Claude Code for Intermediates
You have Claude Code installed and you have run it a few times. It answers, it edits, it helps. But it still feels like a smart autocomplete, not an agent that knows your work. This is the workshop that flips that.
Not a talk. Not a demo you watch. You sit down with your own project and leave with Claude Code configured around it: it knows your codebase, it runs commands you wrote, it reaches your tools, and it can hand work to a second agent while you direct.
This is the second rung of the ladder that runs Vibe Coding Nights and the Claude Code Clinic. Small group, lots of help, everyone working on their real thing.
What you walk out with
A CLAUDE.md that makes Claude Code understand your actual project, not a toy.
At least one command you wrote for a task you repeat, so it is one keystroke from now on.
One MCP server wired in, so the agent can reach a real tool you already use.
One multi-step workflow running on your own work: a subagent fan out, or a command that chains several steps.
The mental model for when to use one agent and when to use several, and how to keep an agent honest so it does not drift or fake its work.
Your seat includes a 1 week AI key
Same as our Beginners room: every ticket comes with a key that powers Claude Code for 7 days, on us. It runs on GLM models through Z.ai, no subscription needed. If you want Anthropic's Claude models, that is the 2 minute switch we show at the end.
How the evening runs (5:00 to 9:30 PM)
5:00 PM. Doors open. Setup check, and everyone gets their key.
5:30 PM. Make it yours: CLAUDE.md, settings, the agent that knows your repo.
6:00 PM. Commands and skills: turn a thing you repeat into one keystroke.
6:45 PM. Reach out: wire one MCP server so the agent touches a real tool.
7:30 PM. More than one agent: subagents, when to fan out, who is the conductor.
8:00 to 9:00 PM. Your build. Pick a real workflow from your work and ship it with helpers.
9:00 to 9:30 PM. Keeping agents honest, where to go next, and time to hang.
Who this is for
People who already use Claude Code, even a little, and want it to do real multi-step work on their own projects: builders, technical PMs, indie hackers, designers who code, operators. You should be able to open a terminal and get a response already. If you have never used a terminal, start with Claude Code for Beginners. If you already run your own multi-agent systems, come to the Clinic with a hard problem instead.
What to bring
A laptop you can install software on, with Claude Code already installed if you can. Mac, Windows, Chromebook, or Linux.
A real project or repo you want the agent to work on. The more real, the better.
One task you do over and over that you wish were one keystroke.
Location
Frontier Tower Floor 10, 995 Market St, San Francisco.
Capacity
Twelve seats. Small on purpose, so everyone gets real help on their own project.