AI Review — a second opinion

Beyond the automatic checks, the AI Review button (top-right) asks an AI to critique a layer for quality — vague names, missing pieces, over-wiring, a workflow that never completes. It's a second opinion, always yours to accept, act on, or reject.

The loop

The AI Review panel is a closure dashboard: each layer shows a status and the Review → act → re-review loop.

1. Review a layer → the AI returns a list of findings, each with a severity (a concern vs. a subjective suggestion), a message, and a concrete suggestion.
2. For each finding, choose how to close it:
   • Fix — apply the suggestion (see Fixing concerns);
   • Ignore — accept it for good (see Ignoring concerns).
3. Re-review to confirm — the panel shows you exactly what changed (see Progress & loops).

It's advisory, and it doesn't converge to zero

The AI reviewer is calibrated to always find something — a few subjective suggestions is a fine place to stop. Findings are judgment calls, not a checklist that hits zero. The panel says so, and nudges you when you've refined a layer enough.

Apply vs. Fix

There are two different "make it better" actions, and the difference matters:

• Apply (the batch button) feeds your accepted points back and regenerates the whole layer. It's good for early, sparse layers — but because it rebuilds everything, it can undo earlier fixes and oscillate. Re-review to confirm it actually helped.
• Fix (per finding) applies one precise edit and nothing else — it converges.

The next pages cover both, and how to tell progress from churn.

Cost

Each Review, Apply, or Fix that calls the model costs a small amount (shown as a spend estimate). This feature needs a real model, so it's available when you run your own instance with an API key — not on the keyless public demo.