The command center.

I build software teams whose members happen to be AI. Here is the crew, how a mission runs, and the principles that keep autonomy honest.

The crew

A live map of the agent team. Click any node to see what it does and how it talks to the others.

Planner

Decomposes goals into a task graph

Breaks the goal into tasks with dependencies, assigns budgets, and re-plans when workers report blockers.

Define

A human writes the goal and its constraints. That is the last mandatory human input until review.

Plan

The planner decomposes the goal into a task graph with dependencies, owners, and hard budgets per task.

Execute

Workers claim tasks in parallel, discover tools through MCP, and hand off artifacts with typed messages.

Review

A reviewer agent runs tests and checks diffs against the goal. Irreversible actions wait for a human tap.

Ship

Approved work merges and deploys. Lessons learned are written to durable memory for the next mission.

Budgets, heartbeats, and checkpoints keep an overnight run safe. A smarter model without supervision is just a faster way to drift.

Agents exchange proposals, reviews, handoffs, and blocks as structured messages, so the orchestrator reasons about state instead of parsing text.

Retrieval follows the task graph. An agent building the billing service never drowns in frontend notes.

Agents own the how; people own the what and the why. Every irreversible action routes through a human.