PMO: AI Agents
The agents that turn the PMO from reporting into prediction, the eleven signals they watch continuously, and the decisions that stay with a project manager.
Markdown. No sign-up, no email.
A traditional PMO reports the past. Someone updates a spreadsheet on Thursday, a status pack goes out on Friday, and by Monday the numbers describe a project that no longer exists.
The change here is not that reporting gets faster. It is that status stops being something people assemble and becomes something the systems already know, which frees the PMO to do the thing that actually moves a project: intervene early.
The eleven signals, watched continuously#
| Signal | Read from | What a bad reading looks like |
|---|---|---|
| Scope | Tickets, change requests | Scope growing while the date holds |
| Schedule | Task completion against plan | Slippage concentrated in one workstream |
| Budget | Spend against forecast | Variance rising before anyone flags it |
| Resources | Assignment and utilisation | One person on the critical path of three projects |
| Risks | The register, plus signals nobody logged | A register untouched for a month |
| Dependencies | Cross-project links | A dependency owned by a team that does not know |
| Tasks | Movement, or the absence of it | Work in progress with no state change for days |
| Client commitments | Contracts, meeting records | A promise made in a call and never entered anywhere |
| Delays | Actual against baseline | Small delays that have not yet compounded |
| Quality | Defects, rework, escaped bugs | Delivery on time by shipping problems forward |
| Productivity | Throughput per period | Falling while headcount is flat |
The four agents#
Status agent#
Reconstructs project state from the systems rather than from anyone's memory. Percentage complete, by delivered scope rather than by tasks closed, because closing tasks is easy and delivering is not.
Reach: ticketing, repositories, finance exports. Read-only. Output: a current position, continuously, with the evidence for each number.
Prediction agent#
Turns the eleven signals into a forecast. This is the shift from reporting to prediction:
Project X, 72% complete. Schedule risk HIGH. Resource risk MEDIUM. Budget variance +8%. Critical dependency: API integration. Predicted delay: 6 days.
Every line carries its basis. "Schedule risk HIGH" is worthless without because three of the last five milestones slipped and the remaining work is concentrated on one person. A prediction that cannot be argued with cannot be acted on either.
Reach: the status agent's output plus history from comparable projects. Output: a forecast with its reasoning, refreshed daily.
Dependency agent#
Maps what waits on what, across projects rather than inside one. Most missed dependencies are not unknown, they are known by one team and invisible to the other.
Reach: plans and tickets across the portfolio. Output: the dependency graph, and specifically the ones whose owner has not acknowledged them.
Commitment agent#
Watches what was promised, to whom, by when. Contract dates, meeting minutes, email. Surfaces slippage against a client commitment before the client does.
Reach: contracts and meeting records. Output: commitments at risk, ranked by how visible the miss will be.
What stays with the project manager#
- Declaring a project at risk. The agent forecasts; naming it is an act with consequences for people and clients, and it belongs to someone accountable.
- Re-planning. Changing scope, date or budget is a negotiation, not a calculation.
- Anything said to a client. Always.
- Escalation. Deciding a problem has outgrown the project.
- Resource reallocation. Moving a person is a decision about a person.
Why prediction changes the job rather than the paperwork#
A PMO that reports tells you a project is eight days late. A PMO that predicts tells you it will be eight days late while there are still fifteen days left, which is the only window in which the information is worth anything.
The measure of this function is therefore not report accuracy or dashboard freshness. It is how often an intervention happened early enough to change the outcome, and that is what the KPI set is built around.