Actions comments. But I would want to do this for all of Prefect. Whatever. I'll look at that."
|
||||
},
|
||||
{
|
||||
"time": "21:02:56",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Yeah, yeah. Should be fine. Okay."
|
||||
},
|
||||
{
|
||||
"time": "21:02:58",
|
||||
"speaker": "SS",
|
||||
"text": "Okay."
|
||||
},
|
||||
{
|
||||
"time": "21:03:00",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "Thank you. Thanks for handling that, Sam. Alex?"
|
||||
},
|
||||
{
|
||||
"time": "21:03:05",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah, I got a proposal in… I got it related to post-mortem of the VS Code, but I think it more suits here. What about adding… updating our process to require adding UI tests in addition to renting some older Mac to run older OSs, but make a requirement to the process to cover any UI change. with a UI test. Even if it's a fix or a quick addition in the uncovered region, especially having all the Agentic workflows, it may be a quick and cheap option. I mean, cheap to add. maybe not to maintain. But it adds, of course, it adds extra cost for the runners. Or maybe we could have like a test suite number two that runs weekly, for example, and not daily."
|
||||
},
|
||||
{
|
||||
"time": "21:04:11",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "I actually think it's… Some Windows team uses something similar. whenever they change UI, they add UI tests, and they have plenty of UI tests, and that's another problem. I believe, but… Yeah, we could… We could discuss it, I guess."
|
||||
},
|
||||
{
|
||||
"time": "21:04:31",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah, because, uh…"
|
||||
},
|
||||
{
|
||||
"time": "21:04:31",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "I think that's what we're considering. Sorry, go ahead."
|
||||
},
|
||||
{
|
||||
"time": "21:04:35",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Because obvious, like, obvious… thing that could fix it is, uh, like, QA, which we are missing, for… everybody knows for… for why, for what reasons, uh, but we have a lot of, uh, supported platforms, a lot of supported macOS versions, and, uh, it's, uh… Like, uh, it's impossible to… make the requirement as in the mentioned. Uh, process, like, to test all the supported macOS versions, because not everybody even have all these machines that, like, I have the older one, but it's not DuckDuckGo, it's my personal machine. And, uh, I can test it, but it takes time. Uh, and nobody… nobody does it, frankly. And we have no QA, and it's impossible to guarantee the quality in such conditions. So, maybe it's required. It should be a required condition to… to… raise the UI test coverage, the end-to-end coverage, so we can validate we are actually producing quality UI tests. Stuff."
|
||||
},
|
||||
{
|
||||
"time": "21:05:50",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Yeah, I think that's worth considering alongside the virtual machine testing stuff that you were talking about earlier, right? I mean, we've known for a long time that we clearly have testing gaps, especially on older versions and older hardware. And… To this point, I think we've just accepted that as the way it is and every now and then something like this goes wrong and we deal with it. um… But we should definitely think about whether there's a promise we're making to our users there that we're not able to fulfill properly. And I think certainly that's true right now. The question is, what does it take to do better at that promise? So I'd love to see some proposals around moving some of these pieces forward."
|
||||
},
|
||||
{
|
||||
"time": "21:06:32",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "All right, that was all for me. Thank you very much."
|
||||
},
|
||||
{
|
||||
"time": "21:06:38",
|
||||
"speaker": "Jonathan Jackson",
|
||||
"text": "Alright, and thank you. Um, so, Alex, we're gonna wait till next week for your other postmortem. Thank you for your flexibility there. And thank you all who stayed to finish that out. I appreciate it. And I look forward to seeing you again soon. Bye, everyone."
|
||||
},
|
||||
{
|
||||
"time": "21:06:54",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Thank you. Bye."
|
||||
}
|
||||
]
|
||||
}
|
||||
-134
@@ -15,137 +15,3 @@ source: "https://us01mynotes.zoom.us/gBBa-iiUQLycw55WRV1tnw"
|
||||
- **20:03:14 · Jonathan Jackson:** Hello, everyone. Let me go ahead and get started. Quick runs through our scheduled rotations as usual. This week, iOS triage is L, followed by Bartek, and then Brindy. Mac OS triage on is on this week by Jorge and then Rachel. Uh, for iOS release, Sam, you're on duty this week, followed by Bartosh, and then Mihao. And for macOS release, Dominic was last week, hasn't closed that task yet. Tom, you're on this week, followed by Juan, and then Joseph. You all may have also seen, I published our… Uh, release schedule changes around the winter holidays, uh, last week. Um, next step there is for me to look at out-of-office stuff and start trying to coordinate and… uh, coordinate swaps and coverage during that time when a lot of us are gonna be out, so keep an eye out for that soon, I'll get that. Um… I'm also using the Zoom transcript feature on top of Zoom summaries, and I'm gonna feed the transcript to Claude and see if we get a better summary that way than we have been out of Zoom's summarization AI. Uh, so… And I'll share maybe both, or I'll see what it comes up with, and I'll share an update after this meeting about what we ended up with there. Um, and one final reminder, next week, uh, is performance review week. The process starts with self-reviews happening on Monday. And then there's a few days of CAs writing reviews, and then a few more days of calibration after that. um… But because there's really only one day in that official schedule dedicated to self-eval writing, I suggest y'all think about starting this week. If you're anything like me, I usually need to write it, and then think about it, and then come back and edit it heavily, and then think about it, and then write it again. So I would suggest giving yourself the extra time if you need to do that. to get started early. Tom, what's up?
|
||||
- **20:05:23 · Tom Strba:** Just a small note, I checked the macOS release rotations. Yes, there is not finished one rotation that's Dominic's, but this week is Diego. Right? Is that true?
|
||||
- **20:05:36 · Jonathan Jackson:** Mm, got it.
|
||||
- **20:05:37 · Tom Strba:** And I think I'm next week. Because I almost got a heart attack when you said that I am…
|
||||
- **20:05:40 · Dominik Kapusta:** Yes, this week is Diego. And sorry for not closing my task.
|
||||
- **20:05:41 · Jonathan Jackson:** Okay.
|
||||
- **20:05:45 · Tom Strba:** that I am this week.
|
||||
- **20:05:45 · Jonathan Jackson:** This week… Why is Diego… I don't see a task… this is a macOS release we're talking about, right?
|
||||
- **20:05:52 · Tom Strba:** Yeah, I think Diego maybe marked it as complete in the beginning of the week, because there's just, like, triage…
|
||||
- **20:05:57 · Jonathan Jackson:** Yes, you're right, Tom, your block is the 26th to the 30th, because that task is missing, I didn't see it, I wasn't looking at the dates, I was just reading the list.
|
||||
- **20:06:04 · Tom Strba:** That's okay. Just wanted to be sure.
|
||||
- **20:06:07 · Jonathan Jackson:** Okay. Diego is responsible for the release this week. It sounds like he's actually doing that. Is that right?
|
||||
- **20:06:08 · Tom Strba:** Perfect.
|
||||
- **20:06:14 · Jonathan Jackson:** He knows. Okay, cool. Thank you.
|
||||
- **20:06:14 · Dominik Kapusta:** That's right.
|
||||
- **20:06:18 · Jonathan Jackson:** Um, no problem. Okay. Anyway, think about starting your self reviews this week. If you feel like you're gonna need more than a day to knock it out. Bartek, over to you.
|
||||
- **20:06:31 · Bartek Waresiak:** Yep, I will be stepping down as IS app DRI. Jacek will take the primary role, and Brindy will take on secondary role. So, yeah, if there is IS drama, if there are IS issues, if there are any urgent things. Go to them! And thank you very much.
|
||||
- **20:06:57 · Jonathan Jackson:** All right. Thank you. An, you're next on the agenda.
|
||||
- **20:07:08 · Anh Do:** Oh, that was quick. Yeah, so in OK, there have been a few projects that involve native changes required as part of the front-end-led projects. Um, so the background is that they, uh, want to contribute some PR, and, uh, um… AI assisted Pi, of course, and then what have been working for us lately is like depends on the size of the task like I would be the one to like either. Dri the change that they are they want to make, or like they prepare the Pi, and I do a thorough review of that. So and now we want to kind of do this more regularly. And so there was a need for some kind of like guidelines that we can give to contributor from outside the apple team. So I was just starting like a small task to basically collect ideas from people. And if you have anything that we already have somewhere, maybe I could get put in. And so. it becomes something that we can point other people to when they want to contribute to the Apple codebase to ensure our standard and make sure that, like, everything is being done the way that we want. So that's about it. Nice.
|
||||
- **20:08:19 · Jonathan Jackson:** Cool, thank you. I just took a quick peek. I see you do mention the engineering expectations. I think that's a good source. We might also want to link… to the onboarding process. Not that they would need to follow all of it, but some of… that's basically our best source of, like, how to get set up with the codebase. So there's probably a little work to do to figure out how to have a slimmed down version of that that folks could follow. And I'm sure other people have other ideas as well. So thank you. Um… Luke, we have a non-Apple person lurking in our midst. Welcome. What can we do for you?
|
||||
- **20:08:54 · Luke Andrew:** Hello. Yes, thanks for having me. Luke from SRE, for people that I haven't met before. Um, this is just a plea for generic Apple help. I don't want to, sort of, get all of your attention to focus on this task. We basically have… a looping client that seems to be sending, uh, requests once a second to us, uh, that is Apple, but not DuckDuckGo Apple, based on the user agent, and doesn't seem to be Safari either. Um… And it's cumulatively across lots of different devices, lots of different queries, sending in the region of 3 million searches a day. Which is substantial, not sort of catastrophic, but substantial enough searches that it's something that we want to deal with. Umm. And I don't really know where to go next to try and understand where it might be coming from or how we might try and reproduce it. So I don't want to solve that in this call, but I just wanted to bring your attention to the task. And if there's anyone who thinks they might be able to help, I'd really appreciate. a second pair of eyes, and pointed out that it's not even worth, with what we have at the moment, that we can, like, escalate this to Apple, because we just don't have enough, uh, information to get any help from them. So yeah, that's it. If anyone has any questions immediately or async, just pop them on the task. Yeah, that's all I have.
|
||||
- **20:10:18 · BA:** Quick one, Luke.
|
||||
- **20:10:18 · Jonathan Jackson:** This is not from a single client, but from a particular app version, potentially, is that right?
|
||||
- **20:10:24 · Luke Andrew:** Yeah. Yeah.
|
||||
- **20:10:24 · Jonathan Jackson:** Okay. You know, be scared that it was one client going crazy. Okay, um…
|
||||
- **20:10:30 · BA:** Yeah, I was going to ask, what's the — we eliminated bots from the equation. Are we happy that it's not bots?
|
||||
- **20:10:36 · Luke Andrew:** It doesn't look like bots. The queries all look human. The rate is really low compared with bots. The moment that I put in a block for it and it immediately stopped, like redirected away.
|
||||
- **20:10:39 · BA:** Okay. Right.
|
||||
- **20:10:52 · Luke Andrew:** It got, like, a 302 from deep, and then just immediately stopped, so that doesn't feel like… bossy behavior either. They normally persist a little bit and then give up. Whereas this was just immediately, it broke out of whatever loop it was in and stopped.
|
||||
- **20:11:09 · Jonathan Jackson:** Interesting.
|
||||
- **20:11:10 · Luke Andrew:** Uh, so don't all take a look, because that would be a waste of time, but if you think you can help, um, I'd appreciate a pair of eyes. I'm gonna drop off. Thank you all for having me.
|
||||
- **20:11:19 · Jonathan Jackson:** Yeah, one last quick question. It sounds like you're treating this as something that definitely needs attention, but not an incident at this point. Is that correct?
|
||||
- **20:11:25 · Luke Andrew:** Yeah, that's correct. It's been going on for… Based on… um, how we categorize queries for, um, like, ads perspectives. These get classified as a separate category based on user agents, so we don't send them to Bing and say, these are definitely good queries. Um, it's been happening for about a month. So it's, at the moment, not urgent. I'm gonna raise it with SRE, maybe, and see if we think it should be an incident that we just let run, but at the moment, that's not how we'.
|
||||
- **20:11:55 · Jonathan Jackson:** Okay. Great. Thank you for the clarity there. Uh, and yeah, if anybody thinks they can help Luke out and figure out what's going on there, feel free to hop into the task. Thanks for coming. Moving on. Alex, you're first in the regular topics list.
|
||||
- **20:12:16 · Alex Martemyanov:** Yeah, hey. I have several topics for today, including the VS Code integration postmortem to run later. And the first one goes with a proposal. The proposal is pretty short. And kind of related to the, uh, recent incident with Mac OS 12 crash. that we are not. Testing for macOS 12 and 13, I guess, because we are missing the infrastructure on CI runners. And the suggestion is pretty simple to extract the macOS UI tests target from the main macOS Xcode project, leaving it as a separate macOS project inside the. uh, Xcode workspace. So it will be available the same way it is now, and, uh, runnable. So only, uh, the only issue I can see for those who use, uh, the actual macOS Xcode project instead of the workspace. they would have to switch to another project to run the tests. So, I just wanted to hear if there is any concerns or objectives against it.
|
||||
- **20:13:39 · Dominik Kapusta:** I think some time ago, we… someone asked… who uses or doesn't use Workspace, and people who didn't use Workspace were in a… In a minority, anyway, like, very few of them, so… I guess… I could move to the workspace.
|
||||
- **20:14:06 · Jonathan Jackson:** I think it'd be good to ask that question in that proposal, at least more explicitly, and just give people a chance to chime in if they feel like that would be a hurdle. Um… Other question for me is, is there… should we think about the iOS side of this as well, and to match… structure, or is this really a macOS only question?
|
||||
- **20:14:26 · Alex Martemyanov:** Okay. Maybe, I… Didn't think about it, actually. And the main concerns here is the compatibility. So we can reach macOS 12 compatibility, at least for some local runners that I'm going to present further. And the build times, so we have package fetching and building for every macOS runner that runs the UI tests and any of the dependencies. None of the dependencies are used during the runs, so it's a 10-second step for each runner that can be easily omitted, and so, yeah. And yeah, it's not related to the main browser targets, so there's no risks. Uh, and they're both, uh, U.S. Probably to think about it in the ER.
|
||||
- **20:15:25 · Dominik Kapusta:** I think iOS is not…
|
||||
- **20:15:25 · Jonathan Jackson:** I mean, I don't have a strong… sorry, I don't have a strong feeling there, I just felt like if we're gonna make this change, like, if it's gonna end up with different structures depending on the app, we should think about whether we really want that to be the case, or if we should do the other.
|
||||
- **20:15:37 · Alex Martemyanov:** No, it actually shouldn't change any structure, any file structure, just the project placement.
|
||||
- **20:15:38 · Jonathan Jackson:** Screwed up. Yeah, yeah, yeah. Dominic, I interrupted you. I'm sorry.
|
||||
- **20:15:44 · Dominik Kapusta:** Ah, no worries, I just wanted to say that iOS is not affected by this at all, because… iOS doesn't need testing on various macOS versions.
|
||||
- **20:15:51 · Jonathan Jackson:** Alright. So the limita… this is about dealing with a limitation that comes from Mac OS 12 build, particularly.
|
||||
- **20:15:59 · Dominik Kapusta:** Yes.
|
||||
- **20:16:01 · Alex Martemyanov:** And bill times.
|
||||
- **20:16:01 · Dominik Kapusta:** I'm Bill Times.
|
||||
- **20:16:02 · Jonathan Jackson:** Got it. Okay. thanks for pointing it out. Uh, like I said, I think you might want to give people a little bit more time to respond to async if they have… particularly if they feel like the… Workspace thing is gonna be a hurdle, but my guess is it's not gonna be a big issue. Uh…
|
||||
- **20:16:22 · Alex Martemyanov:** Okay.
|
||||
- **20:16:24 · Jonathan Jackson:** Next topic.
|
||||
- **20:16:24 · Alex Martemyanov:** We can…
|
||||
- **20:16:26 · Jonathan Jackson:** From you, also.
|
||||
- **20:16:26 · Alex Martemyanov:** Perfect. To the next topic then. Yeah, so, uh, this is another iteration of, uh, the Verfield project. I've been live coding for a while already. Uh, so this is the current state of things. It's changed a little bit, some of you may remember the first, uh, first demo. Uh, so the change is it's, uh, migrated to Go language, and, uh, now it runs on TART system that's used on, uh… We tried runners, I guess. And so the system is here to show you the available images, available runners and locations. So here you see I can have it. stored on the external SSD, because I have my internal 2TB ETH. up to completely, and I have the old MacBook Pro also managed from the same environment. And, uh, you can add other machines as well. Uh, so this is, uh, a wrapper around the TART system that manages, uh, Xcode installs, uh, SIP protection, and, uh, for virtual machine and real machines. Uh, if you use older hardware and, uh. It's, uh… basically providing MCP server for AI clients to. to run these UI tests. Mostly. on the older virtual machines. It can be not older virtual machines, but actual virtual machines. New refreshable machines as well, sir. Uh, for example, I don't update to macOS 27, because, uh… I expect some issues, but I use, uh, testing on macOS 27 using this, uh… a virtual machine system. And so basically the… Again, I have a question to the team in… In the light of a recent discussion of testing on order. macOS versions, for example, whether should I try to push it forward to, like, team-wide? So the idea… the idea is to be able, uh… again, we'll… And in the post-mortem. for the VS Code integration projects. So here you can see VS Code. Basically, it's running the Google browser. It can… run iOS browser as well, in simulator, and in addition, it can also run remote debugging, unlike Xcode, which had supported remote debugging. in a long time ago, but currently lacks the ability, so… Integrating these systems together allows, like, click one button. choose the virtual machine you want to run to test on, and it builds, deploys, and debugs in real time without any lags for library symbols transfer and other stuff. So it just works like you would do on your host machine, but in the virtual machine. So I wanted to hear opinions whether should be. try to adopt this team wide.
|
||||
- **20:20:06 · Dominik Kapusta:** I can offer an opinion.
|
||||
- **20:20:09 · Alex Martemyanov:** Nobody wants to.
|
||||
- **20:20:09 · Dominik Kapusta:** uh… No, well, I mean… No one to be the only one commenting, but anyway. This will be the topic of, uh, of, uh, of the upcoming postmortem, uh, as well, that… The Apple engineering expectation of testing on older macOS version is essentially a dead low. We are supposed to test on older macOS versions, but we rarely do it. Perhaps for new features on macOS? I'm not sure. But certainly not for… Small… tasks. I don't think… well, I don't know, but I don't think we bother to test on all the ROS versions, every small version. Change to the UI, especially. And, uh… Because of that, if we had a way, and also we can't, we are unable to test in CI. So, if… I had a way to… Ask an agent to… run UI tests on a virtual machine. on an arbitrary OS version, that would be great. I would like to see the demo of that, actually. That would be awesome. And that is why we might be interested in adopting it. Perhaps.
|
||||
- **20:21:41 · Alex Martemyanov:** Yeah, again, about the UI tests, correct point. So sometimes you want to run an agent reproducing an issue, for example, and it's not always convenient to upload it to CI, wait for the UI tests to be run. Look for the results, while if you run it locally on your machine, it will be blocked for 5 minutes, and do the stuff while you take a coffee. So this thing allows you to just run directly from the agent on the same machine in background and provide you screen recording with all this and the results.
|
||||
- **20:22:22 · Jonathan Jackson:** Yeah, I mean, I think… You know, your point that our expectations are that we'd be testing on all versions that we support, but in practicality, I think almost nobody actually does that. And particularly with something like as old as Mac OS 12 where we don't have the hardware, we don't have like, most people don't have access to test it to your point, L. Um… I think I'd like to understand a little bit more about how the system would work, what kind of third-party dependencies there are, what it would look like to really set up and what it would cost. I think the idea of we would be able to test. On versions that we say we actually support is pretty appealing.
|
||||
- **20:23:00 · Alex Martemyanov:** Okay, so probably the resolution here would be, um… So, first of all, I do use it as my self-made tooling, and uh… if… I can reach it to the point, I can make it to the point when you click a button, install the system, click a button, and it just runs. I think it would be worth. adopting at least as a selected auditory who can. have, uh, the means to testing, like, enough of RAM or some older hardware, uh, like, choose the early adopters and, uh, see how would we… expand it. And, uh, yeah, correct question by, uh, Federico about, uh, older ores. It's another topic that I wanted to raise during the post-mortem. During the incident, actually. Both for the incident, uh, that, uh, having… Maybe one dedicated machine, older machine, maybe somewhere that would run the UI test suite at least once per week for each release would be a great addition, even if it's like a physical machine somewhere. someone's office or home space.
|
||||
- **20:24:28 · Jonathan Jackson:** Yeah, that's an interesting idea, too. I think… I guess I think that a proposal here that sort of lays out what a strategy for old testing would be that used this would be maybe a useful next step. Um, because then that would be a place that people could rally around, sort of solving that higher order problem of how do we actually make good on our promise of testing on the systems that we support. And it sounds like this would be one, but maybe not the only tool to consider, or maybe even as a combination. Does that make sense?
|
||||
- **20:25:00 · Alex Martemyanov:** Yep.
|
||||
- **20:25:01 · Jonathan Jackson:** Cool. Okay, thank you. Interesting stuff. Sam, you're up next.
|
||||
- **20:25:10 · SS:** Yeah, recently I have been chatting with some folks about being able to put Claude workflows in our CI. I talked about this with OT and they told us that they don't want Anthropic tokens being available in public repos. I've created a private repo to host the agents, uh, which is… Pretty bare bones right now. It lives here in the Apple Team Agents repo, it has CI set up, it has the Claude, uh, action set up. I have a workflow first for testing it, uh, just so we can make sure you can actually talk to Claude and back. Um, the only thing I'm waiting for is whether or not we should get a dedicated API key, because I've been told. Two different things from OT. So I'm just waiting to —. Get that sorted out. I think we should get a dedicated one. once we've done that, yeah, we can start doing things like an automation to clean up feature flags and such, like Fernando's, uh, proposed. So, uh, I'll post to Apple Diary once that's ready, but I think you'd be… if you do want to start building automations in this now, like, feel free, uh. As long as it's, you know, not super expensive. Pete, done.
|
||||
- **20:26:19 · Pete Smith:** Yeah, thanks, Sam. This, this looks really cool. Just because I've seen, um, like, there's a few different conversations. There's a conversation with this, and there's the Bitrise, you know, running, um. uh… Agentic stuff on Bitrise as well. How does it all fit together? Like, I think I understand it, but, like, say I wanna… say we wanted to, you know, uh, build our feature overnight while we're in bed, um… How could we use this to do that? Or can we?
|
||||
- **20:26:43 · SS:** This is not that. This is definitely not that. This is not like the development environment thing. Yeah, these are not like the development machines. I really just expect these to be like GitHub Actions workflows that just do helpful things with Claude. Uh, and we just need a place to put those. Like, if it weren't for the API key limitation, this would live in the monorepo.
|
||||
- **20:26:56 · Pete Smith:** Yep. Yeah, yeah, I guess, I guess what I'm wondering is, so, like, yeah, exactly, we can, we can hit Claude and, um, but, you know, Claude can't, well, build for us, so, like, well, build and check that it's actually working, so where would… say, say even, like, removing feature flag, where would that. On the PR, at the PR station.
|
||||
- **20:27:17 · SS:** Yeah, this is something that Fernando and I was talking about, about doing some kind of coordination between these repos where the monorepo would talk back and forth, do the things that it needs to. This can create a branch. The monorepo can then do the actual testing. Yeah. Right now, I'm just using these for, like, dependability analysis kind of problems, or planning on anyway, which I have a demo for coming up soon.
|
||||
- **20:27:33 · Pete Smith:** Okay. Yep. Cool. Yeah, cool. Thanks.
|
||||
- **20:27:44 · SS:** Dominic.
|
||||
- **20:27:46 · Dominik Kapusta:** Mm-hmm, uh… one of the limitations here is that GitHub Actions workflows are limited to 6 hours of runtime. So…
|
||||
- **20:27:53 · SS:** Also that, yeah.
|
||||
- **20:27:54 · Dominik Kapusta:** Yeah. And also, it's, of course, non-interactive. Um… Which is why we're currently trying to figure out if Bitrise… or Bitrise or not, but anyway, looking for some infrastructure provider for, uh… macOS machines that could run Agentic workflows and be… and offer. interaction with them, right? So, uh… Bitrise RDE remote. development environments is one of the options there. Um… That is in progress. For now. They also… it's also quite a fresh feature on their end, and they have some bugs there that we're trying to resolve right now. And we have a CloudOps team member on that as well. the side of… Connecting them to ClickHouse or Sentry. Yeah, so… We'll see in some time.
|
||||
- **20:28:58 · SS:** Yeah, yeah, to Dominic's point, like, that problem is much bigger than what this repo is. This really is just, I want to be able to run Claude on an automation, like I currently do on my own machine, and I'm just sick of doing it on my machine. I just want to push that up to GitHub Actions. Have it run on its own, and… Show up with PRs and Asana tasks occasionally, so. Okay.
|
||||
- **20:29:19 · Jonathan Jackson:** So the limitations here are that it can… It can inspect code, it can run as an agent, but it can't build and run, so it would… prospectively submit a pull request, and then CIA would tell you whether it did the thing or not.
|
||||
- **20:29:32 · SS:** Yeah, the idea that I'm thinking here is that, yeah, like PRs would go over back to the monorepo. This thing would open PRs against the monorepo. And then CI would do its thing over there. Yeah.
|
||||
- **20:29:38 · Jonathan Jackson:** Right. Yeah, and if we needed to inspect it from there, we could pull a branch and do whatever else we needed.
|
||||
- **20:29:46 · SS:** Yeah.
|
||||
- **20:29:47 · Jonathan Jackson:** Okay, cool. Definitely sounds like a good step, even if you're not trying to solve everything all at once, which is reasonable. You're the next topic as well, Sam.
|
||||
- **20:30:00 · SS:** Oh, is this the demo? Probably should have just done this all at once. Yeah. So one of the things that I.
|
||||
- **20:30:04 · Jonathan Jackson:** Oh, before you… sorry, if you're doing demos, before we get into demos real quick… Uh, we had a note here that we were gonna actually record demos. Um, but we haven't ever done that. Mostly people have opted out of wanting to be recorded, so I wanted to just check in since we have 3 demos here. do we want to record these? Should we rely on the… just the transcript summaries at this point, and whatever tasks people have written? I'm kind of okay either way. Um… I don't know. I don't know that we've ever recorded a demo.
|
||||
- **20:30:35 · Pete Smith:** Alessandro. Alessandro is waiting for his Wednesday morning. Won't be the same if we don't record our demo.
|
||||
- **20:30:39 · Jonathan Jackson:** Right. That's the question, right? I know, yeah, it's like this is sort of a hard way to do this. So, Sam, would you like me to record or not?
|
||||
- **20:30:51 · SS:** I mean, I've already shown this to Alessandro, so I don't think he cares. Yeah. Up to you. I'm good either way. I remember. Don't worry. This is a super quick one anyway, just to show the initial problem that I'm trying to solve with… with a repo like this. Let me share my screen. Um, so I have, for a while, wanted the other half of the flaky tests project. Right now, we have a flaky test project, which keeps track of, uh, you know, tests that are failing. Um… I have traditionally been more interested in the problem of what are the workflows that are failing, uh, and why, because these things were just, like… this stuff comes up a lot, like, during Xcode upgrades in particular, so… This is something that is being used by Claude to pull the monorepo, get the actions, go and look at the code, figure out what's causing a problem, and report back and give a little write-up about what's going on. Um, the idea being that eventually you could then move something into this ready-to-fix section, and the agent would then go and actually open the PR later. Um, you know, this is just moving a… moving an automation that I've had on my machine is like a daily, like, 10 AM workflow for a while. and just putting it remotely so I don't have to think about it anymore. Which is the ultimate goal. So yeah, this is the kind of thing that I'm trying to do with this repo. Um… As an aside, OT is kind of working on… something along this as well, which I learned about as I was setting up this repo. They are setting up this template, which lives on Dub, for what it's worth, for these kind of analyzing report workflows. Umm. So that might be interesting if you are needing like access from things within the system already. Uh, like, I think Lucas is currently looking at getting that thing to have sentry access and all this kind of stuff, so… Yeah. Go check that project out if you're interested. That's it.
|
||||
- **20:32:48 · Jonathan Jackson:** Awesome. Thank you. Pete, you're next.
|
||||
- **20:32:57 · Pete Smith:** Yeah, thanks. Yeah, this is in the demo section. It's not really a demo, I guess, but just wanted to share. I'll speak a little bit about, um. ASTAC, so I put a link in… a link into the chat there, so I'm sure people have heard… of… PSTAC, or maybe people have heard of PSTAC. I'm going to just share my screen. So if you haven't, um… Pstack is a plugin created by an engineer at Cursor or SpaceX. It's basically just a set of skills that this engineer found were very good. at, um, fixing bugs, implementing features. So they created the PSTACK plugin. What ASTACK is, is basically just a port of PSTACK. So PSTACK was. built for cursor. So it has a lot of built in you know, instructions in terms of cursor models and how they work so forth. So I just. Ported PSTAC, uh, over… To, uh, ASTAC, um… Basically replacing any of the cursor references with Cloud Code and Codex references. So what AStack is, is a plugin that you can use in Codex or Cloud Code. It's like it sounds very abstract. It's not. It's really just a. A set of skills, a set of playbooks, some defined principles. And they all work together to what I found to just to produce good output. um… So I'll just, I'll talk a little bit in a moment about how ASTAG works. So how it's set up at the moment, you can see on the left, I just have it there under my public, under my own public repo. But if anybody wants to try it out, I would encourage it. You can see like I didn't do a lot of work here. I really want to undersell this. asked an agent to port this over. So take PStack, port it, make sure it works with Cloud Code and Codex. The most work I did was creating that logo you see on the left of the screen, so that's where all my creativity went into. Um… But yeah, you can go here, you can grab the plugin, you can install it in Cloud Code. And you can try it out. I'd recommend trying it out for… for bug fixes, even like small features that are well defined. So, you know, the evidence, like any evidence I have is just anecdotal. I've just found these particular set of skills, these set of playbooks and the principles that are defined just produce very good output. Again, anecdotal. I haven't measured this. I also think it seems to do it at lower token usage. But again, I haven't measured it. So how it works, how ASTAC or PSTAC work is, you know, you install the plugin, it sets up a hook, so at the start of your sessions, the agents are aware that ASTAC is there. And if you say, hey, build this feature or fix this bug or whatever, you can explicitly say. Enter potato mode. But that's actually part of the hook. The agent will know, oh, if you give it a feature or a bug, it will enter potato mode. All that is like a top level skill, which is basically a router. And depending on the task that you give it, like fix this bug or implement this feature. You will go and find the right playbook and follow that playbook and in the playbook. That would reference what skills to use, et cetera. So. again, you know, there's nothing, like, there's nothing magical here. I mean, there's been other plugins that have been very popular. I was talking to Fernando earlier, and we talked about… was it the Superpowers one? I think was another popular plugin. So things like this will probably be, as I was speaking to Fernando about, they'll probably be, can, you know, eventually end up in the harness. Probably won't need plugins like this at some point. But at the moment, this is something I'm using daily. And I find it very beneficial. It just results in really good output. So I would encourage people to try it out. Like, what what, you know, we've seen probably saw the post from Sebastian, from OM. He's creating, like, auto, which is their version of pstack. He just took pstack. And ported that and kind of made it very specific to Duck AI. That's like the the OM version of Pete's group. So if ASTAC, if people find it useful, we could basically turn that into like the Apple version of PSTAC, Apple team version of PSTAC. and… that's something we could do. So yeah, I would just encourage people to try it out, if you're interested. There's an Asana task there that I posted, and you can give any feedback, and if people want to. People find it useful and want to contribute and make it better. We can do that. It's under AI enablement AOR. I spoke for a long time, if anybody has. Sorry, I shared that artifact. Sorry, key. I didn't go through all this artifact.
|
||||
- **20:37:59 · Pete Smith:** It's linked in… The Apple Weekly Task, so you can go and look at that artifact, it explains at a very high level how it works. Again, it's not magic, it's just a skill, a ruler, lots of other skills. Sorry, my dog is barking. I don't know if you can hear my dog, but. He's over there barking at me. That's very low. That's it. If anybody has any questions, you can ask me questions, or you can ask me questions later. Jane, sorry, your hand up.
|
||||
- **20:38:35 · JF:** Yeah, sorry if you said already and I missed it. What changes have you made over the original piece, Doug?
|
||||
- **20:38:42 · Pete Smith:** Yeah, so… Oh, I'm sorry, it looks like. Okay, dog went dead. Um, yeah, that's why I wanted to undersell it. Like, I didn't, I didn't, I didn't do anything. So, uh, I created that logo. What, what, what I did, or what the agent did, was just port it over, support Pstack over, um, for Cloud Code and Codex. Like, the original Pstack has references to Cursor, and how Cursor. Works. So I just ported it over, so it works in… Um, codecs and cloud code. I didn't add anything additional yet. Um… But the idea will be we can add things if we find them useful. We can add Apple-specific things. If we find them useful. So yeah, no changes really. Sorry, one more thing, James. Thanks for reminding me. I'm going to reshare. Sorry, everybody. Very annoying. I'm going to reshare. So a good question. Like, PStack is being continuously updated. So I have a scheduled task that runs every Monday, looks at the latest PStack, and consumes the latest changes, and, like, ports over the latest changes. So I keep it up to date. So I look at the upstream and keep. a stack up to date. But that's other than that, I just I just port it. I don't make any other changes. Like, you know, of course, there's a chance it's not ported well. It might not work properly, but the intention is it's just a direct port.
|
||||
- **20:40:09 · Jonathan Jackson:** Cool. Thank you, Pete. Paul, I'm protesting.
|
||||
- **20:40:16 · Pau Ballart:** Yes, thank you. I'm going to share my screen. This is pretty much. a demo that I owe you of the snapshot testing. So, quick reminder, this is already available on iOS, not yet on macOS, because we are waiting for macOS 27. runners on big drives on the CI. on, uh… Yeah, pretty much. It's a package that has a very nice README explaining of it, what I will show today. So you can read it or tell your favorite agent to read it. But it has like requirements, how to use it. A bunch of examples for both iOS and macOS, even though, as I said, macOS. is disabled. But let's do a bit of an example. if I find the Xcode that just crashed, apparently. Ah, it's here, yeah. So, uh, how does it work? We start with the bill. Maybe, or maybe not, you will have previews. That is the new, like, hashtag… preview macro but for a snapshot so I created this one you don't need to use previews but it's just if you use them it becomes handy so there is this new preview provider that's the system. protocol, then if you conform it like this, so pretty much. You configure the different states that you want to have in the view. In this example, it's this segmented control. So there are two previews, one for each tab, let's say. You configure it here. So that's the model or the state of the view. And then you configure it here. So it will render one of the previews for each tab in this case. That basically gives you the previews here, but what is nice about it is that then you have the snapshot test. becomes basically a one liner or yeah one four liner uh but yeah just one call because it reuses the same previews that you refine it and then uh this is just a unit test like. Any other that we have. that runs the test. And this. Basically, you can configure it in different ways, for example. Here, if you want, um… to have a preview, but you don't want to snapshot it, because it's not needed. You have different. uh parameters here in this um. Let me show you. In these configurations, so there is… a scope where you can say, I want just for previews, just for snapshots, or for all of them. And then, um… in the test, you can snapshot in a different set of, um, configurations. So, in these asset snapshots. But everything is here in this package. We have different things. We have the different strategies. If you want to snapshot, for example, iPhones, iPads, all the combinations, just a single screen or a custom. Uh, size? And then you have also. the different sizes of components that you can have in an app, right? Majority of things are the screen, but you can also test some intrinsic content size, bottom sheets, or just any big size that you want. So all of that has, like, default parameters. In this example, for example, it's just, you know, normal screen, and I want to judge a snapshot in that mode, because that screen doesn't have light mode. So, it would work like this. And this is a unit test, it runs super fast, and uh… It passes in less than a second. how it works when you change something so let's say remember we have this previous let's say now i change this uh. this view. No, there is a design change, so we have this title here. If I run the tests, now they will fail. I will show you how it looks like in Xcode. So pretty much, issue recording, blah blah blah, does not match reference. And then, you can go here in your tests. We see the test that failed, and if you double-click, there is this gallery. That basically is showing us the reference, the failure, and the difference. So you can preview it like this. That's the original one. And then the new one that got recorded is this one with hello.
|
||||
- **20:45:21 · Pau Ballart:** So you can see a diff that it's generated. Basically, you see everything is moving down. Um… So here it can happen two things. It's a real failure and you need to fix your UI, or it's a design change that is… expected, and you want to re-record the test flight, because this is now the new correct design. To do it, you have different ways. Here, there is a record Boolean. You can pass it to true. But at the same time, instead of changing code. If you go to the folder of the test. In in in Git you can let me show you. in a… in another screen. You can just delete the reference image in the file system. So that's the submodule that we have. In the, um… With all the snapshots. So if you go there and delete the file, it will recreate it on top of it. Um, so yeah, basically you have these two options. If there is no original file in the file system, it will just create it right away. If there is an existing one, it will do the comparison and so on. So the easiest usually is, um, to just remove. the image and re-record it. But let's say that… This works, and you upload a PR, and it fails in the CI. So how does it look like? Pretty much the current PR checks that we have for iOS, we'll report them here with the faillet, and it generates an HTML report, which is this one. that, uh… creates the same thing that we see in Xcode, but in the CI. So if you see a PR failing, you don't need to download the artifacts and so on. You just have to open this, and it has, like, these drop-downs. with the same concept, like the reference, the failure, and the diff. Um, so that's… A bit what I wanted to show. Then, um… So the snapshots live in a submodule, as you probably know. So it's a different repo. It's these DuckDuckGo Apple browser snapshots. And it just contains the PNG files. So that is this. Orchestration, if you want to run when you create snapshots, you need to open, let's say, two PRs. But there is a script that handles everything. So what you would do is here when you record your new snapshots. Maybe I can do it really quick. Just a second. I will. Show you exactly what I mean. So we have the submodule. Snapshot references. I'm gonna remove… The, um… Tests for this one. which is the scan QR code test. So I remove it from the file system. Okay, from this folder. And then when I record it or when I run the test, it will record. So first time it fails, but now it says no reference found on disk, which is fine. And then second time it runs, it already has the references and it passes. Okay, so now we want to open the PR, uh, and we can see that the snapshot reference module has the changes. Again, old and new with a new title. So there is a script in the scripts folder. around doo doo doo scripts. called Open Snapshot Zoom Module PR. You just run this over there, runs it, and it will open the PR, so commit, push, open PR, and so on. Um, and then once in your main branch or main PR, there is a new label that you can tag. Uh, let me show you here. Merge the snapshots. So this will merge the snapshots PR and update the pointer and so on. So, it's these two extra steps. You can do it manually, but, uh, the easiest is once you record it, and you have the changes here in Git, you just run the OpenPR, uh… script, and then once your main PR is approved, you can, uh, just use this label to merge the… the snapshots PR. So you…
|
||||
- **20:50:21 · Pau Ballart:** It's a bit of extra work, but I try to minimize it and automate as much as possible, so… That's how we currently have it set up with these failures and so on. Any questions so far?
|
||||
- **20:50:37 · Dominik Kapusta:** Yes, this is amazing, and do we have a documentation for that in the Apple documentation project?
|
||||
- **20:50:44 · Pau Ballart:** For the CAI part, you mean?
|
||||
- **20:50:46 · Dominik Kapusta:** Yeah, for…
|
||||
- **20:50:47 · Pau Ballart:** Or in general.
|
||||
- **20:50:48 · Dominik Kapusta:** In general. You've mentioned that there's a script that simplifies the work, and that would be great to have it as a task in the Apple Documentations project, if it's not there yet.
|
||||
- **20:51:00 · Pau Ballart:** Okay, I will add it, yeah. Yeah, so the biggest documentation right now is the README, but I understand that it's not, uh…
|
||||
- **20:51:02 · Dominik Kapusta:** Awesome. Thank you so much.
|
||||
- **20:51:09 · Pau Ballart:** in the best place. I mean, it's useful, but…
|
||||
- **20:51:10 · Dominik Kapusta:** Mm-hmm.
|
||||
- **20:51:12 · Pau Ballart:** Yeah, I will let it also in Asana.
|
||||
- **20:51:16 · Dominik Kapusta:** Thanks so much.
|
||||
- **20:51:16 · Jonathan Jackson:** Very cool, thank you, Pao.
|
||||
- **20:51:20 · Pau Ballart:** I have one more thing to share, which is… So two things. Why is this useful, right? Like, because, okay, we can add a snapshot test and so on. There are three things that I see that are important. First one is PR reviews. Since now the two PRs are linked, it makes it easier to review changes, right? You read UI code of SwiftUI, for example, and you see paddings, frames, blah, blah, blah. Seeing a screenshot helps a lot, so now it would be easy to compare side by side, because you will have here, um… For example, this one. The… the file changes. Directly in the… well, it's not the… This one needs a rename. And you will have the file changes on GitHub to review side by side like this. So PR reviews are better than. I think it was already mentioned we can. think about our UI tests, and if we have UI tests that are asserting. Just how things look like. It's a good candidate to replace with. with the snapshots. Um, so… I will open a proposal at some point, but the idea is… Since UI tests are flaky and snapshots and units. are less flaky and run quite fast. We might look at opportunities to move some of these tests and with a good combination of unit and snapshots. Yes, we are still missing the navigation part of I tap this button and this other screen opens. But yeah, it's up for discussion, but I think you have already a good coverage of a feature, right? And the last thing is ship reviews, because here in this other repo, it's becoming basically like an S3 bucket of data. PNG files. So when you prepare your ship review, if you have used this, you don't need, uh… to regenerate the snapshots and so on. You would just even link it from here or pick them from here. And that's just a starting point. We can improve it with future ideas, like we were talking about different languages. So how we could snapshot test different languages, different dynamic sizes for accessibility, like a large text size and so on. Even, I don't know, when the snapshots PR is open. with some automation, check the original Figma design, and compare it, and there are, there are, um… ways to improve this. So this is just a starting point. And last thing I want to mention, it was part of the tech design to track adoption of this. So far, I'm the only one contributing, which is why I'm doing this demo as well. But I created with Claude an automation that every week checks the repo. It's very easy, because we can just go to the insides of this, um… of this repo and see contributors, but also just check, uh, the asset and snapshot, uh, usages. So this is a dashboard that I shared, uh, within DuckDuckGo. It's updated every week, and basically we can see how. how it grows over time. Um… So yeah, if you fancy, please use it. And if there are any issues, let me know. I'm like the unofficial or official DRI of this. Happy to be on top of it if it's causing issues as well.
|
||||
- **20:54:59 · Jonathan Jackson:** It's really cool, Pat. Thank you for sharing. Yes.
|
||||
- **20:54:59 · DR:** Good work, pal.
|
||||
- **20:55:04 · Jonathan Jackson:** Very cool. We are now at 5 minutes to the top of the hour. Dominic and Alex, you both have postmortems on the list. Do you want to… run those now, understanding we'll go past time, and anybody who can stay is welcome to stay or not? Or would you rather find another time for those?
|
||||
- **20:55:24 · Dominik Kapusta:** I'm fine running it. It should also be short, but not sure about Alexa's one.
|
||||
- **20:55:33 · Alex Martemyanov:** Probably can wait till the next Apple Weekly.
|
||||
- **20:55:37 · Jonathan Jackson:** Okay.
|
||||
- **20:55:37 · Alex Martemyanov:** I think it was waiting for a year and it didn't wait another week.
|
||||
- **20:55:41 · Jonathan Jackson:** All right.
|
||||
- **20:55:42 · Alex Martemyanov:** Thank you.
|
||||
- **20:55:43 · Jonathan Jackson:** Thank you, Alex. Um, then, Dominic, why don't you go ahead and get started? Anybody who's not interested in Dominic's incident post-mortem, obviously, is welcome to drop. If we run a little bit late and you need to be somewhere else, that's also okay to drop. And Alex, I'll keep yours on the agenda for next week. Thank you.
|
||||
- **20:56:01 · Dominik Kapusta:** Thanks. I've dropped the link in the agenda as well for the incident postmortem. What happened last Monday is we released the app that was crashing on macOS 12. And, uh, the app was out… well, uh, so the app was out for, like. 3 hours before we, uh, realized that there is a crashing scenario on macOS 12, and then that was handled real quick, uh, with app store rollout stopped. And, uh, Sparkle rollout stopped, uh, as well at some point, um, and the fix was released, actually… well, it… It. Well, almost immediately on DMG, and the next day on App Store, because of the review time. Um, but, uh, how it happened? So, it happened, uh, because we had a change to, um, the save credentials pop-up, which shows up every time you're offered to… you're being offered to save a password. Um… On macOS 12, um… And that change was due to dropping a storyboard, replacing a storyboard with a code-initialized view. So what didn't go as planned? This, uh, change was not tested on old OS versions before merging, and it was strictly macOS 12 issue. Uh, like I mentioned before in this call, we had that engineering expectation in place, but we're not really enforcing it for every change. There's no way of enforcing it, uh, and there's no way of. testing, um… all those versions in CI, so… that's why it happened. Really. Uh, the other thing is, uh, we didn't know about the crash, because, uh, I overlooked it in Sentry. I was the release DRA and I overlooked it in Sentry. There was a single crash on macOS 12 there. Uh, that I treated as noise, and I went ahead and released the app. The sentry summaries weren't generated on that day, because we had another… Whoopsie. Um, the week before, when we updated, uh. Well, some updated the model version and I reviewed it and approved it. and we merged that change, and we updated the model version, but we didn't update the SDK version, and the old SDK didn't… wasn't recognizing the new model, and it failed to… collaborate, so we didn't generate summaries. Um… So that's the other… the other thing. Uh, when the… but! When the summary was generated, uh, it clearly pointed out that there is a crashing scenario on Mac OS 12, which started the incident. A week of internal testing didn't catch a relatively easy-to-reproduce crash. It was a 100% crashing scenario, which wasn't caught. Because we don't have testers on macOS 12. Well, apparently we have one, because we got this one crash at some point. I don't know who it is. I'm…
|
||||
- **20:59:33 · Alex Martemyanov:** Yeah, and by the way, there was another kind of related issue about the crash on Intel machines. It's not directly related, but yeah, it was close.
|
||||
- **20:59:41 · Dominik Kapusta:** Yes, another one, exactly. We also don't have testers, or many testers, on x86. architecture. The App Store review took 15 hours, which is rather long for an expedited review, but also there's nothing we can do about it. At least it got in the queue… I mean, it got in the review pretty quickly after requesting a completed review. What went well is that the sentry summary that was eventually generated, um… Clearly identified the crash, and suggested addressing it, and the fix was simple, quick to prepare and test, it was just assigning an initial frame to a view. And the hotfix release automation workflow less at this time. It's always important to point this out, because… We only talk about CI when it's not working correctly. Despite the long review time, we only had 37 crashes in total and only 9 affected users. That's all. for action items and DRIs, we actually, uh… there was an action item to update agent's instruction prefect… essentially update prefect to ensure that Claude agent SDK version is… compatible with used models. Uh, Sam has handled it, and, uh, well, there is a code command right now to, to update the SDK version whenever you update the model version. Um… The other topic that's not actually mentioned here is what to do about testing on older OS versions, but this is something we will be handling separately through Alex's work, mentioned earlier on the call. And… That's it. Finally, I created an incident task through the form. Asana forum, and marked the incident as high, because it's a crashing scenario, we need a hotfix, and an SRE person joined the call, and we talked a little bit, and he, um… Explain to me that if it's a one-person rollout. It's not a high severity incident, and we dropped it to medium, and they dropped the call. And that's it. And, um… That's something to keep in mind for later, for next incidents, when they are only starting rolling out, we may not need SRA help there. This is all I have.
|
||||
- **21:02:09 · SS:** Regarding the action item, are you happy enough with just having the code comment now that reminds about the Claude SDK?
|
||||
- **21:02:15 · Dominik Kapusta:** I think so.
|
||||
- **21:02:16 · SS:** you know, it's… it's not bulletproof, like, what I… what I wanted to do was to just ask the Claude SDK, what models do you support, and check that it's in the list, but it only includes… yeah, it doesn't include all the actual supported models.
|
||||
- **21:02:18 · Dominik Kapusta:** I think, yeah. I saw that, yeah.
|
||||
- **21:02:28 · SS:** Eh, maybe the comment's okay for now. I tried to make it very obvious what to do and how to confirm this.
|
||||
- **21:02:32 · Dominik Kapusta:** Yeah, I mean, that's okay. Also, that comment would show up, because it's just very close to the line that is changing, it will show up in the GitHub UI when reviewing the code, so it should be fine.
|
||||
- **21:02:37 · SS:** It's right there. Yeah, the only way I was considering extending this was to just have a GitHub Actions check that looks for model changes and adds a comment, GitHub Actions comments. But I would want to do this for all of Prefect. Whatever. I'll look at that.
|
||||
- **21:02:56 · Dominik Kapusta:** Yeah, yeah. Should be fine. Okay.
|
||||
- **21:02:58 · SS:** Okay.
|
||||
- **21:03:00 · Dominik Kapusta:** Thank you. Thanks for handling that, Sam. Alex?
|
||||
- **21:03:05 · Alex Martemyanov:** Yeah, I got a proposal in… I got it related to post-mortem of the VS Code, but I think it more suits here. What about adding… updating our process to require adding UI tests in addition to renting some older Mac to run older OSs, but make a requirement to the process to cover any UI change. with a UI test. Even if it's a fix or a quick addition in the uncovered region, especially having all the Agentic workflows, it may be a quick and cheap option. I mean, cheap to add. maybe not to maintain. But it adds, of course, it adds extra cost for the runners. Or maybe we could have like a test suite number two that runs weekly, for example, and not daily.
|
||||
- **21:04:11 · Dominik Kapusta:** I actually think it's… Some Windows team uses something similar. whenever they change UI, they add UI tests, and they have plenty of UI tests, and that's another problem. I believe, but… Yeah, we could… We could discuss it, I guess.
|
||||
- **21:04:31 · Alex Martemyanov:** Yeah, because, uh…
|
||||
- **21:04:31 · Jonathan Jackson:** I think that's what we're considering. Sorry, go ahead.
|
||||
- **21:04:35 · Alex Martemyanov:** Because obvious, like, obvious… thing that could fix it is, uh, like, QA, which we are missing, for… everybody knows for… for why, for what reasons, uh, but we have a lot of, uh, supported platforms, a lot of supported macOS versions, and, uh, it's, uh… Like, uh, it's impossible to… make the requirement as in the mentioned. Uh, process, like, to test all the supported macOS versions, because not everybody even have all these machines that, like, I have the older one, but it's not DuckDuckGo, it's my personal machine. And, uh, I can test it, but it takes time. Uh, and nobody… nobody does it, frankly. And we have no QA, and it's impossible to guarantee the quality in such conditions. So, maybe it's required. It should be a required condition to… to… raise the UI test coverage, the end-to-end coverage, so we can validate we are actually producing quality UI tests. Stuff.
|
||||
- **21:05:50 · Jonathan Jackson:** Yeah, I think that's worth considering alongside the virtual machine testing stuff that you were talking about earlier, right? I mean, we've known for a long time that we clearly have testing gaps, especially on older versions and older hardware. And… To this point, I think we've just accepted that as the way it is and every now and then something like this goes wrong and we deal with it. um… But we should definitely think about whether there's a promise we're making to our users there that we're not able to fulfill properly. And I think certainly that's true right now. The question is, what does it take to do better at that promise? So I'd love to see some proposals around moving some of these pieces forward.
|
||||
- **21:06:32 · Dominik Kapusta:** All right, that was all for me. Thank you very much.
|
||||
- **21:06:38 · Jonathan Jackson:** Alright, and thank you. Um, so, Alex, we're gonna wait till next week for your other postmortem. Thank you for your flexibility there. And thank you all who stayed to finish that out. I appreciate it. And I look forward to seeing you again soon. Bye, everyone.
|
||||
- **21:06:54 · Alex Martemyanov:** Thank you. Bye.
|
||||
@@ -194,8 +194,8 @@
|
||||
"note_modified_at": "2026-10-20T15:08:30.724Z",
|
||||
"source_url": "https://us01mynotes.zoom.us/gBBa-iiUQLycw55WRV1tnw",
|
||||
"folder": "2026-10-20 Apple Weekly Team Meeting - Zoom ID 2026-10-20 20 01(GMT+6 00) gBBa-iiUQLycw55WRV1tnw",
|
||||
"sha256": "5757a175b493ba2bee769a1bcd33431e752ce73ed296d817ccf8946e60c6dbad",
|
||||
"cues": 137
|
||||
"sha256": "346d4bd302bfc92a437c43ddc0f16d64aa76fd67cdf88603598958c7c785ad2a",
|
||||
"cues": 3
|
||||
},
|
||||
"7JZgnceAT1aZ2H75ok4dmA": {
|
||||
"title": "Costas Brammer's Personal Meeting Room",
|
||||
@@ -218,8 +218,8 @@
|
||||
"note_modified_at": "2026-10-23T15:29:29.141Z",
|
||||
"source_url": "https://us01mynotes.zoom.us/bPprGg5DTye9H-RQXkwMuw",
|
||||
"folder": "undated All-Company Meeting bPprGg5DTye9H-RQXkwMuw",
|
||||
"sha256": "92c8cdc1104e86fdbead212d4f69caf87623d332e2ad67f7807cdc248acd9fb3",
|
||||
"cues": 50
|
||||
"sha256": "67d79d187f34012b3d4e50ea54ba7648c909bfe24d067a4bc04ba72cd8f9cbb6",
|
||||
"cues": 3
|
||||
},
|
||||
"r6weCpT7SnasTUiybJW95w": {
|
||||
"title": "Apple Weekly Team Meeting - Zoom ID",
|
||||
|
||||
@@ -1 +1 @@
|
||||
{"observed_at":"2026-11-16T10:03:14.154Z","scanned":29,"imported":0,"skipped":29,"withoutTranscript":0,"coverage":"complete","sections":[{"name":"Owned by me","pages":5,"notes":84,"complete":true,"terminal_cursor":true},{"name":"Shared with me","pages":1,"notes":0,"complete":true,"terminal_cursor":true},{"name":"Starred","pages":1,"notes":0,"complete":true,"terminal_cursor":true}],"discovered":84,"gaps":[],"policy":"reread_all_owned_shared_and_starred_notes_since_cutoff","since":"2026-09-30"}
|
||||
{"observed_at":"2026-11-16T10:24:34.936Z","scanned":29,"imported":2,"skipped":27,"withoutTranscript":0,"coverage":"complete","sections":[{"name":"Owned by me","pages":5,"notes":84,"complete":true,"terminal_cursor":true},{"name":"Shared with me","pages":1,"notes":0,"complete":true,"terminal_cursor":true},{"name":"Starred","pages":1,"notes":0,"complete":true,"terminal_cursor":true}],"discovered":84,"gaps":[],"policy":"reread_all_owned_shared_and_starred_notes_since_cutoff","since":"2026-09-30"}
|
||||
-235
@@ -23,241 +23,6 @@
|
||||
"time": "21:31:17",
|
||||
"speaker": "Zach",
|
||||
"text": "Yep. Uh, yeah, sorry. Um, yeah, so just quick announcement there. Uh, Julio's moving to the, uh, DRI for that, um, and Mario is going to be, uh, the secondary. Um, so they've both been pretty heavily involved in, uh, that process, and we're excited for this transition."
|
||||
},
|
||||
{
|
||||
"time": "21:31:40",
|
||||
"speaker": "Josh Frank",
|
||||
"text": "I believe I'm next. So hey, everybody. I'm taking a quick break from the gym where you can frequently find me to tell you about random access learning. Just put the link in the chat. We have a session Thursday next week. 12:00 PM Eastern time, I think that's US Eastern, on corporate sustainability and being carbon negative in the AI era. So please, everybody, come to that if you can."
|
||||
},
|
||||
{
|
||||
"time": "21:32:17",
|
||||
"speaker": "AC",
|
||||
"text": "Everybody Ă last week we had our away mobile browser experience. Assessment our goals days a same they car our Browser mobile, browser experience from good to great in the previous cycle. We ship the android. And dialog."
|
||||
},
|
||||
{
|
||||
"time": "21:32:55",
|
||||
"speaker": "AC",
|
||||
"text": "Internal the keyboard now always on the new tape. Device to device migration. We want to ship this before black friday. A ui that s case for split. Screen. And we want to what support for many bar. We are going to start also the new tupage experiment for new users. And they think next these mark."
|
||||
},
|
||||
{
|
||||
"time": "21:34:04",
|
||||
"speaker": "Mark Ingram",
|
||||
"text": "Hey, everyone. I'm Mark from OL, the Desktop Browsers Objective. Some exciting news this week is Ivy is taking over as the ODRI, and I'll be moving on to the advisor role. I'll keep it short because I know we're on hack days this week, but I also, just before introing Ivy, I want to announce my new midlife crisis to everybody, which is. I decided to start learning the drums, and I can see Carl's excited face on the screen. Okay, that's my midlife crisis update. Over. Ivy, do you want to say hi?"
|
||||
},
|
||||
{
|
||||
"time": "21:34:31",
|
||||
"speaker": "I(",
|
||||
"text": "Yeah, I'm just really excited to take over OL, and stay tuned for some exciting news about extensions on macOS coming really soon."
|
||||
},
|
||||
{
|
||||
"time": "21:34:43",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "All right. Congrats, Ivy. That's exciting. So I'm kicking us off today. I'm going to actually share my whole screen because I'm going to be moving from one side to the other. So I have two demos today. So we have the Julia special, two demos. Um, my first demo is… since the meetup, I've been thinking about how, um… We need to, like, I've been thinking about the new objectives that we have about word of mouth and how we talk about DuckDuckGo and how, especially for ducklings, is hard to answer some of the questions that we get once you first join DuckDuckGo. So I wanted to create something that helps them answer those questions and understand our products better. So I do want to shout out Ben Smith because he created this. I just basically said the idea. Um, so this is an artifact, um, and you will see the different tabs here, so you have different questions per tab. Um, and then you have the specific questions here, which you can click on and see them in writing. Or, if you want to, you can actually listen to it as. kind of like a podcast episode. Um, so… let me know if you can actually hear this. I'm just gonna play a couple of seconds. Episode 1. The Big Picture. Last week, a friend of mine looked up, is it normal for your knee to click? Just once. For the next week, every website they visited tried to sell them knee supports, walking sticks, and one of those… So you get the gist of it. I hope you got to hear that. Um, so this is still a work in progress. We're going to be creating this content. Um, I really didn't want to… to be in audio, because I feel like if you get that in writing, it's different explaining as a human. So, this is what we're building. And then, the second thing I'm building… dang, the Zoom… The Zoom thing doesn't let me go to the other one. Okay, here. So this is the second thing that I built. This is coming from feedback from people who went on leave, which was around how to catch up with everything that had been happening since they were gone. Um, so this is a how… what do you miss? Uh, so basically, you add here the date you started your leave, and then the start of the date you return. Um, you click catch me up. Uh, it takes a couple of seconds, but it basically gives you, like, a summary of a bunch of things that happened while you were out. So, um… company-wide posts, uh, what was published, what was answered in odds and ends, so all the Q&A, which objectives are up-to-date, like, which assessment was the last thing, and what we're working on right now, uh, things that were shipped, uh, also, like, new AORs, retired AORs, so here, as you can see, like, one, um, sample of, like, the announcements. It gives you just, like, a summary of all the announcements. Um, and we also have a people tab, so, like, new hires, which teams they are in, uh, people who left, and then it also shows you, um, green bow ties, spotlights, in case you want to catch up with all the spotlights that were posted that. Um, the time you were away. I do want to add a tab for specifically functional team updates, so, like, if there's a new process or anything specific to the functional team, I'm just not sure yet where to get that information from. Um, so… working on that. If you have suggestions, or if you would like to be… try out these things, let me know. I would love the feedback. Um, and now, this is it for my demos. Now I have a demo by, um, Robert, who couldn't be here. So, also, please ping me if you can't, uh, hear this. Hello, hello, hello. I worked on three hacks this time around. I'll try to get through it all quickly. The first is to do with increasing Duck AI prompts. I thought that maybe a nice way to do that would be to add a bit of discoverability into our browser. So if you are looking at a foreign language page, then you get a suggestion to translate it. Can someone, uh, unmute themselves and just let me know if you can hear this? Because I'm, like, concerned because I can't see that. Okay, thank you."
|
||||
},
|
||||
{
|
||||
"time": "21:38:59",
|
||||
"speaker": "Anne Hildebrand",
|
||||
"text": "Yes."
|
||||
},
|
||||
{
|
||||
"time": "21:38:59",
|
||||
"speaker": "Zaahir Moolla",
|
||||
"text": "We can hear."
|
||||
},
|
||||
{
|
||||
"time": "21:39:02",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "And, um, so yep, that sends the page content off into Duck AI and translates it. We already had this feature, but, you know, just adding the prompt up there. That works. And there's lots of… this is very exploratory. There's lots of different ideas along these lines that we could look into, such as, if it's a long article, maybe we, um, tell the user how long it'll take and prompt to summarize, or on a shopping page, we could prompt them to, um, compare. Uh, the cool thing is that these suggestions here, these three options, that's all generated by our local LLM, so it's completely private. Um, nothing's, uh, hitting our servers until you click on one of those. Um, most recipes too. And this is a cool one, if you, um, are looking at an event, you can ask DuckAI to compute the event details. And I added this, um… dynamic UI, this card at the bottom of the Doug AI answer, and it gives you a button to add it to your calendar. Uh, one cool idea is, if you're looking at a privacy page, it could offer to help the user understand what it is they're agreeing to, so very within our brand. And another cool idea, just because YouTube features always seem to perform really well for us, is we could, um, prompt the user to, uh, do things using the transcript that appears in YouTube here. So this actually extracts the transcript from this panel. And it can send it to DocAI. Okay. Hack number two is, um, fixing a personal frustration of mine. So, I'm a Raycast user, and I'm always doing things like… um, converting currencies and, like, complex queries like this. And, uh, it frustrates me to no end that, um, that doesn't work when you plug it into DuckDuckGo. And so what I did is I had Claude reverse engineer the raycast calculator, and I've plopped that into SERP, and so now you can, um, jump into SERP running on this test machine, and do complex queries like that. And there's all sorts of things you can do. Mixed currencies is kind of cool. Um, one thing I use a lot is, like. Adding different units. And there's also, like, sort of… Complex math bits that are kind of fun. And the way that I handled this in the UI is that —. The simple calculator stays if you're doing a simple expression. And same if you're just doing a regular. conversion, then we get the existing UI for that too, because it's nicer since it has the dropdowns. Okay, and my third hack is yet another frustration that I've had for a while. So here we are looking at my phone. Oops. And, um, so we've always been able to switch tabs by swiping on the address bar at the bottom there, and that is quite nice. But what we haven't been able to do is swipe up to go to the tab switcher. And so I added that, and it turns out that Bartosz had already been working on this, so he has helped me a lot with this. Thank you very much. And yeah, it's really nice that you can… it tracks your finger, so you can kind of drag out halfway. Real smooth. And Bartosz added, you can drag out halfway and swipe to the right. Those are my hacks. Thank you very much. All right, and we're not done yet. I have another one by Pablo. When people want their data on another device they don't go looking for things and sync it back up doesn't really tell them what it does. Chanel's usability insight report made that very clear. And the problem is getting bigger. Zac AI chats, uh, can easily be lost. Um, the more stuff we add to Zac AI, um, the more. things that are worth keeping, Sink and Backup will just not be enough. So, we are proposing MyTacticGo, a private account. It's private because it's from us, and it needs no personal data to sign in, so no email and no password to remember. It is encrypted, your data is backed up, unsynced, and we show what's ready to go, like 100 passwords or 244 bookmarks, so the value is very concrete. On desktop, MyTacticGo lives in the toolbar, where people actually look. It lists what will be synced, like your chats and passwords and bookmarks, but it also brings your subscription. So, this can become, actually, your home base, and it shows your plan and your connected devices, so it is a great opportunity for… for dependability, um, and cross-promoting other… other features and understanding our… our ecosystem. All in one place. That's it for me, and I think we're. It's Andrea next."
|
||||
},
|
||||
{
|
||||
"time": "21:44:01",
|
||||
"speaker": "Andrea Soto",
|
||||
"text": "Okay, so I was thinking that we want to increase word of mouth, and stickers are a great way to do it. We also have our Easter egg. Um, logos, that would be great to… sell as stickers, so I added the option next to the logo when you click it to be able to buy the stickers on desktop and mobile. Or $2, so… That's my hack days project. Valent, you're next."
|
||||
},
|
||||
{
|
||||
"time": "21:44:35",
|
||||
"speaker": "Balint Farkas",
|
||||
"text": "Hey, folks. So I'm working on the DuckDuckGo Chromium browser, also known as Duckium. We use Agentic workflows all the time, and it can, of course, see the code, but it cannot see the internals of the browser. So it can see the code, it can drive the browser via Chromium DevTools protocol, but it cannot see our own services like Sync. pixels, burn, feature flags, and so on. So I've exposed them to AI. You can just connect to the browser from your own AI terminal and then ask stuff, for example, why can't I see sync and backup in the menu? It should be here. The answer is because the feature flag in this live session isn't on. Or you can ask spicier stuff, such as performance search for I love cats, which it can do via Chromium DevTools protocol, and observe what pixels have been fired, which comes from the internals observation that I just added. And you can correlate these things and observe the browser's behavior as it happens, which will enable us more sophisticated debugging scenarios, and what's more important, more sophisticated Agentic workflows. That's my first thing. And the second thing is for Mark Sabu, who is out currently, and I'm sharing his video in a second."
|
||||
},
|
||||
{
|
||||
"time": "21:45:52",
|
||||
"speaker": "Andrea Soto",
|
||||
"text": "Thank you."
|
||||
},
|
||||
{
|
||||
"time": "21:46:02",
|
||||
"speaker": "Balint Farkas",
|
||||
"text": "Should be running. Hey folks! I'll walk you through the life of a Windows Release DRI in the next two minutes. After the last quick one days, we iterated on the design together with the release team, and I spent the last two days building it. The Windows Release Command Center. Until now, all of this was manual. We kept Asana and GitHub Actions in sync."
|
||||
},
|
||||
{
|
||||
"time": "21:46:10",
|
||||
"speaker": "Andrea Soto",
|
||||
"text": "Ooh."
|
||||
},
|
||||
{
|
||||
"time": "21:46:22",
|
||||
"speaker": "Balint Farkas",
|
||||
"text": "and dispatch every workflow by hand. Now the ODRI can check the state at a glance, and the release DRI can get through their duties by following the steps in the app. This is the state of every channel, which build is where, and at what rollout percentage. It all comes live from Asana. No more opening 5 tasks and 3 workflow pages to figure out where we are. The calendar shows what's coming up, branch cuts, rollout steps, so nobody gets surprised by a deadline. Clicking the release opens the DRI view. It shows the next section and flags anything that needs attention. Say we want to take 177 to 100% stable. I click this button and it dispatches the workflow. Because it's at 100%, the app asked me to follow these steps and upload it to the Microsoft Store. But I'm actually the DRI of 176, and I will need to do a hotfix. Hotfixing used to be the most manual part. Now I just pick from develop. Like PES, a PR that needs to be hotfixed. and just displays the bot with creation, which will actually just do this. I already done that to not block the demo. Umm. So I will need to upload the release notes. It loads the release notes automatically. So I can just approve it and upload. And now that it's done, I can just go back and simply follow the remaining steps to finish the release. And obviously, every flow syncs back into Asana, so the runbook still works if the app is down. That was Release Command Center. I hope it makes release weeks a little calmer and gives every DRI back some time for actual work. Oh, and sorry about the editing. I'm a dev, so this video was shipped as an MVP. Thanks, Odo! That's it for me."
|
||||
},
|
||||
{
|
||||
"time": "21:48:10",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah, hey. Hey, everyone. This is me, Alex. There are so many new faces. I'm pretty sure some of you have no idea who I am. Which is fine, sometimes I'm not entirely sure myself. I'm Alex, from the Apple team, working on our macOS browser. And I thought I'd make it greater again. What would you expect from a macOS browser? During this hack… in the hack days. They say Julius was a hacker, performing live demos, chewing water into wine, accepting some cookies. Not bad for a hack days project, right? All right. Showtime. Could you please stop sharing?"
|
||||
},
|
||||
{
|
||||
"time": "21:48:55",
|
||||
"speaker": "Mark Ingram",
|
||||
"text": "Thank you, Alan."
|
||||
},
|
||||
{
|
||||
"time": "21:48:56",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Got it."
|
||||
},
|
||||
{
|
||||
"time": "21:48:57",
|
||||
"speaker": "Balint Farkas",
|
||||
"text": "Yeah, yeah, yeah, I'm trying, uh…"
|
||||
},
|
||||
{
|
||||
"time": "21:48:59",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "It's a very…"
|
||||
},
|
||||
{
|
||||
"time": "21:49:03",
|
||||
"speaker": "Alex Martemyanov",
|
||||
"text": "Yeah. Okay, first try failed miserably. Let's try again. Yeah, Kira. Hello. I hope you can hear the sound. Yeah, hope you can hear the sound. But seriously. What's happening to the world? AI is everywhere, too much information, too much liquid glass and beige. Sometimes I just want to burn it all down. Like this. Yeah, much better. Let's do it again. Where shall we start now? Sometimes I just want to run away back to something familiar. Or maybe look for some happiness. Or maybe I just need to talk to somebody. Yeah, maybe the problem isn't the internet. Maybe it's just how we see it. I wish I could change the whole settings of my life. how it appears. What if we could go back? Somewhere like this. And buttons look like buttons. Windows look like windows. Smells like Christmas morning, hot chocolate and mandarins, back when everybody was still around. Remember, on the websites. had some personality. When every page told you someone had paid attention to it. And speaking of the little things… You see? Even the tooltip. Respect your personal space. Your privacy. You just need to start and ask for some help. Here we are. Your protection, your privacy, your peace of mind. Thank you. That's all from me."
|
||||
},
|
||||
{
|
||||
"time": "21:50:57",
|
||||
"speaker": "Deepak",
|
||||
"text": "I'll go next. Let's add the photo. Cool. Clearing my screen… What? I didn't work. Trying again. Okay. Can you all see my screen now?"
|
||||
},
|
||||
{
|
||||
"time": "21:51:30",
|
||||
"speaker": "John Boileau",
|
||||
"text": "Yep."
|
||||
},
|
||||
{
|
||||
"time": "21:51:31",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "We can see it."
|
||||
},
|
||||
{
|
||||
"time": "21:51:32",
|
||||
"speaker": "Deepak",
|
||||
"text": "Okay. Cool. Playing a video, if you can't hear the audio, please. DuckyEye can generate a lot of great images, but only if you know exactly what to promote. ChatGPT, Gemini, and other tools have solved this with pre-existing templates for images, which has also led to a lot of virality for them. Well, now we have them too. If you go to new images, we have 15 of these different templates ranging from anime to 80s portraits to caricatures. Clicking on any of them populates the input field with the corresponding prompt. Some of them require you to upload a reference image, which is indicated by this icon, and then a user is notified to do that with this tooltip. I already did this one. So if you look at this, this was the reference image and I think it did a pretty good job. We also prompt the user to try more styles, so they can click on any of the, uh, styles, and then the original image gets, uh, pre-filled here with the. Uh, a defense bill. Another thing that really irked me was that after all of these images were generated, they were scattered across charts, and there was no one place to see them. Well, now we have this My Images tab. where you can see all of the different images that we have generated. You can delete one or all of them from here, and you can also go to the chart, where. they were generated, or use them as a reference for your next image. One cool thing that we do here is, if you do delete this image, then in the original conversation, where this was generated, we have this image deleted placeholder here, so that the continuity of the chat is not lost. That's it. Thank you. Yeah, that's it. Thank you."
|
||||
},
|
||||
{
|
||||
"time": "21:53:20",
|
||||
"speaker": "DR",
|
||||
"text": "All right, so I go next. Let me see. Here. Can you see the screen? All right. So, uh, starting next Monday, our internal release, actually internal users, uh, in the internal release, will have, uh, the possibility of installing from the web, uh, Chrome Web Store 1Password. And this video showcases how it works, that it connects to the desktop app that you have installed. You can see that it works for registering your email and autofilling your password for it. You have to unlock it first, though. You can see here how it works with Touch ID. Um… You can fill the login, password. Then it's going to ask you to save it. You can also, of course, use it to log in to a site, those details. So you guys see here. Then it also works for passkeys, where you can register a passkey for one of the accounts that you already have in 1Password. Um… Again, it's going to probably ask for authentication. And then use that same passkey. To log into the site. In this example, it didn't ask for additional authentication because it was so close to the previous one, but it's working quite nicely. And we'll be posting an announcement with details on how you can provide feedback for this. I said it's quite complicated on the technical side. Thank you. Hold on."
|
||||
},
|
||||
{
|
||||
"time": "21:54:58",
|
||||
"speaker": "AO",
|
||||
"text": "Hey, Elle. This is a demo that Alex, Elle, and I did. So double Alex demo. or DACAI artifacts. Hey everyone! During Hack Days, we explored DACAI artifacts. Documents, presentations, and visuals. But you know what? It's boring, so here's the twist. This presentation was built entirely in DACAI. See? It's right here, inside the chat. And you know what? I can even download it as a PowerPoint file. Now check this out. Let's ask how much I could save. I get an answer, but only for this one case. Different numbers, we need to try again. Or I can get an app that does that for me. Duck AI builds it right here, next to the chat. You can watch it come together. Not a code block, not a screenshot, a real working app. Now I can change the amount, the time, the rate, and it updates instantly. Have a PDF? You can turn it into just about anything. Study notes, a brief, key takeaways, whatever you need, as a document. You have full control over your doc and markdown. You can also let Duck AI rework it. When you are done editing, you can now export your doc. With Artifacts, you can build your own website too. If you want something different, you can just point DuckAI at it and ask for an edit. Every version is saved, so if you don't like your change, you can go back in one click. Artifacts also work well with our browser features. Let's say I have a really long article, and absolutely no desire to read all of it. Attach the top, and Dakia will summarize the main takeaways visually. Just a few seconds, and it's ready, clear, structured, and much easier to understand. And of course, artifacts work great on mobile too. Try them yourself! That's it. Thank you."
|
||||
},
|
||||
{
|
||||
"time": "21:57:18",
|
||||
"speaker": "Gary K",
|
||||
"text": "All right, I'm up. Let me share with sound. a video… Here we go. Any time there is an incident, the first question is always, what changed? And the answer is potentially in a lot of different places. So I've created a command line tool that will aggregate the events that are likely to be interesting for anyone investigating new behavior. What changed is a tool that can access Azure activity logs, Asana deployment plans, Jenkins jobs, GitHub pull requests and actions, REX commands, and Mattermost notifications. Because there can be a lot of data in these sources, the tool ships with default settings that are designed to give high signal-to-noise ratio. These settings can be overwritten via a config file. or a runtime flag. Users can search backwards from now or over a specific window of time, from minutes to days. There are a host of flags designed to make it easy to use. Users can opt in to format the output as JSON and perform further filtering. and transformations using JQ. What changed uses your own credentials, so security posture is maintained. What Changed is available now at www.duckduckgo.com/gcurtis/whatchanged. Thank you. Like and subscribe for more great content. Thank you."
|
||||
},
|
||||
{
|
||||
"time": "21:58:52",
|
||||
"speaker": "Tom Strba",
|
||||
"text": "Nice. Hey everyone. So over the last couple of weeks in OJ and OM, we have been trying to build tools for Duck AI. What does it mean? We would like Duck AI to be able to get context automatically from your browser. And not only that, we would like to. Duck AI to be able to control the browser. So, during hack days, I tried to put all the pieces together, and uh… I would like to show you a proof of concept that I managed to build, so… Let's say I have, uh… Couple of, uh… shoes I am trying to buy. And I would like Duck AI to recommend me which shoes should I buy. So… I will ask you. Which shoes should I buy from all these options? And I'm not providing more context. It will basically get the list of my tabs. It will read the content and it will recommend me, oh, this is the best for speed for daily trainer. whatever but this is not where the fun ends this is where the fun starts so how about now we ask Duck AI. to buy our choice. So for example, buy me stride arrow three. And Duck AI will basically open a new tab and start buying shoes for me. As you can see, here is the store and it's starting to fill the form and it doesn't know which. shoe size it should buy, so it asks me. I will just respond, for example. 10, and it now reads the form. As you can see, it filled. The 10, and now there is a… the website needs the address. So the fun part is that we already have the identity in the browser. So Dakia asks me, hey, can I use this identity from the browser to fill it? And yes, of course, let's share it. If we take a look at the store… by the way, this is the store I built for myself for testing. The address is now filled. and it asks me, it asks me last time, hey, how about, uh, I'm gonna, I'm gonna submit the form, are you, are you sure? Yes. I am sure. And my shoes are now ordered. I didn't need to click on that website. I just wanted you to see what is happening in the background. But Duck AI now or soon will be able to buy. stuff for you or do anything in the browser you want it to do. Uh, that's the first demo I wanted to show you. And the second one, this one will be really quick, and it's, like, internal tool enhancement. Uh, if you… Often test a Mac. make ship reviews, you know the process. You need to download the DMG, attach it manually. Then drag and drop it. The instance is running, so you need to quit it. Then drag and drop again, wait for installation and then run it. So I build the button and now you will just click to run and you can start reading the. ship review, and everything will happen in the background, and after the build is downloaded, I'm gonna speed this up a little bit, it will basically just runs, and you can start testing. Um, that's it for me, thank you."
|
||||
},
|
||||
{
|
||||
"time": "22:02:22",
|
||||
"speaker": "I(",
|
||||
"text": "All right. I'm just getting my screen up. Please let me know if you can see it. All right, looks like it's there. All right. So as a user, why should I care about the protections report on our browser? It's a list of numbers and icons. What is the actual benefit? So I reworked it. I started with the bottom line. Companies are watching you. It's creepy. Also highlighted right away that there's an issue to be fixed. Privacy isn't your default, and here's why that matters. The more of your everyday browsing happens in our app, the less these companies get to see. So let's expand this. Immediately, we get a creepy, interactive, prying eye that follows your cursor around. And we get useful info. We block tracking attempts, but why? They target you with ads, they watch what you read, and they record what you click. It's creepy. So who's watching? Well, Google does a lot of spying, and it was on 4 out of the 5 sites he visited. And Facebook was on almost half. Again. It's really creepy. So I wondered how we could make this more like a personal infographic. Here we highlight the nosiest site of the week, and we show a ring of trackers from that site that blink like eyes to drive home the visceral idea that you're being watched. Now, we empower you to understand some of what we're doing by showing you how we prevent a digital fingerprint view from being created. You can hover over each component of the fingerprint and see what we've hid from creepy companies. And how many times? And finally, we contrast this with Chrome. Where companies swarm around your cursor, and cookie pops and ads fill the screen. We drive home the fact that Chrome is owned by Google, and that unlike on Chrome, DuckDuckGo blocks Google's tracking attempts. So maybe this really creeped you out. Why are so many companies watching you read the news? Let's share that. And just like that, a promotional infographic can be shared with a link to download our browser. So try it out for yourself on Windows on the link task. Thanks."
|
||||
},
|
||||
{
|
||||
"time": "22:04:41",
|
||||
"speaker": "Vinay Nadig",
|
||||
"text": "All right, that's me next. Let me share my screen. Alright, so for hack days, I have. built an app called, QuackTime. What it does is, it lets you schedule ad hoc meetings, through Mattermost without having to do the whole song and dance of matching time zones across. uh, different geographies. So, to use that, go to QuackTime, go to Metamost, or any DM, QuackTime and DM duration, in this case 60 minutes, and then we'll set a type build. You will get a link, you can open it. you will end up on this interface, basically. You will see that all your, uh… uh, working hours and free, uh, slots through Outlook has already been filled in. And that would be the same case for another person as well. So, the other person will also open the same link. And yeah, so you can see that I'm working with Adrian, so I have this much overlap with him, so you can select whichever task that you want. And, oh, sorry. You can propose it. And the other person will… See an option to accept that. And once you accept it, an invite will be sent automatically at the correct time, along with the Zoom link for… which uses your personal Zoom link. That's pretty much it. I'll stop there."
|
||||
},
|
||||
{
|
||||
"time": "22:06:17",
|
||||
"speaker": "Nicklas Jensen",
|
||||
"text": "All right, I'm next up here, and I am going to be sharing some porn. out of a conversation, actually, in a Zoom chat. So, one of the things on the meetup that happened was that we, uh… At least for a lot of the ducklings, we were freaking out about how much merch and how many things have been over the years. So the idea was to create a hall of fame for specifically T-shirts. So Miguel and I worked on that. For this, uh… For these hack days, the goal is you have a timeline, you can see all of the different designs that have been over the years. These are all the ones that Miguel could get a hold of so far. You can vote for your favorite ones. And also, if you don't feel like going through them all manually, you can just… pick out of sort of a random generated thing. And yeah, that's the T-shirt Hall of Fame. How do I stop sharing? Thank you."
|
||||
},
|
||||
{
|
||||
"time": "22:07:30",
|
||||
"speaker": "Dominik Kapusta",
|
||||
"text": "All right, here's a demo of the… Long overdue feature on… macOS, which is the reader mode. Um… Reader mode. extracts text from websites. And here you can see that, uh, we're navigating to a website that, uh, is eligible for being displayed in reader mode, which contains relatively more text, uh, and, um… Just with a single button in the address bar, um… we turn on the reader mode, then we navigate away from that website, it just resets to a regular view. here, uh, you can see, uh, visiting another website, uh, that's also eligible for reader mode, and here's, uh, here's how it works, but it's not all, because I've worked on that, uh, during Hug Days, but. Um… a parallel to my work, actually. Marios was doing the same thing for iOS, so here it is, um… Here, the reader mode, uh… Button is in the website, uh… menu… And this demo will also show you that it respects the browser theme. So that, um… shows correctly in the dark theme as well. Even if you visit a website that… uh, doesn't honor the system theme, uh, then the reader mode, of course, uh, will be dark in this case, when the Guardian was in, uh, in light theme. Yeah, and this all was possible. to do thanks to a library by Meek that extracts. text from, from websites. So really the integration. It was fairly minimal here, and it was just a bit of work. That's why I said it was long overdue, because that library was published a few months ago already. That's it. Thank you."
|
||||
},
|
||||
{
|
||||
"time": "22:10:01",
|
||||
"speaker": "Valerie Kraucunas",
|
||||
"text": "Alrighty, hello. Um, so, I'm Valerie. I work on the front-end team. Um, the thing I've been really heavily focused on all year is the rebrand, and that… with that has come… turning over, uh, style values from one… one way to a new way. Um… with that, we have these things called design tokens, which is basically, like, a codified design decision. And so, when we're styling… the new rebranded stuff we're using, or at least we want to be using all of these things. That said, um, a lot of times when you are handed designs that are rebranded. You end up with this scenario of… It'll be a mix of using the new style tokens and the old style tokens. So in this case, like the SDS styles are a legacy thing. Um, and I want to be very clear, this is not a shame. This is not a shame campaign or anything on designers, because oftentimes when they are building these things, they are… they have the old way in the same file as the. new way. And so it's really, really tedious to get down into every single nook and cranny and node or layer to make sure that you're using the new styles on everything and to make it even more complicated. So when developers see this, we just see this and we're like, why is this still here? Why are we still using SDS things? But when you get over into the design look, so this first column is when they're picking the first padding value, so space X07. When they're picking the second column value, space X03. They look the same. It's really easy to co-mingle the old style and the new. And so, um, basically my concept is to… be able to hand a Claude skill a Figma link, and have it go and traverse the entire. All child nodes of that, um… of whatever you linked over. So. I'm going to try this. I doubt it will… I doubt it'll work. It does not go as fast as I would like it to, but we'll give it a go. So, basically, what happens is you, you know, invoke it with the Figma link, then you give it an Asana link. And give it a name. Um… What that ends up spitting out is a nice report for designers to be able to go and update all the things that are using, uh, old values, or possibly things that are just out of system. As well, um, now I want to be clear, I'm standing on the shoulders of giants here, because, um, while we've been doing the SERP rebrand, uh, my partner in crime, Rachel Lim, created a… skill called Marlin Rebrando, which is basically a dictionary of, uh, knowledge for this was the old way, this is the new way, these are the things we should be checking for when we are executing the rebrand and the code. Um, so my skill is heavily relying on that, but, um… What it'll basically do is it will just… fully traverse like a node tree, if you will, all the way down. Um… And yeah, the things that we're looking for are the use of legacy variables. This is run from within the DDG. repo, for now, at least. So what it will do also is it'll look at the currently installed design token package and make sure that there aren't values being used in the designs that don't exist in. the installed package, then also looking for just colors that aren't tied to a system value, unselled text, et cetera. So what this ends up doing is it will create a. design token report, and. give the designers a link to the layer that. his, uh… has a property on it that's a problem, if you will, and then give it an idea of, like, this is an unexpected value, and then has an expected or a suggested value. The nice part about this as well is that developers. don't necessarily have to wait on designers to go. What?"
|
||||
},
|
||||
{
|
||||
"time": "22:14:28",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "Well… I'm so sorry, I hate to do it, but I have to interrupt you, because we're at 4 minutes now."
|
||||
},
|
||||
{
|
||||
"time": "22:14:35",
|
||||
"speaker": "Valerie Kraucunas",
|
||||
"text": "Oh, shoot, I'm so sorry. Okay, well…"
|
||||
},
|
||||
{
|
||||
"time": "22:14:36",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "Sorry."
|
||||
},
|
||||
{
|
||||
"time": "22:14:38",
|
||||
"speaker": "Valerie Kraucunas",
|
||||
"text": "Making everybody's life better, that's all."
|
||||
},
|
||||
{
|
||||
"time": "22:14:50",
|
||||
"speaker": "Stano",
|
||||
"text": "Okay, I believe I was next. Um, I'm gonna… I hope you can see my screen now. Someone please confirm."
|
||||
},
|
||||
{
|
||||
"time": "22:15:00",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "No, we cannot."
|
||||
},
|
||||
{
|
||||
"time": "22:15:01",
|
||||
"speaker": "Jason Ashby",
|
||||
"text": "Nope."
|
||||
},
|
||||
{
|
||||
"time": "22:15:02",
|
||||
"speaker": "Stano",
|
||||
"text": "You cannot? Let me try again. Where is… The sun got stuck."
|
||||
},
|
||||
{
|
||||
"time": "22:15:24",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "Did you find the share button?"
|
||||
},
|
||||
{
|
||||
"time": "22:15:27",
|
||||
"speaker": "Stano",
|
||||
"text": "Give me a sec, I reconnected. You can take it out of my budget, I'll make it quick. It does not want to work. You can skip me. Go on."
|
||||
},
|
||||
{
|
||||
"time": "22:15:45",
|
||||
"speaker": "Anh Do",
|
||||
"text": "Okay, looks like I'm next. So let me try to share my screen. Wow. Uh, how do I share, like, a… Particular windows, um… Sorry, um, I had so many windows opening at the moment. Oh, come on. Guys, it is not fun. Sorry. OK, I'm going to share my desktop then. One sec. Hopefully, you can see that on my screen. Can anyone see it? Okay. So for context, in OK, we are helping people in the US to remove personal information from data broker by submitting our request on their."
|
||||
},
|
||||
{
|
||||
"time": "22:16:21",
|
||||
"speaker": "Julia Abreu",
|
||||
"text": "I can see it."
|
||||
},
|
||||
{
|
||||
"time": "22:16:33",
|
||||
"speaker": "Anh Do",
|
||||
"text": "behalf. But recently, we have a lot of issues with broker block our archive with CloudFlare and CAPTCHA, so we can submit the form for the users. So this demo has been built based on Ben Smith design prototype. Let me play that."
|
||||
}
|
||||
]
|
||||
}
|
||||
-47
@@ -15,50 +15,3 @@ source: "https://us01mynotes.zoom.us/bPprGg5DTye9H-RQXkwMuw"
|
||||
- **21:29:37 · John Boileau:** And this way.
|
||||
- **21:30:59 · Gabriel Weinberg:** Hello, everybody. We have a few quick updates that we're going to mainly do the Hack Day's showcase today that Julia's leading. Everyday AI Guild. That's Zac. Okay. Yeah, there it is. All right. Yeah.
|
||||
- **21:31:17 · Zach:** Yep. Uh, yeah, sorry. Um, yeah, so just quick announcement there. Uh, Julio's moving to the, uh, DRI for that, um, and Mario is going to be, uh, the secondary. Um, so they've both been pretty heavily involved in, uh, that process, and we're excited for this transition.
|
||||
- **21:31:40 · Josh Frank:** I believe I'm next. So hey, everybody. I'm taking a quick break from the gym where you can frequently find me to tell you about random access learning. Just put the link in the chat. We have a session Thursday next week. 12:00 PM Eastern time, I think that's US Eastern, on corporate sustainability and being carbon negative in the AI era. So please, everybody, come to that if you can.
|
||||
- **21:32:17 · AC:** Everybody Ă last week we had our away mobile browser experience. Assessment our goals days a same they car our Browser mobile, browser experience from good to great in the previous cycle. We ship the android. And dialog.
|
||||
- **21:32:55 · AC:** Internal the keyboard now always on the new tape. Device to device migration. We want to ship this before black friday. A ui that s case for split. Screen. And we want to what support for many bar. We are going to start also the new tupage experiment for new users. And they think next these mark.
|
||||
- **21:34:04 · Mark Ingram:** Hey, everyone. I'm Mark from OL, the Desktop Browsers Objective. Some exciting news this week is Ivy is taking over as the ODRI, and I'll be moving on to the advisor role. I'll keep it short because I know we're on hack days this week, but I also, just before introing Ivy, I want to announce my new midlife crisis to everybody, which is. I decided to start learning the drums, and I can see Carl's excited face on the screen. Okay, that's my midlife crisis update. Over. Ivy, do you want to say hi?
|
||||
- **21:34:31 · I(:** Yeah, I'm just really excited to take over OL, and stay tuned for some exciting news about extensions on macOS coming really soon.
|
||||
- **21:34:43 · Julia Abreu:** All right. Congrats, Ivy. That's exciting. So I'm kicking us off today. I'm going to actually share my whole screen because I'm going to be moving from one side to the other. So I have two demos today. So we have the Julia special, two demos. Um, my first demo is… since the meetup, I've been thinking about how, um… We need to, like, I've been thinking about the new objectives that we have about word of mouth and how we talk about DuckDuckGo and how, especially for ducklings, is hard to answer some of the questions that we get once you first join DuckDuckGo. So I wanted to create something that helps them answer those questions and understand our products better. So I do want to shout out Ben Smith because he created this. I just basically said the idea. Um, so this is an artifact, um, and you will see the different tabs here, so you have different questions per tab. Um, and then you have the specific questions here, which you can click on and see them in writing. Or, if you want to, you can actually listen to it as. kind of like a podcast episode. Um, so… let me know if you can actually hear this. I'm just gonna play a couple of seconds. Episode 1. The Big Picture. Last week, a friend of mine looked up, is it normal for your knee to click? Just once. For the next week, every website they visited tried to sell them knee supports, walking sticks, and one of those… So you get the gist of it. I hope you got to hear that. Um, so this is still a work in progress. We're going to be creating this content. Um, I really didn't want to… to be in audio, because I feel like if you get that in writing, it's different explaining as a human. So, this is what we're building. And then, the second thing I'm building… dang, the Zoom… The Zoom thing doesn't let me go to the other one. Okay, here. So this is the second thing that I built. This is coming from feedback from people who went on leave, which was around how to catch up with everything that had been happening since they were gone. Um, so this is a how… what do you miss? Uh, so basically, you add here the date you started your leave, and then the start of the date you return. Um, you click catch me up. Uh, it takes a couple of seconds, but it basically gives you, like, a summary of a bunch of things that happened while you were out. So, um… company-wide posts, uh, what was published, what was answered in odds and ends, so all the Q&A, which objectives are up-to-date, like, which assessment was the last thing, and what we're working on right now, uh, things that were shipped, uh, also, like, new AORs, retired AORs, so here, as you can see, like, one, um, sample of, like, the announcements. It gives you just, like, a summary of all the announcements. Um, and we also have a people tab, so, like, new hires, which teams they are in, uh, people who left, and then it also shows you, um, green bow ties, spotlights, in case you want to catch up with all the spotlights that were posted that. Um, the time you were away. I do want to add a tab for specifically functional team updates, so, like, if there's a new process or anything specific to the functional team, I'm just not sure yet where to get that information from. Um, so… working on that. If you have suggestions, or if you would like to be… try out these things, let me know. I would love the feedback. Um, and now, this is it for my demos. Now I have a demo by, um, Robert, who couldn't be here. So, also, please ping me if you can't, uh, hear this. Hello, hello, hello. I worked on three hacks this time around. I'll try to get through it all quickly. The first is to do with increasing Duck AI prompts. I thought that maybe a nice way to do that would be to add a bit of discoverability into our browser. So if you are looking at a foreign language page, then you get a suggestion to translate it. Can someone, uh, unmute themselves and just let me know if you can hear this? Because I'm, like, concerned because I can't see that. Okay, thank you.
|
||||
- **21:38:59 · Anne Hildebrand:** Yes.
|
||||
- **21:38:59 · Zaahir Moolla:** We can hear.
|
||||
- **21:39:02 · Julia Abreu:** And, um, so yep, that sends the page content off into Duck AI and translates it. We already had this feature, but, you know, just adding the prompt up there. That works. And there's lots of… this is very exploratory. There's lots of different ideas along these lines that we could look into, such as, if it's a long article, maybe we, um, tell the user how long it'll take and prompt to summarize, or on a shopping page, we could prompt them to, um, compare. Uh, the cool thing is that these suggestions here, these three options, that's all generated by our local LLM, so it's completely private. Um, nothing's, uh, hitting our servers until you click on one of those. Um, most recipes too. And this is a cool one, if you, um, are looking at an event, you can ask DuckAI to compute the event details. And I added this, um… dynamic UI, this card at the bottom of the Doug AI answer, and it gives you a button to add it to your calendar. Uh, one cool idea is, if you're looking at a privacy page, it could offer to help the user understand what it is they're agreeing to, so very within our brand. And another cool idea, just because YouTube features always seem to perform really well for us, is we could, um, prompt the user to, uh, do things using the transcript that appears in YouTube here. So this actually extracts the transcript from this panel. And it can send it to DocAI. Okay. Hack number two is, um, fixing a personal frustration of mine. So, I'm a Raycast user, and I'm always doing things like… um, converting currencies and, like, complex queries like this. And, uh, it frustrates me to no end that, um, that doesn't work when you plug it into DuckDuckGo. And so what I did is I had Claude reverse engineer the raycast calculator, and I've plopped that into SERP, and so now you can, um, jump into SERP running on this test machine, and do complex queries like that. And there's all sorts of things you can do. Mixed currencies is kind of cool. Um, one thing I use a lot is, like. Adding different units. And there's also, like, sort of… Complex math bits that are kind of fun. And the way that I handled this in the UI is that —. The simple calculator stays if you're doing a simple expression. And same if you're just doing a regular. conversion, then we get the existing UI for that too, because it's nicer since it has the dropdowns. Okay, and my third hack is yet another frustration that I've had for a while. So here we are looking at my phone. Oops. And, um, so we've always been able to switch tabs by swiping on the address bar at the bottom there, and that is quite nice. But what we haven't been able to do is swipe up to go to the tab switcher. And so I added that, and it turns out that Bartosz had already been working on this, so he has helped me a lot with this. Thank you very much. And yeah, it's really nice that you can… it tracks your finger, so you can kind of drag out halfway. Real smooth. And Bartosz added, you can drag out halfway and swipe to the right. Those are my hacks. Thank you very much. All right, and we're not done yet. I have another one by Pablo. When people want their data on another device they don't go looking for things and sync it back up doesn't really tell them what it does. Chanel's usability insight report made that very clear. And the problem is getting bigger. Zac AI chats, uh, can easily be lost. Um, the more stuff we add to Zac AI, um, the more. things that are worth keeping, Sink and Backup will just not be enough. So, we are proposing MyTacticGo, a private account. It's private because it's from us, and it needs no personal data to sign in, so no email and no password to remember. It is encrypted, your data is backed up, unsynced, and we show what's ready to go, like 100 passwords or 244 bookmarks, so the value is very concrete. On desktop, MyTacticGo lives in the toolbar, where people actually look. It lists what will be synced, like your chats and passwords and bookmarks, but it also brings your subscription. So, this can become, actually, your home base, and it shows your plan and your connected devices, so it is a great opportunity for… for dependability, um, and cross-promoting other… other features and understanding our… our ecosystem. All in one place. That's it for me, and I think we're. It's Andrea next.
|
||||
- **21:44:01 · Andrea Soto:** Okay, so I was thinking that we want to increase word of mouth, and stickers are a great way to do it. We also have our Easter egg. Um, logos, that would be great to… sell as stickers, so I added the option next to the logo when you click it to be able to buy the stickers on desktop and mobile. Or $2, so… That's my hack days project. Valent, you're next.
|
||||
- **21:44:35 · Balint Farkas:** Hey, folks. So I'm working on the DuckDuckGo Chromium browser, also known as Duckium. We use Agentic workflows all the time, and it can, of course, see the code, but it cannot see the internals of the browser. So it can see the code, it can drive the browser via Chromium DevTools protocol, but it cannot see our own services like Sync. pixels, burn, feature flags, and so on. So I've exposed them to AI. You can just connect to the browser from your own AI terminal and then ask stuff, for example, why can't I see sync and backup in the menu? It should be here. The answer is because the feature flag in this live session isn't on. Or you can ask spicier stuff, such as performance search for I love cats, which it can do via Chromium DevTools protocol, and observe what pixels have been fired, which comes from the internals observation that I just added. And you can correlate these things and observe the browser's behavior as it happens, which will enable us more sophisticated debugging scenarios, and what's more important, more sophisticated Agentic workflows. That's my first thing. And the second thing is for Mark Sabu, who is out currently, and I'm sharing his video in a second.
|
||||
- **21:45:52 · Andrea Soto:** Thank you.
|
||||
- **21:46:02 · Balint Farkas:** Should be running. Hey folks! I'll walk you through the life of a Windows Release DRI in the next two minutes. After the last quick one days, we iterated on the design together with the release team, and I spent the last two days building it. The Windows Release Command Center. Until now, all of this was manual. We kept Asana and GitHub Actions in sync.
|
||||
- **21:46:10 · Andrea Soto:** Ooh.
|
||||
- **21:46:22 · Balint Farkas:** and dispatch every workflow by hand. Now the ODRI can check the state at a glance, and the release DRI can get through their duties by following the steps in the app. This is the state of every channel, which build is where, and at what rollout percentage. It all comes live from Asana. No more opening 5 tasks and 3 workflow pages to figure out where we are. The calendar shows what's coming up, branch cuts, rollout steps, so nobody gets surprised by a deadline. Clicking the release opens the DRI view. It shows the next section and flags anything that needs attention. Say we want to take 177 to 100% stable. I click this button and it dispatches the workflow. Because it's at 100%, the app asked me to follow these steps and upload it to the Microsoft Store. But I'm actually the DRI of 176, and I will need to do a hotfix. Hotfixing used to be the most manual part. Now I just pick from develop. Like PES, a PR that needs to be hotfixed. and just displays the bot with creation, which will actually just do this. I already done that to not block the demo. Umm. So I will need to upload the release notes. It loads the release notes automatically. So I can just approve it and upload. And now that it's done, I can just go back and simply follow the remaining steps to finish the release. And obviously, every flow syncs back into Asana, so the runbook still works if the app is down. That was Release Command Center. I hope it makes release weeks a little calmer and gives every DRI back some time for actual work. Oh, and sorry about the editing. I'm a dev, so this video was shipped as an MVP. Thanks, Odo! That's it for me.
|
||||
- **21:48:10 · Alex Martemyanov:** Yeah, hey. Hey, everyone. This is me, Alex. There are so many new faces. I'm pretty sure some of you have no idea who I am. Which is fine, sometimes I'm not entirely sure myself. I'm Alex, from the Apple team, working on our macOS browser. And I thought I'd make it greater again. What would you expect from a macOS browser? During this hack… in the hack days. They say Julius was a hacker, performing live demos, chewing water into wine, accepting some cookies. Not bad for a hack days project, right? All right. Showtime. Could you please stop sharing?
|
||||
- **21:48:55 · Mark Ingram:** Thank you, Alan.
|
||||
- **21:48:56 · Alex Martemyanov:** Got it.
|
||||
- **21:48:57 · Balint Farkas:** Yeah, yeah, yeah, I'm trying, uh…
|
||||
- **21:48:59 · Julia Abreu:** It's a very…
|
||||
- **21:49:03 · Alex Martemyanov:** Yeah. Okay, first try failed miserably. Let's try again. Yeah, Kira. Hello. I hope you can hear the sound. Yeah, hope you can hear the sound. But seriously. What's happening to the world? AI is everywhere, too much information, too much liquid glass and beige. Sometimes I just want to burn it all down. Like this. Yeah, much better. Let's do it again. Where shall we start now? Sometimes I just want to run away back to something familiar. Or maybe look for some happiness. Or maybe I just need to talk to somebody. Yeah, maybe the problem isn't the internet. Maybe it's just how we see it. I wish I could change the whole settings of my life. how it appears. What if we could go back? Somewhere like this. And buttons look like buttons. Windows look like windows. Smells like Christmas morning, hot chocolate and mandarins, back when everybody was still around. Remember, on the websites. had some personality. When every page told you someone had paid attention to it. And speaking of the little things… You see? Even the tooltip. Respect your personal space. Your privacy. You just need to start and ask for some help. Here we are. Your protection, your privacy, your peace of mind. Thank you. That's all from me.
|
||||
- **21:50:57 · Deepak:** I'll go next. Let's add the photo. Cool. Clearing my screen… What? I didn't work. Trying again. Okay. Can you all see my screen now?
|
||||
- **21:51:30 · John Boileau:** Yep.
|
||||
- **21:51:31 · Julia Abreu:** We can see it.
|
||||
- **21:51:32 · Deepak:** Okay. Cool. Playing a video, if you can't hear the audio, please. DuckyEye can generate a lot of great images, but only if you know exactly what to promote. ChatGPT, Gemini, and other tools have solved this with pre-existing templates for images, which has also led to a lot of virality for them. Well, now we have them too. If you go to new images, we have 15 of these different templates ranging from anime to 80s portraits to caricatures. Clicking on any of them populates the input field with the corresponding prompt. Some of them require you to upload a reference image, which is indicated by this icon, and then a user is notified to do that with this tooltip. I already did this one. So if you look at this, this was the reference image and I think it did a pretty good job. We also prompt the user to try more styles, so they can click on any of the, uh, styles, and then the original image gets, uh, pre-filled here with the. Uh, a defense bill. Another thing that really irked me was that after all of these images were generated, they were scattered across charts, and there was no one place to see them. Well, now we have this My Images tab. where you can see all of the different images that we have generated. You can delete one or all of them from here, and you can also go to the chart, where. they were generated, or use them as a reference for your next image. One cool thing that we do here is, if you do delete this image, then in the original conversation, where this was generated, we have this image deleted placeholder here, so that the continuity of the chat is not lost. That's it. Thank you. Yeah, that's it. Thank you.
|
||||
- **21:53:20 · DR:** All right, so I go next. Let me see. Here. Can you see the screen? All right. So, uh, starting next Monday, our internal release, actually internal users, uh, in the internal release, will have, uh, the possibility of installing from the web, uh, Chrome Web Store 1Password. And this video showcases how it works, that it connects to the desktop app that you have installed. You can see that it works for registering your email and autofilling your password for it. You have to unlock it first, though. You can see here how it works with Touch ID. Um… You can fill the login, password. Then it's going to ask you to save it. You can also, of course, use it to log in to a site, those details. So you guys see here. Then it also works for passkeys, where you can register a passkey for one of the accounts that you already have in 1Password. Um… Again, it's going to probably ask for authentication. And then use that same passkey. To log into the site. In this example, it didn't ask for additional authentication because it was so close to the previous one, but it's working quite nicely. And we'll be posting an announcement with details on how you can provide feedback for this. I said it's quite complicated on the technical side. Thank you. Hold on.
|
||||
- **21:54:58 · AO:** Hey, Elle. This is a demo that Alex, Elle, and I did. So double Alex demo. or DACAI artifacts. Hey everyone! During Hack Days, we explored DACAI artifacts. Documents, presentations, and visuals. But you know what? It's boring, so here's the twist. This presentation was built entirely in DACAI. See? It's right here, inside the chat. And you know what? I can even download it as a PowerPoint file. Now check this out. Let's ask how much I could save. I get an answer, but only for this one case. Different numbers, we need to try again. Or I can get an app that does that for me. Duck AI builds it right here, next to the chat. You can watch it come together. Not a code block, not a screenshot, a real working app. Now I can change the amount, the time, the rate, and it updates instantly. Have a PDF? You can turn it into just about anything. Study notes, a brief, key takeaways, whatever you need, as a document. You have full control over your doc and markdown. You can also let Duck AI rework it. When you are done editing, you can now export your doc. With Artifacts, you can build your own website too. If you want something different, you can just point DuckAI at it and ask for an edit. Every version is saved, so if you don't like your change, you can go back in one click. Artifacts also work well with our browser features. Let's say I have a really long article, and absolutely no desire to read all of it. Attach the top, and Dakia will summarize the main takeaways visually. Just a few seconds, and it's ready, clear, structured, and much easier to understand. And of course, artifacts work great on mobile too. Try them yourself! That's it. Thank you.
|
||||
- **21:57:18 · Gary K:** All right, I'm up. Let me share with sound. a video… Here we go. Any time there is an incident, the first question is always, what changed? And the answer is potentially in a lot of different places. So I've created a command line tool that will aggregate the events that are likely to be interesting for anyone investigating new behavior. What changed is a tool that can access Azure activity logs, Asana deployment plans, Jenkins jobs, GitHub pull requests and actions, REX commands, and Mattermost notifications. Because there can be a lot of data in these sources, the tool ships with default settings that are designed to give high signal-to-noise ratio. These settings can be overwritten via a config file. or a runtime flag. Users can search backwards from now or over a specific window of time, from minutes to days. There are a host of flags designed to make it easy to use. Users can opt in to format the output as JSON and perform further filtering. and transformations using JQ. What changed uses your own credentials, so security posture is maintained. What Changed is available now at www.duckduckgo.com/gcurtis/whatchanged. Thank you. Like and subscribe for more great content. Thank you.
|
||||
- **21:58:52 · Tom Strba:** Nice. Hey everyone. So over the last couple of weeks in OJ and OM, we have been trying to build tools for Duck AI. What does it mean? We would like Duck AI to be able to get context automatically from your browser. And not only that, we would like to. Duck AI to be able to control the browser. So, during hack days, I tried to put all the pieces together, and uh… I would like to show you a proof of concept that I managed to build, so… Let's say I have, uh… Couple of, uh… shoes I am trying to buy. And I would like Duck AI to recommend me which shoes should I buy. So… I will ask you. Which shoes should I buy from all these options? And I'm not providing more context. It will basically get the list of my tabs. It will read the content and it will recommend me, oh, this is the best for speed for daily trainer. whatever but this is not where the fun ends this is where the fun starts so how about now we ask Duck AI. to buy our choice. So for example, buy me stride arrow three. And Duck AI will basically open a new tab and start buying shoes for me. As you can see, here is the store and it's starting to fill the form and it doesn't know which. shoe size it should buy, so it asks me. I will just respond, for example. 10, and it now reads the form. As you can see, it filled. The 10, and now there is a… the website needs the address. So the fun part is that we already have the identity in the browser. So Dakia asks me, hey, can I use this identity from the browser to fill it? And yes, of course, let's share it. If we take a look at the store… by the way, this is the store I built for myself for testing. The address is now filled. and it asks me, it asks me last time, hey, how about, uh, I'm gonna, I'm gonna submit the form, are you, are you sure? Yes. I am sure. And my shoes are now ordered. I didn't need to click on that website. I just wanted you to see what is happening in the background. But Duck AI now or soon will be able to buy. stuff for you or do anything in the browser you want it to do. Uh, that's the first demo I wanted to show you. And the second one, this one will be really quick, and it's, like, internal tool enhancement. Uh, if you… Often test a Mac. make ship reviews, you know the process. You need to download the DMG, attach it manually. Then drag and drop it. The instance is running, so you need to quit it. Then drag and drop again, wait for installation and then run it. So I build the button and now you will just click to run and you can start reading the. ship review, and everything will happen in the background, and after the build is downloaded, I'm gonna speed this up a little bit, it will basically just runs, and you can start testing. Um, that's it for me, thank you.
|
||||
- **22:02:22 · I(:** All right. I'm just getting my screen up. Please let me know if you can see it. All right, looks like it's there. All right. So as a user, why should I care about the protections report on our browser? It's a list of numbers and icons. What is the actual benefit? So I reworked it. I started with the bottom line. Companies are watching you. It's creepy. Also highlighted right away that there's an issue to be fixed. Privacy isn't your default, and here's why that matters. The more of your everyday browsing happens in our app, the less these companies get to see. So let's expand this. Immediately, we get a creepy, interactive, prying eye that follows your cursor around. And we get useful info. We block tracking attempts, but why? They target you with ads, they watch what you read, and they record what you click. It's creepy. So who's watching? Well, Google does a lot of spying, and it was on 4 out of the 5 sites he visited. And Facebook was on almost half. Again. It's really creepy. So I wondered how we could make this more like a personal infographic. Here we highlight the nosiest site of the week, and we show a ring of trackers from that site that blink like eyes to drive home the visceral idea that you're being watched. Now, we empower you to understand some of what we're doing by showing you how we prevent a digital fingerprint view from being created. You can hover over each component of the fingerprint and see what we've hid from creepy companies. And how many times? And finally, we contrast this with Chrome. Where companies swarm around your cursor, and cookie pops and ads fill the screen. We drive home the fact that Chrome is owned by Google, and that unlike on Chrome, DuckDuckGo blocks Google's tracking attempts. So maybe this really creeped you out. Why are so many companies watching you read the news? Let's share that. And just like that, a promotional infographic can be shared with a link to download our browser. So try it out for yourself on Windows on the link task. Thanks.
|
||||
- **22:04:41 · Vinay Nadig:** All right, that's me next. Let me share my screen. Alright, so for hack days, I have. built an app called, QuackTime. What it does is, it lets you schedule ad hoc meetings, through Mattermost without having to do the whole song and dance of matching time zones across. uh, different geographies. So, to use that, go to QuackTime, go to Metamost, or any DM, QuackTime and DM duration, in this case 60 minutes, and then we'll set a type build. You will get a link, you can open it. you will end up on this interface, basically. You will see that all your, uh… uh, working hours and free, uh, slots through Outlook has already been filled in. And that would be the same case for another person as well. So, the other person will also open the same link. And yeah, so you can see that I'm working with Adrian, so I have this much overlap with him, so you can select whichever task that you want. And, oh, sorry. You can propose it. And the other person will… See an option to accept that. And once you accept it, an invite will be sent automatically at the correct time, along with the Zoom link for… which uses your personal Zoom link. That's pretty much it. I'll stop there.
|
||||
- **22:06:17 · Nicklas Jensen:** All right, I'm next up here, and I am going to be sharing some porn. out of a conversation, actually, in a Zoom chat. So, one of the things on the meetup that happened was that we, uh… At least for a lot of the ducklings, we were freaking out about how much merch and how many things have been over the years. So the idea was to create a hall of fame for specifically T-shirts. So Miguel and I worked on that. For this, uh… For these hack days, the goal is you have a timeline, you can see all of the different designs that have been over the years. These are all the ones that Miguel could get a hold of so far. You can vote for your favorite ones. And also, if you don't feel like going through them all manually, you can just… pick out of sort of a random generated thing. And yeah, that's the T-shirt Hall of Fame. How do I stop sharing? Thank you.
|
||||
- **22:07:30 · Dominik Kapusta:** All right, here's a demo of the… Long overdue feature on… macOS, which is the reader mode. Um… Reader mode. extracts text from websites. And here you can see that, uh, we're navigating to a website that, uh, is eligible for being displayed in reader mode, which contains relatively more text, uh, and, um… Just with a single button in the address bar, um… we turn on the reader mode, then we navigate away from that website, it just resets to a regular view. here, uh, you can see, uh, visiting another website, uh, that's also eligible for reader mode, and here's, uh, here's how it works, but it's not all, because I've worked on that, uh, during Hug Days, but. Um… a parallel to my work, actually. Marios was doing the same thing for iOS, so here it is, um… Here, the reader mode, uh… Button is in the website, uh… menu… And this demo will also show you that it respects the browser theme. So that, um… shows correctly in the dark theme as well. Even if you visit a website that… uh, doesn't honor the system theme, uh, then the reader mode, of course, uh, will be dark in this case, when the Guardian was in, uh, in light theme. Yeah, and this all was possible. to do thanks to a library by Meek that extracts. text from, from websites. So really the integration. It was fairly minimal here, and it was just a bit of work. That's why I said it was long overdue, because that library was published a few months ago already. That's it. Thank you.
|
||||
- **22:10:01 · Valerie Kraucunas:** Alrighty, hello. Um, so, I'm Valerie. I work on the front-end team. Um, the thing I've been really heavily focused on all year is the rebrand, and that… with that has come… turning over, uh, style values from one… one way to a new way. Um… with that, we have these things called design tokens, which is basically, like, a codified design decision. And so, when we're styling… the new rebranded stuff we're using, or at least we want to be using all of these things. That said, um, a lot of times when you are handed designs that are rebranded. You end up with this scenario of… It'll be a mix of using the new style tokens and the old style tokens. So in this case, like the SDS styles are a legacy thing. Um, and I want to be very clear, this is not a shame. This is not a shame campaign or anything on designers, because oftentimes when they are building these things, they are… they have the old way in the same file as the. new way. And so it's really, really tedious to get down into every single nook and cranny and node or layer to make sure that you're using the new styles on everything and to make it even more complicated. So when developers see this, we just see this and we're like, why is this still here? Why are we still using SDS things? But when you get over into the design look, so this first column is when they're picking the first padding value, so space X07. When they're picking the second column value, space X03. They look the same. It's really easy to co-mingle the old style and the new. And so, um, basically my concept is to… be able to hand a Claude skill a Figma link, and have it go and traverse the entire. All child nodes of that, um… of whatever you linked over. So. I'm going to try this. I doubt it will… I doubt it'll work. It does not go as fast as I would like it to, but we'll give it a go. So, basically, what happens is you, you know, invoke it with the Figma link, then you give it an Asana link. And give it a name. Um… What that ends up spitting out is a nice report for designers to be able to go and update all the things that are using, uh, old values, or possibly things that are just out of system. As well, um, now I want to be clear, I'm standing on the shoulders of giants here, because, um, while we've been doing the SERP rebrand, uh, my partner in crime, Rachel Lim, created a… skill called Marlin Rebrando, which is basically a dictionary of, uh, knowledge for this was the old way, this is the new way, these are the things we should be checking for when we are executing the rebrand and the code. Um, so my skill is heavily relying on that, but, um… What it'll basically do is it will just… fully traverse like a node tree, if you will, all the way down. Um… And yeah, the things that we're looking for are the use of legacy variables. This is run from within the DDG. repo, for now, at least. So what it will do also is it'll look at the currently installed design token package and make sure that there aren't values being used in the designs that don't exist in. the installed package, then also looking for just colors that aren't tied to a system value, unselled text, et cetera. So what this ends up doing is it will create a. design token report, and. give the designers a link to the layer that. his, uh… has a property on it that's a problem, if you will, and then give it an idea of, like, this is an unexpected value, and then has an expected or a suggested value. The nice part about this as well is that developers. don't necessarily have to wait on designers to go. What?
|
||||
- **22:14:28 · Julia Abreu:** Well… I'm so sorry, I hate to do it, but I have to interrupt you, because we're at 4 minutes now.
|
||||
- **22:14:35 · Valerie Kraucunas:** Oh, shoot, I'm so sorry. Okay, well…
|
||||
- **22:14:36 · Julia Abreu:** Sorry.
|
||||
- **22:14:38 · Valerie Kraucunas:** Making everybody's life better, that's all.
|
||||
- **22:14:50 · Stano:** Okay, I believe I was next. Um, I'm gonna… I hope you can see my screen now. Someone please confirm.
|
||||
- **22:15:00 · Julia Abreu:** No, we cannot.
|
||||
- **22:15:01 · Jason Ashby:** Nope.
|
||||
- **22:15:02 · Stano:** You cannot? Let me try again. Where is… The sun got stuck.
|
||||
- **22:15:24 · Julia Abreu:** Did you find the share button?
|
||||
- **22:15:27 · Stano:** Give me a sec, I reconnected. You can take it out of my budget, I'll make it quick. It does not want to work. You can skip me. Go on.
|
||||
- **22:15:45 · Anh Do:** Okay, looks like I'm next. So let me try to share my screen. Wow. Uh, how do I share, like, a… Particular windows, um… Sorry, um, I had so many windows opening at the moment. Oh, come on. Guys, it is not fun. Sorry. OK, I'm going to share my desktop then. One sec. Hopefully, you can see that on my screen. Can anyone see it? Okay. So for context, in OK, we are helping people in the US to remove personal information from data broker by submitting our request on their.
|
||||
- **22:16:21 · Julia Abreu:** I can see it.
|
||||
- **22:16:33 · Anh Do:** behalf. But recently, we have a lot of issues with broker block our archive with CloudFlare and CAPTCHA, so we can submit the form for the users. So this demo has been built based on Ben Smith design prototype. Let me play that.
|
||||
Reference in new issue
Block a user