Agentic Project Management (APM)
An open-source framework that coordinates a team of specialized AI agents, a Planner, a Manager and Workers, to manage a large software project through structured planning documents and reviewed task assignments instead of one long, degrading chat.
Agentic Project Management (APM) structures work on an ambitious software project across a team of AI agents rather than one continuous chat, which tends to degrade as it grows: the assistant loses track of requirements and starts producing bad code or hallucinated details.
A Planner conducts structured project discovery and breaks requirements into a specification, a plan for how work is organized, and rules for how work is performed. A Manager assigns tasks to Workers, reviews completed work and maintains project state from execution summaries rather than raw code. Workers execute tasks within a defined domain such as frontend, backend or API work, each starting from a self-contained task prompt, then validate their own work, log it to memory and report back.
Unlike a fresh subagent per task, a Worker accumulates working context across assignments within its domain, and a structured handoff transfers that context to a new instance of the same agent when its context fills, rather than discarding it. Project state lives in files outside any agent's context, and a person mediates every exchange between agents by running the command each agent names, keeping every step visible and reviewable.