Sign in

Sebastian Huckleberry

@sourishkrout.com
334 followers 284 following 178 posts

cofounder at stateful.com - makers of runme.dev

PostsRepliesMedia
Sebastian Huckleberry @sourishkrout.com · 14/08/2026
Ha. #LLMs are bad jokers and a tough audience at the same time. This one scored 78% 🤷‍♂️.
000
Sebastian Huckleberry @sourishkrout.com · 08/07/2026
If my “World Cup picks report” #AI agent skill had a theme song, it’d be this one. Eval results: world-cup-picks-report-evals.sourishkrout.workers.dev Skill and evals: github.com/sourishkrout...
Meme image of Busta Rhymes pointing at the viewer with the caption: “Don’t cha wish your skills were eval’d like me? Don’t cha?”
000
Sebastian Huckleberry @sourishkrout.com · 07/07/2026
“F*** your skills” is the joke. “And mine too” is the point. We stopped shipping software without tests. But we’re still putting production-grade behavior into agents on lucky runs, screenshots, and folklore. Use `runme eval` with tests and results in the repo: runme.dev/blog/runme-e...
Terminal showing a successful runme eval regression run and comparison, with all checks passing and the reward improving from 0.832 to 0.978.
000
Sebastian Huckleberry @sourishkrout.com · 15/06/2026
I still hate you, #FIFA, but I did love everything about this ⚽️ soccer game yesterday. #CURGER
000
Sebastian Huckleberry @sourishkrout.com · 21/05/2026
The cost of product launches has collapsed drastically. Yesterday, visr.dev had: - no blog, - no content, - no design, - no video clips, And, we had yet to polish the actual product feature. It dropped today and it ain't AI-slop: visr.dev/changelog Used to take ~2wks for a small team.
020
Sebastian Huckleberry @sourishkrout.com · 20/05/2026
Like specifying a user's fax number when you run "useradd". Sure thing.
000
Sebastian Huckleberry @sourishkrout.com · 13/05/2026
Trude Kernig, back in business. My OpenClaw forced my hand. I’m fully migrated to Codex. Claude didn’t lose on taste. Anthropic degraded out of first-choice status. At parallel coding speed, throughput wins.
Trude's recent coding agents threads
000
Sebastian Huckleberry @sourishkrout.com · 12/05/2026
Gearing up for the early Visr.dev release. Simple goal: help AI agents stop making the same mistake twice. We’re turning past runs into reusable skills, checks, and evals for agent harnesses. If your team's building with agents, this is probably a problem you’ve already felt. Waitlist is open.
Skill derived from a combined terminal and agent session.
010
Sebastian Huckleberry @sourishkrout.com · 23/04/2026
It's wild to see the weirdest rate-limits hit you once you dial up agentic coding concurrency.
000
Sebastian Huckleberry @sourishkrout.com · 21/04/2026
Memory getting in the way?
000
Sebastian Huckleberry @sourishkrout.com · 21/04/2026
You forked Codex? I've been doing evals in Visr. Helped a lot while draining my bank account. So far I've gotten away with Mini models.
100
Sebastian Huckleberry @sourishkrout.com · 20/02/2026
One from horse-less carriage category. Against @jeremy.lewi.us's efforts Golang must no be particular popular at OpenAI 🥸.
100
Sebastian Huckleberry @sourishkrout.com · 13/02/2026
Jumped on the #openclaw hype train Nailed choosing its name
030
Sebastian Huckleberry @sourishkrout.com · 01/02/2026
Here are a few pieces of evidence:
000
Sebastian Huckleberry @sourishkrout.com · 18/12/2025
It's always fun to see a big traffic spike hit @runme.dev's website. Today @console.dev featured Runme in their newsletter. Thanks for the shout out and also appreciate the likes / dont likes.
000
Sebastian Huckleberry @sourishkrout.com · 21/10/2025
000
Sebastian Huckleberry @sourishkrout.com · 20/09/2025
Mezgroni 🥃? There’s got to be a better name for a Mezcal Negroni. One week sample from the barrel 🪵 is excellent 🤯.
010
Sebastian Huckleberry @sourishkrout.com · 08/09/2025
You and I are gonna live forever 🎶 #oasis
000
Sebastian Huckleberry @sourishkrout.com · 03/09/2025
Sometimes Javascript is just killing me.
110
Sebastian Huckleberry @sourishkrout.com · 08/08/2025
I have to give I to the backpacks made by peakdesign.com. They don’t just look sleek. They continue to save the day. Look how effortlessly I strapped it to the motobike on my impromptu Provonce 🇫🇷 road trip. Viola!
010
Sebastian Huckleberry @sourishkrout.com · 08/08/2025
Marseille is an excellent place to visit. It’s vibrant, eclectic, beautiful, and full of interesting cuisines 🍲🐟. Can’t say no to a Bouillabaisse even on a hot August day!
000
Sebastian Huckleberry @sourishkrout.com · 05/08/2025
Monaco 🇲🇨 is a reverse Las Vegas. In Vegas the glitz is fake and the fun is real. Draw your own conclusions 😆.
000
Sebastian Huckleberry @sourishkrout.com · 05/08/2025
The best the Liguria’s Genova has to offer. - Fish with ground garbanzo bean flour chips - Pesto Genovese, invented here Yummys 😋
100
Sebastian Huckleberry @sourishkrout.com · 30/06/2025
Just shy of two decades, I can call myself a dual-citizen 🇩🇪🇺🇸 today. Thanks, Uncle Sam 🎉
110
Sebastian Huckleberry @sourishkrout.com · 02/06/2025
Two weeks of barrel aging commenced. Here’s the final product: Black Manhattan 🍸. I love how it turned out.
Photo of bottle with finished Black Manhattan next to the barrel it aged in for the last two weeks.
010
Sebastian Huckleberry @sourishkrout.com · 22/05/2025
Another edition of the soda plus a splash of fortified wine! Today it’s Sherry. The same Sherry that spent two weeks aging in an oak barrel. Tasty 😋!
Photo of a La Croix can next to a glass with soda water, ice cube, and a splash of barrel aged sherry.
010
Sebastian Huckleberry @sourishkrout.com · 19/05/2025
After 15d of seasoning with cream sherry, the barrel is ready for aging Black Manhattan. This should take anywhere up to 15d with sampling every other day to avoid too much oak flavor.
Photo of retrieving the aged sherry from the barrel.Photo of lined up ingredients for Black Manhattan that’s going in the barrel.
010
Sebastian Huckleberry @sourishkrout.com · 06/05/2025
My favorite new low ABV drink is a Faux Americano. It’s really just a good splash of sweet vermouth with soda water. No Campari.
Photo of glass with faux americano next to a emptying can of lacroix
100
Sebastian Huckleberry @sourishkrout.com · 04/05/2025
All egresses and ingresses plugged on this barrel full of sherry. Ready to breeze right through a SOCs review, just like a good ol’ #Kubernetes cluster. Two weeks of seasoning is starting now to get it ready to age some cocktail 🍸.
Photo of oak barrel which now holds the full contents of the cream sherry bottle.
010
Sebastian Huckleberry @sourishkrout.com · 03/05/2025
After rinsing you fill the barrel with water to seal it. It’s like: kubectl apply -f agua.yaml Takes ~48h to finish 😆.
Photos of oak barrel plugged full of water to make it swell and seal up.
111
Sebastian Huckleberry @sourishkrout.com · 03/05/2025
This is what a 1 liter oak barrel for home-aging spirits looks like. What’s funny about the 1L is that it’s displacing not holding liquid.
Photo of a brand new 1L oak barrel used for aging spirits
010
Sebastian Huckleberry @sourishkrout.com · 30/04/2025
Looking forward to JS/TS tooling with true concurrency and near-native speed. The memory consumption of building something like "VS Code" is staggering.
Screenshot of node running into its JS heap memory limit
000
Sebastian Huckleberry @sourishkrout.com · 28/04/2025
Integration tests for specific versions, including pre-releases, are now easy! Previously, it required special file placements and specific operation orders to avoid overwriting the file. ### Exported in runme.dev as KernelBinary RunmeKernel | link-release --version v3.13.2-rc.2 $TARGET_PLATFORM
Screenshot of the integration tests running against a release candidate
000
Sebastian Huckleberry @sourishkrout.com · 18/04/2025
Inside the checked-out repo, it's as easy as running a single command: $ runme run IntegrationTests
000
Sebastian Huckleberry @sourishkrout.com · 18/04/2025
Now it takes me a hot 30s to check if the pre-conditions for my integration tests are met, and not a 15min wait until the pipeline trips over missing values. Can your YAML do that?
011
Sebastian Huckleberry @sourishkrout.com · 18/04/2025
Here's the first successful completion of the unit and integration pipeline running in GitHub Actions. github.com/runmedev/vsc... The ENV is completely portable with @runme.dev's direnv support. Tests are aware of PRs, so the shell calls an extra "gha-job" @dagger.io function to pass metadata.
Running the integration pipeline with the artifact under test, entirely built from its sources. The only difference is the additional "gha-job" function that passed GitHub Actions-specific metadata.
211
Sebastian Huckleberry @sourishkrout.com · 17/04/2025
First successful pass of @runme.dev's e2e test suite running fully @dagger.io-ized. This is no easy feat, it's running VS Code (aka Chromium) inside of Virtual X Frame Buffer (vxfb), literally mimicking user actions leveraging @webdriver.io. It's a one-liner now. gist.github.com/sourishkrout...
End-to-end test suite reporting success.
132
Sebastian Huckleberry @sourishkrout.com · 14/04/2025
Software #documentation nihilism is absurd; especially in the middle. 🤔 X-axis, as in "junior, over senior, to staff/principal". Thanks for the inspiration, @jeremy.lewi.us. I couldn't help but make a meme.
The journey from Junior, over Senior, to Staff/Principal where both ends of the spectrum value docs but the middle dismisses it.
010
Sebastian Huckleberry @sourishkrout.com · 07/04/2025
Did you know @runme.dev natively supports #DotEnv and #Direnv? Makes it super simple to leverage your project's existing environment configuration. More in the docs: docs.runme.dev/configuratio...
A screenshot showing an environment populated with direnv/.envrc
010
Sebastian Huckleberry @sourishkrout.com · 02/04/2025
It's incredibly satisfying to roll a deploy of a docs website on that local #Kubernetes cluster you haven't touched in a long time, and it just works. Just kidding, it never works 🤣. What works is Workload Identity that does not discriminate between Humans and Machines. runme.dev/blog/typed-e...
Show how a Runme docs preview was successfully deployed using human-centric workload identity.
000
Sebastian Huckleberry @sourishkrout.com · 27/03/2025
Building @runme.dev, we've dreamed of a Cloud-Native Shell—leverage everything great about the command line while effortlessly transcending any single machine. That's @dagger.io Shell. Express artifacts, builds, and AI agents "Code as Docs™️" style with Runme for Dagger: docs.runme.dev/guide/dagger
042
Sebastian Huckleberry @sourishkrout.com · 25/03/2025
That's a bit aggressive, no?
000
Sebastian Huckleberry @sourishkrout.com · 19/03/2025
I am surprised how far I have gotten just using Cursor's trial plan.
Screenshot showing how Cursor usage limits were exceeded
000
Sebastian Huckleberry @sourishkrout.com · 12/03/2025
Gotta give it to @windsurfai.bsky.social... the centrally overlaid command palette is much better. It's just too easy to overlook VS Code's top anchored UX elements.
The command palette opens in the center versus anchored at the top like a browser address bar
000
Sebastian Huckleberry @sourishkrout.com · 12/03/2025
The VS Code platform & extension, being the Chromium of IDE, is a bit of a dream come true. Competitors can now stand out with key UX details and productivity features, with AI integration being the most hyped. @runme.dev just works across @windsurfai.bsky.social, Cursor.com, and VS Code flavors.
Runme notebook inside of the AI-enabled Windsurf IDE
000
Sebastian Huckleberry @sourishkrout.com · 11/03/2025
Hot off the @runme.dev press, native #direnv support! While still deactivated by default, it will become enabled out-of-the-box once it holds up in real-world repos. Looking good. It won't be long.
000
Sebastian Huckleberry @sourishkrout.com · 06/03/2025
The fact that it is so effortless to switch between Code, Cursor (running @runme.dev), Coder, Codespaces, etc., proves that the VS Code project is factually the Chromium of IDEs. When building an IDE, you better differentiate beyond the core experience of editor, terminal, LSP, and extensions.
The screenshot shows a Runme notebook hosted inside of the Cursor IDE.
000
Sebastian Huckleberry @sourishkrout.com · 05/03/2025
It's odd how #direnv has a neutral state when it first encounters a new .envrc'd directory. Once you ran allow or deny, you can't go back using the CLI. To restore neutral, a "direnv reset" would be helpful. The only other option is monkeying with direnv state files.
This is a screenshot of a terminal demonstrating the allow/deny/neutral issue.
000
Sebastian Huckleberry @sourishkrout.com · 16/01/2025
Just for the record. The new merging UX looks tight. I want that "auto-merge once checks pass" nonetheless.
Screenshot of New Merge UX
000
Sebastian Huckleberry @sourishkrout.com · 15/01/2025
A tremendous timesaver in the non-AI department for @github.com would be a simple checkbox to "auto-merge once checks pass" as part of this drop-down. Right now, I choose between waiting for e2e tests to pass or moving on and forgetting to come back 🤔. Not fantastic.
Screenshot of a Pull Request merging options drop down
210