From the desk
Run it first.
Then write it down.
Field notes on AI agents and developer tools. Tested on real machines, not summarized from press releases.
Contents
- 01Choosing a Coding Agent
- 02Green Tests Prove Nothing
- 03Don't Trust an Agent's "Done"
- 04Run an AI Agent 24/7 on a VPS
- 05MCP Servers 101 and Five Design Rules
- 06What Claude Code Really Costs
- 07Trim Your CLAUDE.md
- 08Claude Code vs Cursor vs Codex in Practice
- 09Claude Code Hooks in Practice
- 10Subagent Failures in Parallel Runs
- 11Permission Design for AI Agents
- 12Pitfalls of Always-On Bot Monitoring
Four house rules
- 01Run it first
We only write about what we have actually run on our own machines or VPSes. No press-release rewrites. - 02Keep the commands
Steps, config files, even the failing logs — enough for you to reproduce it. - 03Date everything
Prices and specs drift. Every claim carries the day we checked it. - 04Show the money
Affiliate links are marked. Tools that pay us nothing still get recommended when they are good.