How VS Code Went from Monthly to Weekly Releases with AI — Harald Kirschner

AI Engineer · 19 min · 184 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:12Hey uh everyone.
00:15This is a hard tone to follow.
00:16Why did everybody leave?
00:17No. Um this is the rerun of a talk that she gave yesterday upstairs.
00:22So this will be better than the one yesterday.
00:24So uh kudos to you for catching this one.
00:27I'm working on VS Code and we move from a monthly release to a weekly release cycle thanks to AI.
00:35And this is about the things we learned, the systems that made this work, not just to ship faster, but actually to ship more effectively.
00:43So, it's not about how we build agents into VS Code, which we can try out anytime.
00:48You can come by the booth and see us showing of agents, but this is about how we use agents to build the product itself.
00:56And that's VS Code, that's the copci and other areas, but also how as you use agents and more AI, things are breaking and you have to rethink actually how you work.
01:07And that's a very small team shipping to over 50 million users.
01:12What happened is over the past year is we started tracking already early on this code survival metrics in VS code.
01:20So that means the percent of agent written code that actually gets committed.
01:24how much kind of garbage has the agent produced that you threw out the window again or didn't trust and that you that you read it u and and scrapped.
01:33So GBD 4.1 started at 55%.
01:37But then over time and over improving the harness and new models cloud opus 4.6 is now today at 86%.
01:46And that increase clearly shows developer trust and confidence in shipping AI code faster and more AI code generated.
01:56And this success created problems on the VS code side.
01:59We're one of the biggest or the biggest open-source project on GitHub.
02:04So we saw a massive increase in issues because AI helps to file issues.
02:09So we actually see an increase in quality issues but also the counterpart of more automated lowquality issues.
02:17We also see more open PRs by the team by our own velocity and by our community.
02:24And interesting part we actually see everybody would assume we see a lot of garbage PRs but actually the number of community merged PRs is going up as well which is really amazing that we can unlock more community contributions as well.
02:37And as part of this new workflow, we actually moved several months ago, we moved from monthly releases, which happened over 10 years.
02:44Since the inception of VS Code 1.0, we have been shipping a new version of VS Code on a monthly basis.
02:52Including fun facts like the January release actually ships in February.
02:56So when you read the release notes in VS Code, you're wondering why it's February and you're reading the January release notes if you're a month late.
03:03No, that's just how we tracked our iterations.
03:08So that cadence was to match to velocity, but it also showed us that the it's not about using more AI as you work dayto-day, which a lot of companies are pushing.
03:20Like you're not token maxing.
03:22You're not spending every day using AI.
03:25It's about evolving the whole system, how you're shipping software to evolve to actually make better use of AI along the whole process.
03:34And that what makes this velocity actually work and doesn't just generate more code.
03:40I'm breaking this down into this first part which a lot of people have been experiencing.
03:44I see shipping code faster and that's the easy part.
03:47You have like this 100x engineers that has built this beautiful skill MCP plug-in system and they're just shipping code like crazy and slowly maybe more in the more people in the team
03:58uh adopt the same patterns and you see a lot more code shipping faster.
04:03But the problem is once you ship faster, you get into code review.
04:07You get into how do you actually ship high quality code that doesn't constantly break.
04:12And that's important for us as VS Code.
04:14We're shipping binaries to user machines.
04:17So if something breaks, that recovery is even more costly.
04:22But once you hit that velocity with shipping high quality code faster with confidence, how do you actually know you're shipping the right thing or if we should ship this thing?
04:34And that's where AI kicks in to actually learn faster in the end and shipping better products which a lot of people are missing out as as they write more code is applying that product taste and applying that learning.
04:49So let's start shipping faster.
04:51Now that everybody can vibe, we can really all ship faster.
04:54And it's a big part is getting your codebasation ready.
04:58So that's kind of the no-brainer slide that everybody should already been doing is thinking about how to get your code bases agent ready is creating agents.mmd that give it that are lightweight enough and give the agents a good kind of map of the codebase where to look at
05:15mapping out that is every agent now at your uh at your fingertips has a slash in it that gives you a good draft and then you keep itering those are living documents
05:24that should evolve as the agent makes mistakes and works.
05:29And that's where I see a lot of teams that have invested in developer experience actually gain a lot because once they have good documentation and onboarding for a codebase
05:39that is read by the agent as well and that can help the agent as much.
05:44Next up you have skills.
05:45VS Code has always messed heavily in accessibility.
05:48So one skill we added early on was infusing all the accessibility best practices and how we think about accessibility into a skill that everybody will use and that previously was one person
06:00that had to then got get pulled in or give feedback and now that skill was reviewed and is maintained by the kind of area owner of accessibility.
06:10So to bring that that accessibility that short work everywhere into everybody's hands.
06:15So think how you can encode your expert knowledge to unblock this like few users that might be always asked for this input into a skill.
06:25And lastly once you do it right for us in VS Code the litmus test is can I as a PM effectively VIP code in the VS code repo which I do
06:38um and how is that accepted by engineering?
06:42How much work does it cost?
06:43Is it out of the box working?
06:46Is it for me working locally?
06:48Or is it eventually breaking when I ship it?
06:51So, a lot of these quality improvements will show in the end that you can allow everybody to be more effective in the repo.
07:01Another improvement, and that's a massive one, is we switched to TypeScript Go.
07:08And if you've been in any developer codebase that takes some time to build or run linting or run CI/CD, you've noticed it's already annoying.
07:17But developers find other ways.
07:19They can review a PR.
07:20They can context switch into something else while runs and then do something else.
07:24But once you hit that with agents, you have 10 agents, 20 agents running and hitting the same bottleneck, these any slow part of your CI/CD will compound.
07:36And for us, Typescript Go was a 10 times improvements in our builds, which at the scale of automatic PRs running with agents and them fixing getting needing that CI/CD
07:49loop for feedback to improve the code is a massive improvement.
07:55And that leads to me uh filing big PRs.
07:59That was beginning of the year to get the ask questions tool in which I believe we really needed and then bringing it to the team to show to let them try it out in the product.
08:09Within a few weeks we had a perfectly polished and designed experience based on my initially very scrappy PR.
08:17But it unblocked that this these discussions the participation of everybody else because we had the ground set laid and the idea and the experience was clear.
08:30Next up, when agents actually work on UI, and we all had this, that the agent gives you a UI and it says it looks perfect, then you open it and it's all misaligned.
08:40Even with SVG, even with the latest models, you still get it.
08:44And it's like the number one thing I keep telling everybody now these days.
08:48If your agent cannot use your application, your product directly to get this feedback loop of is it all working, then that's a really big investment that pays off every time you work on UI.
09:02And there's two areas we invested in.
09:05One is the component browser, which is an automated build that runs every time VS Code changes something.
09:13Let me open this in this issue as an example.
09:19In this case, we added a back button to our customization screen.
09:23And because it's changing wide, we actually have this automated process in the back that takes screenshots of every component in VS Code and points out the differences.
09:33It allows us to detect unexpected changes like we're changing one component as the ripple effect of something else moving or some other icon disappearing.
09:42and allows you to quickly review a PR as well because you also get directly a view and what the PR is actually doing and that's what we ask every developer before we ask them to at least attach a video or screenshot on what they have been implementing so we don't have to run everything from scratch and we
09:59can quickly assess and and give feedback and this is now automated with the component explorer next up we have the self-correcting loop because in VS Code we run uh a web browser and open a website.
10:14Basically, that's what VS Code is.
10:16It's all HTML.
10:17We can actually use Playright to automate a lot in the browser.
10:22So, we have a /launch skill that would just open up VS Code and give it a specific scenario to click through in VS Code.
10:29So, it's a really powerful way to diagnose issues because you can also get the locks along the way, but also validate any fixes you did before and after without you having to click through it.
10:40So you can just hit launch move on to the next task and then come back once the agent verified it's worked because it can access your application and to call out this is nice because we have playright and we are a web app uh running in electron
10:55but there's many other solutions out there one of my favorites is Xcode MCP because on my site I work on iOS apps and once you see Xcode MCP just creating in VS code without Xcode involvement
11:07just making an iOS app opening it up clicking through it, taking screenshots.
11:12These moments are magical and give you a really tight feedback loop as well.
11:15So there's other tools available for Android as well to make this work.
11:21Let's talk about code review, which I think is the next big investment.
11:24We know in the past when we enabled code we on GitHub the automated GitHub copiloted one we're not convinced yet but over the next months and weeks they have massively improved and now it's a mandatory
11:37uh review every time somebody opens a PR and there's more improvements coming actually dial in the uh the effort that a code review does depending on how much risk you have in this repo so you have a low medium high effort for code review so you can dial in kind of the cost benefit
11:52as well but for us is if a review after review is done, humans will not even look at PR until all the comments are resolved and and addressed.
12:03Okay, now we're holding we're shipping faster.
12:06We got to hold quality.
12:08And a big signal for us because we're on GitHub is of course issues.
12:12Uh who has filed a GitHub issue?
12:15Who has had problems in VS Code before?
12:19Yeah, cool.
12:20Kind of similar.
12:21Awesome.
12:21This the last crowd had less hands up.
12:23So it's good.
12:24Um, yeah.
12:25So it's a massive, it's a really good signal.
12:26And as I said, we get sometimes even better issues now because of AI and more people with different language backgrounds being able to file very good issues.
12:36But we also get a lot of issues.
12:37And previously engineers actually triage them manually.
12:41And now we have AI doing the filtering to get out spam enriching and translating in the end assigning area owners for all the issues to raise the signal to noise quality of this.
12:54We also add a human layer to that once you are in issue because the agent will make mistakes and we need that feedback loop when it makes mistakes.
13:02We create a Chrome extension that allows you to fix duplicated issues as well.
13:07So we have the agent doing the work and we have a feedback loop where humans can still fix it that we can bring back into the into the agent work up front.
13:16That's a really important thing to think about as you have agents doing the work that you get the human feedback as well.
13:23The other input we have is a lot of error stacks.
13:27Every time there's an exception and most apps will have this you get an error stack reported and we actually don't use any external tool for that.
13:34It's all in our own data and we built all the tooling around it and it's mostly we already had it.
13:40We already had an ML doing sorting on top and this is now actually done by a mix of criteria.
13:48So one is we collect the raw telemetry.
13:50We filter it down and that's like 51 billion per day.
13:53We filter it down to just error stacks that actually have the full stack.
13:57Then we group it and bucket it.
13:59That's more of the fingerprinting exercise.
14:02And then in the end after we all of that we have 10 issues filed assigned to specific area owners that a PR is autocreated to actually already try to fix the problem.
14:15This is our error page where we see all the stacks including how many hits and how many users affected so we can quickly react to anything that's coming.
14:23And once we have this working we get a issue filed and a PR automatically opened.
14:30This is the issue.
14:31is the error with some initial diagnosis and then the PR got opened with and the agent actually goes in and finds out why does this happen who was responsible in this case um
14:45added a change that was connected to this to find f find symbols and then it figured that the on cancellation request wasn't in the RPC protocol so the agent already opened the PR we could just merge it and that's a really great automated way you don't care about errors you just want to get a more stable code
15:02base.
15:05So all of this automated the triage initially mix of agents and deterministic systems then the issue is filed handed off to an agent in the PR multi- aent system and in the end we have fixing fixes landing mostly automatically
15:18with some human oversight to still approve the PR.
15:25There you go.
15:26So previously VS Code did yolo releases in a way that it was release day everything was stable.
15:32We did our testing and we shipped the release to 100%.
15:35We just opened the floodgates but because of agents we want to derisk that process.
15:40We should have probably done it earlier.
15:42So now actually we do stage rollouts and as we do the roll out we monitor error locks and issues and everything else.
15:48So just like good citizens in a in an web application, we now apply the same because rollbacks on an install app are so expensive.
15:58Okay, now we have this shipping faster, shipping at quality.
16:02How do we actually do this in a in a speed that we actually learn faster, which is the which is a bigger challenge with AI?
16:08You actually want to build a better product faster.
16:12So one of the examples is our VSC bench.
16:14We're shipping an agentic product.
16:16So it's important that we have our own evals of our product always at hand and being really easy to extend.
16:23It's all in the GitHub issue.
16:25It's all in GitHub.
16:26When we file an issue to add a new scenario, it's actually an agent picking up from a template and bootstrapping the scenario for you.
16:33So we make it as easy as possible to bring in all the developer scenarios that are coming up in issues through customer conversations through all these channels to bring them back into our emails.
16:43So we then start hill climbing actually against them and then com also confirm that it's all improving offline and then an online experimentation.
16:52So classic product life cycle for um for us and we see these interesting things in our stack.
17:01Understanding your evals is critical.
17:03One experiment we did that you can read about on our blog is that we gave a very simple eval scenario of just write a file with hello world just testing out our eval harness end to end and how different models react.
17:18The same five character file took the most expensive model 70x more tokens and that was not the high reasoning model.
17:26You can read more about unblock.
17:27We actually don't talk about the models, but just that insight in how your models react with your harness is extremely is a very important exercise.
17:38Next up, we have prototyping, which I do a lot of because I noticed that I don't actually want to land most of my work into VS Code PRs itself.
17:47It's a great way to start a conversation, but most often I want to have a quick conversation because now we're meeting on a daily basis where we just want to have very quick feedback loops
17:57on here's the idea, here's how it could look like, what if we do this.
18:01Next day you come back with updated uh prototypes and you just keep having this conversation.
18:06So much quicker feedback from a month release cycle to weekly to daily sprints where you just keep working on problems and prototypes unlock those deeper discussions and how the experience should look like.
18:21It also means that we have the durable ownership in internal work as I mentioned daily sprints um smaller squads, smaller work streams that really prioritize and drive smaller sets
18:34um of areas to move faster.
18:37So try these things, find out where your bottlenecks are to actually ship higher quality, ship faster and learn faster and find fix that next bottleneck.
18:48And don't just tune how you work with agents, but tune where how how do they get feedback?
18:52How do you build these loops?
18:55And then figure out after you have these these success moments, how can you improve that feedback loop for the next moment of what's blocking you to move even faster
19:04and it's a constant iterative cycle across all these cycle ones.
19:09That's it.
19:09Come by code booth.
19:10Happy to chat now.
19:12Otherwise, happy coding.
19:16[applause]
19:33[music]