Writing

How I work with agents. What works and what doesn’t.

• Der Agent ist nur ein Ordner (German)

  No framework, no swarm. A folder with context, skills and an AGENTS.md. That is the agent. Here is what it looks like in practice.

  14 April 2026

• Dein Wissen gehört dir. Du kannst es nur nicht benutzen. (German)

  Everyone has bought courses, listened to podcasts, scattered notes everywhere. The knowledge is there, but you can’t get at it. How I pulled everything out of Notion, Obsidian and Heptabase and made it available to my agent.

  9 April 2026

• 5,700 Messages in 16 Days

  I built a skill that analyzes my AI agent sessions. 5,700 messages, 65 projects, 111 context switches per day. Here are the numbers.

  16 February 2026

• A Full Stack Migration in One Day

  PostgreSQL to SQLite, React to Hotwire, 8 gems removed. An autonomous AI agent ran 37 iterations and migrated a production Rails app in one day.

  8 February 2026

• My AI Coding Workflow, January 2026

  How I use Claude Code with custom skills, voice input, and autonomous agents to ship 5 client projects and a dozen side projects per month.

  1 February 2026

• What Levelsio Wants, I Already Have

  Levelsio tried to get vanilla Claude to automate startup creation. It refused. I already have the stack: domains, servers, DNS, deployment, all via chat.

  25 January 2026

• Just Build It?

  Why I build custom AI agent skills instead of waiting for features. Session analysis, handoffs, model-specific tools, all from existing data.

  16 January 2026

• Agent-Native CLI Design

  Design patterns for CLIs that AI agents actually use well. Structured output, composable atoms, accumulated state, and minimal round-trips.

  13 January 2026

• Agent-Native Skill Feedback

  A PostToolUse hook that makes Claude document friction every time it uses a skill. The agent becomes its own QA tester.

  12 January 2026