The build log.

Why I Don't Believe an Agent When It Says "Done"

Before lessons and forecasts, the one thing it all stands on: a model's work has to be checked from the outside, not taken on its own word.

How Swarmery Came to Be. Part 3: How the Agent System Stopped Being a Black Box

How I started to see what my agents are really doing, and how the system learns from its own work.

How Swarmery Came to Be. Part 2: Why I Replaced the Kanban Board with Plans

Why cards on a board did not work for my agents, and how the planner became the heart of Swarmery.

How Swarmery Came to Be. Part 1: From ChatGPT to Claude Code

What a Book from 2004 Taught Me About My Agents

How to tell if an agent really understood the task, and how a system can learn when the model can't.

Retro: The Agent System That Improves Itself

How my agent system learned to learn from its own mistakes.

Swarmery: How I Stopped Copy-Pasting Agents Between Projects, and Started Driving Them

Syncing agents was the easy part. The interesting problems started after: seeing what 15 agents across 6 projects actually do, measuring whether they're any good, and letting a board drive them.