AI Decide
Asks a model a yes/no question about the run and branches on the answer.
Branch based on an AI decision
Reach for it when
- A judgement no selector expresses — “does this screen look broken?”
- Deciding whether a log is noisy or genuinely bad before waking anyone.
Every setting this block has. A field marked only when appears once you have chosen the mode above it — those are alternatives to each other, not extra things to fill in.
- Decision criterionRequiredstarts at —
True takes the top branch, false the bottom. Drag in a field from the left to test it.
State the criterion
Required, unlike Analyze: a branch with no stated criterion is a coin toss dressed up as a decision. Write the question so that “true” has one clear meaning.
the run attaches itselfevery step and its settingswhat each step returnedcaptured logsthe last screenshotDid the app show a crash dialog or an error screen at any point?No error screen appeared; the flow completed to the confirmation page.
truefalseConnect both sides
`true` takes the top branch, `false` the bottom — the same shape as Assert and If / Branch.
AI DecideIdletruefalseIt runs in graph order, on the worker
Unlike AI Analyze, this one cannot wait for a post-run pass in the browser: by then the branch below it would already have executed. So it runs where it sits, including on a scheduled run.
AI DecideFail the Run (on false)
- `true` or `false`, and the reasoning behind it.
- A model is not a selector. Where a concrete check exists — an element, a status code, a log line — Assert and If / Branch are cheaper and will not change their mind.
- A vague criterion gives a confident answer to a different question. “Is it OK?” is not a criterion.
- Inference is metered against your plan's monthly allowance, and this one runs on every device the branch reaches.