Skip to main content

AI Project Manager

The AI Project Manager (AI-PM) reads everything your team produces — requests, dates, standups and the activity log — and turns it into a forecast: what's on track, what's at risk, where the bottlenecks are, and who's overloaded. It's the predictive counterpart to the Operations dashboard.

Open Manage → AI Project Manager (/ai-pm).

Almost all of it is free

The analytics below are deterministic — computed with plain queries over your data, no LLM, no credits. Only the optional AI briefing (a written summary) makes a model call, and that's billed as a small platform cost, on demand — it doesn't touch your AI credits.

What it computes​

Health score​

An overall org health rating — on track, at risk, or overdue — rolled up from the state of all active requests.

Timeline risk (per request)​

For each request, the AI-PM estimates a completion risk:

  • It computes velocity from the standup history (how fast progress is actually accruing).
  • It projects a completion date from that velocity and compares it to the due date.
  • Overdue requests are flagged at ~95% risk; behind-pace requests get a heuristic risk score; on-pace requests sit low.

Each request gets a risk % bar and a projected forecast so you can see trouble before the deadline, not after.

Bottleneck detection​

By pairing STAGE_STARTED / STAGE_COMPLETED events from the activity log, the AI-PM measures average dwell time per workflow stage and shows which stage consumes the largest share of total time — e.g. "the Publish stage accounts for most of the delay."

Capacity analysis​

Open requests per assignee, flagging anyone overloaded (roughly five or more open requests) so you can rebalance.

Recommendations​

A set of heuristic suggestions based on the above — what to reassign, what to escalate, what's blocking the most work.

The AI briefing​

Press ✨ AI briefing to generate a concise, natural-language summary of the current state. This makes a single call to Claude (claude-haiku-4-5) and produces a sharp, specific narrative — it names the requests at risk, the dominant bottleneck, and the recommended action.

This is the only part of the AI-PM that calls a model. The numbers it narrates are computed deterministically; the model only puts them into words.

How it stays accurate​

The AI-PM needs no special data capture — it's a read-model over data your team already generates by working normally:

  • post standups and velocity/forecasts get sharper,
  • use workflow stages and bottleneck detection lights up,
  • set due dates and timeline risk becomes meaningful.

In other words, the more disciplined your Requests and Workflows hygiene, the better the forecasts.

Next​