Developer Uses Git-Based Board to Manage AI Coding Agents
A developer describes a system using markdown files and git snapshots to coordinate AI agents and reduce cognitive overload.
The author experiences 'AI Brain Fry' when managing multiple threads with terminal-based AI tools. To address this, every ticket in the system is a markdown file located in the git repository next to the relevant code. The state of the work board is maintained as a committed snapshot in git. An agent works on this board in cycles to research tickets, build safe items, run tests, and verify deployments. When building sensitive code, the agent works on a branch and requests approval before proceeding. The agent never deploys to production, a task that remains with the human author.
Items requiring human judgment are collected in a single place called the 'owner inbox'. A script reads the entire board and provides a prioritized list of items waiting for the author's decision. The author's reading surface is a Claude artifact that displays one page rebuilt from git on every cycle. The author takes one item at a time, loads its context, makes a decision or requests an action, and then closes it. Closing a ticket puts it back into the queue so the context does not remain in the author's memory.
The author notes that the system is still exhausting but prevents feelings of distraction or being lost. A remaining challenge is ensuring tickets do not enter endless loops while waiting for a ruling. The author describes the system as a living organism that they continue to improve.
Sources
Every sentence of this story was checked against facts quoted from these articles. How we make the news
Spotted an error? Tell us
