You already know how to build production AI agents.

The .claude/ folder you edit every day IS the agent.

You just didn't know it counted as production.

Last week I ran a spike. Same claude.md, same skills, same MCPs, same hooks I edit every day in Claude Code. Pointed at a Next.js app.

Shipped a working Knowledge Chatbot in an afternoon.

No new framework. No managed platform. Zero files I hadn't written a dozen times before.

Everyone's covering agents that help you build. Almost nobody's covering this: The same primitives ARE the production runtime.

Five things. All markdown or JSON. All diff cleanly in a PR:

• claude.md — the agent’s identity in markdown. Tone, rules, constraints.

• Skills (.claude/skills/) — reusable behaviors, also markdown. Same file format in dev and prod. Same loader.

• MCPs (.claude/mcp.json) — servers that extend the agent’s tool surface.

• Hooks — PreToolUse blocks path traversal. PostToolUse logs every file the agent read.

• Sub-agents (.claude/agents/). Each directory is a self-contained agent. New agent = new folder.

The SDK is one line:

npm install @anthropic-ai/claude-agent-sdk

There is no translation layer between dev and prod. The claude.md that guides Claude Code in your terminal is the exact same claude.md running in production.

For PMs, this is the part that matters: changing the agent's behavior is editing a markdown file. You iterate on AI at the speed you iterate on a Notion doc.

—

Full walkthrough (the cloneable demo, the three TypeScript casts the SDK types don't warn you about, and the postbuild trick that stops prod from inheriting your dev claude.md): productcompass.pm/p/cla…