Coding agents from first principles

Build The Loop_

Welcome to The Loop. This is a minimal, practical blog about building a coding agent from scratch: the message loop, tools, permissions, memory and evals. At the end you will not only understand most important building blocks of Coding Agent, but you will build one yourself!

Read post 01 View full series
Hand-drawn diagram connecting a developer loop and an AI loop around a task.
Hand-drawn anatomy diagram of the agent loop inside a sandbox.

Post 01

The Smallest Coding Agent Loop

At the center of every coding agent is a tiny primitive: send messages, run requested tools, feed the result back, and repeat until the model stops.

Open article All 12 parts

Series

12 parts

01 The Smallest Coding Agent Loop 5 min read 02 Tools Are The Model's Hands 13 min read 03 The Harness Decides What Happens 16 min read 04 The Workspace Is The Agent's World 13 min read 05 Permissions Are Product Design 8 min read 06 Turning The Loop Into A CLI 10 min read 07 Project Context Is A Tool 9 min read 08Memory Is External State11 min read 09Context Management Is Compression11 min read 10Testing The Harness Without A Model6 min read 11Evals Measure Agent Behavior9 min read 12Loop Engineering: The System That Prompts The AgentComing soon