One week of running a judge/build/human AI task loop over 41 stale tasks: 10 closed without writing code, one unattended cycle, and exactly one decision that reached a human. Field notes, n=1.
Author split 41 backlog tasks among AI loop roles (judge/build/human); 27 closed day one because the bottleneck was judgment, not code. For engineering teams: stagnant backlogs lack a judgment layer to re-validate premises, not developer capacity. Tasks rot without periodic re-eval.
Is there a region at the bottom of your task ledger you haven't scrolled to in weeks?
Mine held 41 tasks across two repositories. I use AI agents every day, and yet the ledger never shrank.
One morning I ran my homegrown "list the tasks that are ready to start" command. The answer was empty.
python3 ~/.claude/scripts/claims.py ready
# (no output)
My coding agent keeps its backlog in a public GitHub repo, restored fresh into every session. A morning's audit found three of…
A real session with Claude Fable 5: a botched audit, a full undo, and the question that turned the...
6 sessions, 150 standardized tasks, 2 rule formats. The mechanical gate won. Everything else was...
Claude Code writes in six or nine parallel sessions. The bottleneck is one person taking in that output and issuing the next…
I keep a running list of small, well-defined tasks I never get to: a missing health-check endpoint, a...
My coding agent ends almost every task the same way. "Done! All tests pass." For months I just...
