Autopilot — what it actually does

Once configured, Autopilot works from its tab on the Noot, which has three parts.

Plan this project

Generate a plan turns the brief into board columns, phases, tasks with owners, estimates and dates, plus the meetings it thinks the project needs. The result appears as a Proposed plan — a summary and a table of Task / Phase / Priority / Assignee / Due / Est. — which you then apply in one go or discard. Nothing lands until you say so.

Run a management cycle

A cycle reads the project's meetings, action items, board, schedule health and team capacity, then does what a project manager would — inside the autonomy level you set. What it may not apply on its own lands below for approval.

What a cycle can do

• Assign work by real capacity — judged across *every* project a person is on, using team specialty as a signal. If your guardrails empty the pool, it stops rather than falling back to "anyone in the workspace".
• Ask owners for status when work is behind pace, and read the natural-language answer — see Status check-ins.
• Ripple dependencies — when a prerequisite slips, it moves the work waiting on it (start *and* due date, duration preserved, only ever forward) and tells the owner before they're blocked.
• Propose closing a task whose linked pull request has been merged. A proposal, never a silent close — only the team knows whether the task meant more than the PR.
• Log risks to RAID from four sources: blocked dependencies, risks raised in meetings, check-ins answered "at risk", and capacity problems. Each carries its kind, its impact, the meeting that raised it and the task it threatens.

Autopilot activity — and undoing things

Every change Autopilot proposed or made is listed with its reasoning. Approve what's waiting, dismiss what you disagree with, and undo anything it already did.