How It Works
PAPI runs a repeating cycle. Each cycle produces working software and a clear record of what happened, what surprised the team, and what to do next.
The team (or the AI agent) reviews the backlog of tasks and selects a batch for the cycle. Each task gets a detailed specification — a “BUILD HANDOFF” — that describes exactly what to build, what not to build, and how to verify it's done.
The AI agent builds each task, one at a time. Every build produces a report: actual effort vs estimated, any surprises encountered, and architectural notes for future builders. The human reviews each task as it completes.
A human reviews each completed build. They can approve it, request changes, or reject it entirely. This is where quality gates live — nothing ships without a human saying it's good.
When all tasks are approved, the cycle closes with a versioned release. PAPI generates a changelog from the build reports, creates a git tag, and pushes to the repository. The project is now ready for the next cycle.
What makes it different
Traditional project management tools (Jira, Linear, Asana) track tasks but don't provide structure for AI-assisted work. They don't generate specs, track estimation accuracy, or surface architectural decisions. PAPI does all of this because it's designed for the specific challenge of keeping AI agents on track.
The key insight: AI coding tools are fast but directionless. PAPI provides the direction through structured cycles, and the accountability through build reports and human reviews. The result is software that ships predictably, with a complete audit trail of every decision along the way.