Kai Context

Find and fix bugs beyond the diff. An independent code reviewer that traces your changes through a semantic map of your codebase. →

[Jacob @jakecodes]

We wrote up how to read Martian’s bench → https://t.co/wJKFa1IxXP

[Jacob @jakecodes]

In the next version of Kai… it flags bugs the diff never touched.

[Jacob @jakecodes]

Code review in 2026:

✅ Diff looks great
✅ Tests pass
✅ 3 AI bots approved
❌ The new code can never actually run

Review the code around the change, not just the change.

[Kai Context @kaicontextdev]

🕰️ We made an interactive history of code review: 1976 → today.

From printouts in meeting rooms, to email patches, to CI bots, to multiple AI agents reviewing every PR.

Favorite finding: AI-written PRs get merged 32.7% of the time. Human-written: 84%.

https://t.co/R0zlplEzxU

[Kai Context @kaicontextdev]

@saranormous 👋

[Kai Context @kaicontextdev]

@PremiumGoblin Do you also see it as a new challenge?

[Kai Context @kaicontextdev]

@bendee983 Greenfield feels low friction to get started, however a mature codebase we've cultivated requires a psychological shift.

(legal compliance aside)

[Kai Context @kaicontextdev]

@MatthiasWagner Agree!

[Kai Context @kaicontextdev]

@andrewchen Definitely an issue that's being worked on.

In the meantime an easy solution is to fire off prompts to the cloud to open PRs and then have an agent merge them in.

[Kai Context @kaicontextdev]

@Prathkum Agree. Things are moving in this space!

[Kai Context @kaicontextdev]

@boshen_c Have had good results with Cursor, Claude, Codex. Any specific needs/constraints?

[Kai Context @kaicontextdev]

@LyalinDotCom I guess it depends 😅 Device? Prompt needs/complexity?

[Kai Context @kaicontextdev]

@icanvardar You don't have to wait for them to finish. You can run them in a worktree, in the cloud, a second machine, or maybe increase the scope of the work to get more done and go watch a movie 🙂

[Kai Context @kaicontextdev]

@Tech_girl Maybe for the same reason folks have subscriptions to Netflix, Apple TV, Disney, etc...

My guess: optionality, variety, affinity, and to avoid lock-in.
🍿

[Kai Context @kaicontextdev]

🚀 We've been heads down.

Kai's gone from 0.11 → 0.35:

- Proper benchmark harness (roughly doubled our SWE-- bench pass rate)
- GitHub CI integration
- Lots of new features, lots of issues resolved

More to share soon.