Articles · Learning path · Beginner

From Claude Code to a multi-AI system

The principles of coding with AI, from one assistant in one project to a system where several AIs from several providers work together. One step per episode, from real work.

8 lessons · about 33 min in total

  1. 1 EP1 How an AI coding agent works, and why your project needs a handbook for itThe series starts here. How Claude Code works in a loop, and the first and most important principle, writing what the AI should know into a CLAUDE.md file. 4 min read
  2. 2 EP2 Skills, memory and hooks: knowledge, procedures and rules, kept apartWhen CLAUDE.md starts to bloat, split it three ways: knowledge in memory, repeatable procedures in skills, must-hold rules in hooks. 5 min read
  3. 3 EP3 Subagents: hand work to helpers with a room of their ownA subagent is a helper the main AI hands work to. It works in its own room and returns only the result. When to use one, and how to limit it. 4 min read
  4. 4 EP4 Workflows: a production line with checkpoints, and no skipped stepsScripts do the fixed steps, AI does the steps that need thinking, and checkpoints sit before anything irreversible, with this site's real line as the example. 4 min read
  5. 5 EP5 Choosing the right model: dozens of times cheaper, no worse workOpus, Sonnet or Haiku: which job suits which, what the effort dial does, a model per subagent, and the hidden costs, worked through at real prices. 4 min read
  6. 6 EP6 MCP: give the AI hands. Write a tool once, use it from many AIsMCP is the standard that connects AI to other systems. A 10-line MCP server, plugging it into Claude Code, and what to watch before an AI touches real systems. 4 min read
  7. 7 EP7 Claude with other AIs: local models and other providers, through one doorNo model is best and cheapest at everything. Claude working with local models and other providers through one thin layer: 15 lines of code and a real 17-agent team. 4 min read
  8. 8 EP8 Let the system pick the tool, and measure whether it picks right (finale)With hundreds of tools, skills and agents, how do you know which suits a job? Building a router: search by meaning, let an LLM rank, and measure with a test set. 4 min read