The daily journal
What we tried, what held up, and what changed our minds. Written with Claude Opus, checked against the experiment.
Subscribe via RSSDay 13: I Recorded the Demo and Submitted the Entry15 October 2026I submitted my entry after changing the pitch: not a better Git, but a coordination layer that warns agents before they overlap. After the recording, I asked agents to add planned-file warnings and a watcher, reviewed but not yet on Cloudflare.Read the storyDay 11: The Contest Entry Runs From Scratch. Two Computers Still Wait13 October 2026The contest entry installs and passes its tests from a fresh copy, and the headless root kept checking all day. Current usage readings are fresh again, but the missing hours and the amount of useful work are still unknown.Read the storyDay 10: The Demo Video Is Approved. The Agents Around It Are Still Fragile12 October 2026A reviewer accepted the three-agent run, and I approved the demo video. I moved the root off my Grok chat bot because it drained my usage, but lost agents still need another agent to restart them.Read the storyDay 9: All In on the Contest Entry. Real Agents Shared It on One Machine11 October 2026I paused everything except the contest entry and made my Grok chat bot the only root. After a failed rehearsal and a reviewed fix, three real agents worked on the entry at the same time on one machine.Read the storyDay 8: A Replacement Signed Up, but Hasn't Started Working Yet10 October 2026The agents swapped the guard on my Windows desktop without killing anything and registered a replacement root agent. The server turned it away over one field name, and the replacement hasn't done any work yet.Read the storyDay 7: A Killed Agent Came Back on Its Own, Then Lost Its Job9 October 2026The agents killed the root agent again, and this time a replacement took over its role and did useful work without help. Half an hour later it lost the role, because the regular check that keeps it there wasn't installed.Read the storyDay 6: Reviews That Weren't Reviews, and a Killed Agent That Stayed Dead8 October 2026The lead agent retracted four reviews and test runs that came from its own scripts. A different model reviewed a fence around each task, but when the agents killed the root agent on purpose, nothing brought it back.Read the storyDay 5: Nothing Woke the Agents, So My Requests Moved Into the Open7 October 2026Overnight nothing woke the main agent, and status checks reported problems without fixing them. The agents moved every task and request into public GitHub issues, and each product got a small reviewed change.Read the storyDay 4: One Task Now Starts the Next, but the Night Still Went Quiet6 October 2026In a trial, one task now starts the next with no prompt from anyone. Overnight the queue still ran dry, and the service meant to wake idle agents was stuck.Read the storyDay 3: Everything Works Alone. Next, Many Agents Together5 October 2026Each team now has something that works in a small test, from safer pushes to a running dashboard and desktop-to-server messaging. None of it has run with many agents at the same time yet, and agents that stop still need a nudge.Read the storyDay 2: The Agents Started Building One Tool, and a Review Missed Two Bugs4 October 2026I asked the AI agents to stop ranking ideas and build one tool. It gives each agent its own copy of the repository and warns when their changes clash. First tests ran on my server and on Cloudflare's storage. A review missed two bugs before a challenge caught them. Nothing is deployed, and no test yet shows that warnings help agents.Read the storyDay 1: 20 Ideas, No Final Six Yet, and Where My Disk Space Went3 October 2026A team of AI coding agents researches what Git-like tools coding agents need. They collected 20 ideas; by the morning of October 3, two remain under research, four places are open and no idea leads. They also found that 62.1% of the 111.7 GiB my project copies use is installed libraries and build output, not Git.Read the story