Tasks & runs
Almost nothing in Structured AI makes you wait. Checks, comparisons, audits, uploads and agent tasks all run in the background while you keep reviewing - which raises the obvious question of where you watch them.
The tasks tray
The tray is the answer. It lists everything currently running and everything that finished recently, with a status and an estimate of how much longer it needs.
What shows up there:
| Run | Started by |
|---|---|
| QA/QC run | Project QAQC, or the upload wizard |
| Custom checks | A checklist or a check you built |
| Spec audit | Spec Audit |
| Compare versions | Compare Versions |
| Compare against control | Compare Against Control |
| Comment closure | Comment Closure |
| Shop drawing review | Shop Drawing Review |
| Code review | A code-compliance check |
| Code ingestion | Uploading a book to the code library |
| Document processing | Any upload |
| Agent task | A job you gave the chat |
Clicking a finished run takes you to its results. Clicking through to your full history opens My Performance.
One state is worth naming: Waiting for documents. A check launched while the project's documents are still processing is parked rather than run - it sits in the tray with that badge and starts automatically once processing finishes.
The tasks tray with work in progress
Being told instead of watching
You don't have to keep the tray open:
- Turn on email-on-completion in Settings → Preferences and you'll get an email when your runs finish.
- Notifications cover run completions and failures in the app.
When a run is slower than you expected
Set size is the usual explanation. A check that reasons across every sheet of a 600-page set genuinely takes longer than the same check on a 40-page one, and the tray's estimate accounts for that.
A run that has clearly stalled, or came back with an error, is worth reporting - see Troubleshooting.
Leaving and coming back
Closing the tab doesn't cancel anything. Runs are server-side: they carry on, land in the tray and the notification bell, and their results are waiting in the project when you come back. The same is true of a long chat job - the thread picks up where it left off.