Jack Brown

automating testing @getlark (YC S25). prev @stripe. climb with me

[Jack Brown @jackbr513]

Hands down best way to handle agentic memory

[Vijit Dhingra @VijitDhingra1]

https://t.co/tYEuz2l8ty

[Jack Brown @jackbr513]

Am I the only one who thinks Claude Opus 4.8 hasn't really improved over 4.7 wrt coding tasks?

[Lark @getlark]

Lark is sponsoring AI Dev Summit this week.

If you’re here, come say hi.

We’d love to chat about faster releases, fewer regressions, and AI-powered testing.

[Jack Brown @jackbr513]

Nationalize utility companies

[More Perfect Union @MorePerfectUS]

Nearly 50,000 people in the Lake Tahoe area have been told that their utility will stop providing power to them, because it's redirecting that power to data centers.

NV Energy, the Nevada utility that has supplied most of Lake Tahoe’s electricity for decades, says that next year it will stop servicing homes in the area, and instead direct that electricity to the growing demand from Nevada data centers.

Northern Nevada is one of the fastest-growing data-center corridors in the country.

https://t.co/nTGJAMbwAP

[Jack Brown @jackbr513]

tag this person

[Jack Brown @jackbr513]

@haydenbleasel How does this handle file storage in untrusted environments? https://t.co/eRygtObtsh handles this with pre signed upload URLs streamed over websocket

[Lark @getlark]

E2E tests shouldn’t break every time your UI evolves.

Today, one of Lark’s own E2E tests failed after we shipped a UI change. Lark summarized the failure, identified it as a test issue rather than an app bug, and repaired the test automatically.

The test has been passing consistently since.

[Jack Brown @jackbr513]

@FredKSchott How does this compare to something like runtimeuse? https://t.co/eRygtObtsh

[Jack Brown @jackbr513]

@ClaudeDevs Okay actually idek how to pronounce this

[Jack Brown @jackbr513]

Okay keep it together @ClaudeDevs

[Jack Brown @jackbr513]

AI solving a 60-year-old math problem is impressive. what's more impressive is that most teams can't get their AI to reliably fill out a form. the gap between research demos and production workflows is still enormous.

[Jack Brown @jackbr513]

the test suite that takes 45 minutes to run is the one nobody fixes broken tests in. speed is a feature, and in CI it's the only feature that affects whether engineers actually trust the output.

[Jack Brown @jackbr513]

still the best tax strategy tbh

[Jack Brown @jackbr513]

Who else is using openclaw to do their taxes?

[Vijit Dhingra @VijitDhingra1]

Over 100 QA reports generated since launch.

We’re seeing real issues uncovered across SDKs, APIs, and dashboards—helping teams catch and fix problems before users report them.

Fast, structured, reproducible.

[Vijit Dhingra @VijitDhingra1]

OpenAI just added support for popular sandbox providers in their Agents SDK — using a “harness separate from compute” model.

At @getlark, we take a different approach: “harness in compute.”
We want agents to behave more like local Claude or Codex — fully embedded, with everything running inside the compute environment.

That’s why we open-sourced https://t.co/7UPA8tD0HC — an agent-agnostic framework for running long-lived agents in any sandbox.

Works with @e2b, @daytonaio, @vercel, @modal, and more.

[Lark @getlark]

Create Linear issues automatically when tests fail or bugs are discovered.

Catch bugs → create issues → fix faster.

No missed failures. No manual triage.

[Vijit Dhingra @VijitDhingra1]

We just shipped QA Reports.

Deploy AI agents to test your product, uncover issues, and get back a structured report with reproducible test cases.

We ran it on the Vercel Sandboxes API. It mapped test areas, executed flows, and surfaced real issues in minutes.

Try it: https://t.co/84Ebxuw8KR
Example report (Vercel Sandboxes API): https://t.co/IOqc7sMtSr

[Jack Brown @jackbr513]

@bcherny We're launching bug fixes as features now?

[Jack Brown @jackbr513]

@tenobrus @Jason facts but also chill on the R word god damn