Product and Engineering @sourcegraph.
Bremen, Germany
- Tweets 139
- Following 100
- Followers 228
- Likes 456
Made 17 PRs restructuring a bigger service to support some upcoming features.
Well my agent did.
But I reviewed the architecture, I had the agent battle test it E2E, at scale, verified backward and forward compatibility. The latest E2E run was a beautiful 4.5 hour agent loop.
2
5
370
Agents running remotely is good for health!
Going on more walks recently to brainstorm and think and hack on stuff. Always wanted this Jarvis experience where I’m not tied to an ergonomic desk and still can build stuff. But I needed a keyboard. Dictation ftw.
1
5
119
Forever favorite testing technique: Snapshots.
You get to see down to the byte what has changed, and you can very easily verify work, both from humans and agents. And the agent can look at it and say "that ain't quite right yet".
Example: github.com/sourcegraph/scip-…
6
505
Everyone should commit generated code.
You can browse the code on GitHub, your agents can search it, and don't spend 200k tokens on "how to run generate" before anything builds.
Agents write code at the speed of light now, surely not the main driver of repo bloat anymore?
5
212
Kubernetes is great for orchestration .. agents need good sandboxes though. gvisor is cool but not a full full dedicated "computer". EKS has no native sandboxing runtime classes right now ..
Surprisingly @Azure has with Kata!
When will the other hyperscalers follow?
1
3
746
Our frontend teams probably asked about 50 times about some vercel-like preview environment for PRs on @Sourcegraph .. expensive and slow to boot up all the 153 GCP resources..
@AmpCode orbs will now get us that for close to free: "dear agent, add the portal link to the PR thx"
1
2
16
1,413
Are university classes and SWE books more useful or less useful to be a software engineer now than it was 6 years ago?
Back then typing correct syntax was a huge part of the day to day. Now it’s mostly higher level thinking, which they teach. So maybe they’re more useful now?
3
5
333
When shipping code is getting easier than ever.. it's also harder than ever to build the right stuff. Join my team @Sourcegraph as *the* Product Manager!
grnh.se/7hca2iem4us
1
15
4,252
If you are opinionated and pragmatic about building products on top of large language models, I'd love to work with you!
grnh.se/59eagky04us
8
6
126
9,615
Better than 500 coding agents is one orchestration agent, say leading physicists who say things about orchestration agents.
Jokes aside, when we prototyped this, it clicked immediately. We're using this daily now, despite not being a 10,000 people org.
sourcegraph.com/changelog/ag…
7
456
Back from summer vacation and all I've been hearing about while I was mostly offline driving through Europe was GLM-5.2.
Great to know what to do on day 1 :D
108
It drove me so insane recently that we prioritized this work: sourcegraph.com/changelog/ne…. 60fps scrolling, even at 100,000 changed files. We’re experimenting with lots of UX ideas there right now. Send me your wishlist
This is why PR diff speed matters. This isn't a dunk on GitHub specifically, because GitLab, Forgejo, etc. are all equal or worse. But this is the kind of thing that drives me nuts, because this is a core workflow and its slow enough I literally take my hands off the keyboard.
Btw, when my mouse jiggles on the left, its because the page is literally skipping frames and I'm instinctively shaking my mouse to see if it'll respond. And on the keyboard input you can literally here me finish typing before a letter even shows up.
For someone like me who is an expert at these tools, my brain navigates the tool dramatically faster than it can keep up, and that is not good. The tool should not get in the way.
4
217
Applied AI that works:
@linear please file a ticket for this. Add it to the XX team, set urgency to high. Add XX as a customer request. Link the ticket to the linked product gap submission. Assign <coworker>.
This used to take me like 20 clicks or keyboard shortcuts.
1
10
9,434
We've been experimenting with something called "compute search" 5 years ago. But who wants to write a script to get their answer out of a one-off code search, even if the data was valuable.
LLM := Take away the script writing, and still get the valuable result!
Huge update today: we gave our Deep Search agent the ability to write and execute scripts to help analyze and deeply understand a codebase.
SOTA models love, love, love to write code. Problems that the agent simply can't do in it's own loop can now be offloaded to a script, with the results shared back.
Custom tools on demand. This is code mode in action.
sourcegraph.com/changelog/de…
1
1
8
407
The best 7.0 release we've ever done!
Today, we're announcing Sourcegraph 7.0, a release that marks the beginning of a new chapter for our company and product.
Over the past several releases, we've shifted our focus. We're doubling down on being the intelligence layer that developers and AI agents rely on to navigate, understand, and operate on large codebases.
sourcegraph.com/blog/a-new-e…
8
644
We had a lot of fun in the past couple of weeks at @Sourcegraph ...
1
1
11
586
In the era of AI, tooling changes a lot, more than anything else in engineering.
Building things yourself is suddenly cheaper than off-the-shelf tools.
The #1 mandate in engineering at @Sourcegraph is to make things agent friendly.
1
8
322
Once on a WYSIWYG platform, our new engineer-written Changelog is now a beautiful and simple app.
Our engineers own the storytelling of what they ship and write in markdown.
Very agent friendly.
Btw, lots of things shipping this week, check it out!
sourcegraph.com/changelog
3
19
4,367