Field notes, written as I ship. Newest first.
39 posts
A field note on Claude Code commands: which custom slash commands actually earned their place, which I deleted, and the 2026 question the listicles skip, when a command should have been a skill.
I ran Claude Code hooks daily for a month. Four earned their place, two got deleted, and one quietly made my workflow worse. The honest field log.
A month of running Cursor and Claude Code daily on the same Next.js project. What each one is actually for, where they bit me, the cost difference, and why I kept both instead of picking a winner.
An AGENTS.md file helps AI coding agents, but only when it is short. A two-week field log on trimming mine from 412 lines to 38, and why less context beat more.
Running three to five AI coding agents at once with git worktrees for two weeks: the exact setup, honest merge and cost numbers, the failure mode nobody warns you about, and when parallel agents are the wrong call.
Reviewing AI-generated code is the new bottleneck. Here is the 15-minute triage system I run on every agent pull request, and when I just rewrite instead.
One weekend, one invoice tracker, three AI app builders (Lovable, Bolt.new, Totalum). Honest field log: where each one shone, where each one stalled, and which one survived to Monday morning.
A short, honest June 2026 retro on what got cut from a working Claude Code config: the custom statusline, the over-engineered MCP wrapper, the autosave permission. Plus the one thing I almost killed and why I kept it.
A 30-day field log of running an MCP server in production: four failure modes the tutorials skip, honest latency and cost numbers, and when an MCP server is the wrong tool.
Seven Claude Code subagents written in a week. Four still run daily, three are in the trash. The YAML, the tool whitelists, the cost-per-invocation numbers, the three anti-patterns I won't repeat, and the honest carve-outs where subagents are the wrong tool.
Seven days of running Cursor background agents on a real Next.js side project. 23 tasks, 14 PRs, 9 merged, $11.40 spent. What shipped, what failed, the task spec that doubled the merge rate, and when to reach for an AI app builder instead.
After a year of AI pair programming, I optimize my editor for reviewing, not writing. I format only modified ranges to keep diffs honest, switch to inline diffs, bind word-by-word accept and next-hunk navigation to fast keys, keep a short instructions file in version control, auto-approve only safe local commands, and commit constantly. The settings all follow one shift: I read far more code than I write now.
