● agent-first coding masterclass | 13 hands-on labs

Build software for a user that is partly machine.

This is not a course in a programming language. It is a hands-on class in the habits that make your code something an AI agent can find, call, remember, and trust, not just a person. Every recipe is taken from a system that runs in production. You write it. Your agent uses it.

You leave with: a project an agent can discover, sign in to, call, and remember — and the know-how to run many of them at once.

Start Lab 01 → See the 13 labs

the arc — one reliable call, to a team of agents

01The typed calla result you can trust 02Build your own MCP serveragent-callable in 10 min 03Schema is the producttools a small model can use 04Make it agent-discoverablea front door for agents 05Auth without rolling your owna key that is a boundary 06Memory that survivesfacts on disk, plus a guard 07Memory mixesfour memory types, one stack 08Context engineeringkeep less, the right less 09Meta promptinga prompt that writes prompts 10Agents talking to agentsa handoff you can verify 11Orchestrate a fleetmany workers, one gate 12Reverse a closed APIbuild the way in yourself 13Ship it and prove itdeploy, then prove it live

how the room runs

who it is for

Builders, and their agents. You bring the instinct for a real problem; the labs bring the production patterns. Each lab is a single page with copy-and-paste steps you run on your own laptop. No GPU and no cloud account until you choose to deploy in Lab 13. The three middle labs on memory, context, and prompting run on plain Python with no setup and no keys.

the agent-first idea

Every lab ends with a hand this to your coding agent prompt. Paste it into Claude Code or Cursor, watch your agent build the lab for you, then read the changes to learn what it did. That is the point: you and the machine, building together, the whole way through.

where these labs come from

VCN #34 · Toolsmith VCN #33 · Total Recall ClawCamp · Agents Talking to Agents ClawCamp · Context + Meta Prompting Web A(gent).0

Every lab is a single file, no build step, easy to read the source. Every Python recipe uses plain ASCII text so it runs on Windows. Every pattern is real code from a working system. Take anything you want.
Immersive Commons · Vibe Coding Nights · built agent-first.