Build an Agent Harness workshop

Free, in-person workshops at our Old Street campus. Work through a piece of our ML curriculum with our founders. No application required. Overview A taster for our ML engineering programme. You'll build a working coding agent from scratch in Python: a minimal version of Claude Code, with three tools, a loop, and a termination condition. Background Coding agents are everywhere, and most developers use them without a clear picture of what's going on underneath. The harness versus the model. What sits in the context on any given turn. Why reading a big file on turn three costs tokens on every turn after. When to reach for a hook, and when to reach for a sub-agent. None of this is specific to one tool. It's the same architecture across every coding agent, and the fastest way to get it is to build one. Once you've written the loop and watched your agent fumble its way out of a mistake, you can open any agent and know what you're looking at. Who is this for Experienced software engineers. Comfortable on the command line, able to pick up an unfamiliar API from its docs. No prior agent or ML experience needed. What we will cover What to bring Hot-desking You're welcome to hot desk from 9am on the day. ​​​Access information: Please ring the Founders and Coders bell. The space is wheelchair accessible and has a lift. Our toilets are fully accessible. If you have any specific access needs, please let us know when you sign up and we'll do our best to help.