Live LabsBuildersCoding agents3 hours

Build a coding agent from scratch: the loop, the tools, the sandbox

Coding agents look like magic until you write one. We build the loop, the file and shell tools, and the sandbox they run in, so you know what Claude Code and Codex do under the hood.

You leave with

A coding agent that reads a repo, edits files and runs tests inside a sandbox, with every step logged.

How the time is spent

  1. I do45 minParam builds it live

    I build the agent loop with two tools and let it fix a failing test, then show the transcript step by step.

  2. We do72 minTogether, step by step

    We move the shell tool into a sandbox and add a permission check before any write.

  3. You do63 minOn your own code

    You point the agent at a bug in your own repo and decide which tools it may use.

What you will be able to do

Write the agent loop

Know what happens between the prompt and the edit.

Sandbox the tools

Run commands where a mistake cannot reach your machine.

Set permissions

Choose what the agent may read, write and run.

Starts from

Before you come

For
Engineers who use coding agents every day and want to build their own.
You need
Python. An LLM API key.
Stack
Python, Claude Agent SDK, OpenAI Agents SDK, E2B
Track
Hands-on agentic AI for engineers who ship.

More builders labs