AutomateAI  /  Self-healing
Self-healing replay

Why re-run everything when only one step broke?

Full AI reasoning is expensive and unnecessary for a test that already passed. AutomateAI records the passing path once, replays it deterministically, and only calls the model back in to fix the exact step that stopped working.

How a run behaves

Reason once. Replay cheaply. Heal narrowly.

01

Record a passing run

The agent completes the test using full reasoning, step by step, the first time it runs.

02

Generate a deterministic script

The passing steps — including their assertions — are converted into a replayable script, versioned per test.

03

Replay on every future run

No model calls. The script plays back directly against your app, verifying the same assertions each time.

⚠ ON FAILURE

Heal just that step

The agent reasons about the one broken step only, fixes it, and immediately hands control back to replay.

replay · checkout-flow.web · v4
live
Bounded, not open-ended

A circuit breaker, not an infinite retry loop

If a run needs more than a handful of heals, that's usually a sign the app changed more than one step's worth. Past that point, AutomateAI falls back to finishing the run in full reasoning instead of patching indefinitely.

3 heals max per run then full reasoning fallback every fallback is logged, not silent
Scripts don't go stale silently

Editing the test invalidates the old script

If you change what a test is actually supposed to do, AutomateAI retires the old deterministic script automatically instead of replaying a script that no longer matches the intent.

auto-invalidated on edit re-recorded on next run