Linux's b4 Adds AI-Assisted Code Review TUI - Now Dogfooding On Its Own Patches
The b4 tool that many kernel developers rely on for patch workflow just hit a new milestone: a text-based UI for AI-assisted code review is up and running, and it's already been used to review b4's own patches. It's opt-in, sits alongside your normal b4 flow, and aims to save time while catching issues earlier.
Konstantin Ryabitsev at the Linux Foundation is leading the "b4 review tui" work. He shared that the first actual agent-assisted review is complete, noting, "Lots of refinements still needed, but it's doing useful things, at least." The initial runs have used Claude Code as the reviewing agent.
With b4 being the de facto tool across kernel development, having AI help available right in a TUI is a practical step. It doesn't replace human review or maintainer judgment. It's another lens you can turn on when you want it.
Why it matters for kernel developers
- Faster triage: surface risky diffs, missing checks, or suspicious patterns across long series.
- Prompted consistency: apply the same review heuristics on every iteration to reduce drift.
- Fits the workflow: runs in a terminal, respects email/patch series habits, and remains optional.
- Signal, not verdict: encourages better commit messages, clearer intent, and tighter diffs.
How to put it to work (practically)
- Run it as a pre-flight check on v1 patchsets to catch low-hanging issues before sending.
- As a maintainer, use it to prioritize which patches need deep attention first.
- Ask the agent to cross-check commit messages against the actual diff for mismatches.
- Keep a small set of prompts for your subsystem style and gotchas; reuse them across series.
What to watch
- Early days: expect rough edges and keep human review as the source of truth.
- Prompts matter: better guidance in the TUI leads to more useful feedback.
- Scope it: use it where the stakes are appropriate (e.g., public series, non-sensitive code paths).
Related work
Separately, Chris Mason at Meta has been working on new AI code review prompt helpers for kernel development. The direction is clear: make AI a helpful assistant that fits the way the kernel community already works.
Where to follow and try
- b4 source on kernel.org for updates on the review TUI and patch activity.
- If you're experimenting with Claude for code review, this resource can help: Claude Certification for Developers.
Bottom line: this is a practical add-on for people who live in b4 and want AI signal without changing their patch flow. It's optional, useful already, and getting better.
Your membership also unlocks:
