Case 01
ProjectView PM: an AI PMO office with a human in the loop
- Problem
- Weekly portfolio and status reporting is slow, manual and inconsistent.
- What I built
- A staged pipeline of seven tool-using AI agents. A Program Coordinator; then Risk & Dependency, Resource and Finance agents working in parallel; a PMO Analyst that scores health (Green / Amber / Red) and KPIs; an Executive Communications Assistant that assembles one report and a steering deck; and a Stakeholder Communicator. Agents share findings through a common findings board and ground their work on the app's own data: budget, tasks, timeline, meeting notes and calendar.
- Governance pattern
- Every consequential action (sending email, creating tasks, editing timelines, updating resources) is queued to a human inbox to review, edit, reject or approve, and nothing executes without approval. Every run is logged with the tools used and tokens spent, for full traceability. It runs weekly on a schedule or on demand.
- Result
- About 40% less manual reporting effort in my own use.
What it taught me
"Put human approval at the point of consequence, not everywhere. That's what makes agents safe enough to use."