Let's build our own coding agent
Time and room for this session are published in October. when the timetable goes live.
Coding agents can feel like magic: point one at a codebase, ask it to fix a bug, and it reads files, edits code, runs commands, and iterates until the task is done. But under the hood, the core ideas are surprisingly approachable.
In this talk, we’ll build up a small coding agent from scratch to understand how tools like Claude Code, Cursor, and Codex work. We’ll look at the agent harness: the application layer that talks to the model, manages context, executes tool calls, and keeps the work moving.
Starting with a simple conversation loop, we’ll add file and command access, use a browser to verify the agent’s work, and explore what happens when its context fills up. We’ll then look at subagents, project instructions, MCP, and reusable skills. Along the way, I’ll connect these building blocks to engineering examples from Lovable, including how feedback from the agent can help us improve its harness.
You’ll leave with a working mental model of coding agents and concrete patterns you can use to build your own, customize the tools you already use, and debug them when they go wrong.
