Agentic Engineering Patterns
by Simon Willison
Simon Willison's free, growing guide to getting reliable, reviewable code out of coding agents like Claude Code and OpenAI Codex.
Overview
Agentic Engineering Patterns is a free, chapter-based guide on simonwillison.net by Simon Willison, co-creator of the Django web framework, creator of Datasette and the LLM CLI tool, and one of the most widely read independent writers on practical LLM use. He began it in February 2026 and presented it on 4 March 2026. The Hacker News launch thread scored 543 points with 310 comments, and the 'Writing code is cheap now' chapter drew 384 points and 499 comments. Willison defines agentic engineering as professional engineers using agents that can both generate and execute code to amplify their existing expertise, as distinct from vibe coding. The guide has six sections. Principles covers what agentic engineering is, why writing code is cheap now, hoarding working examples an agent can reuse, using AI to produce better code, and anti-patterns such as filing unreviewed agent-written PRs. Working with coding agents explains how agents work (tokens, system prompts, the tool-call loop, replayed context, cached input tokens, reasoning), using Git with agents, and subagents. Testing and QA covers red/green TDD, 'first run the tests' and agentic manual testing. Understanding code covers linear walkthroughs built with his Showboat tool and interactive explanations that counter cognitive debt. The last two sections are annotated real prompts from his own projects and an appendix of the prompts he uses day to day. It is tool-agnostic and aimed at Claude Code and Codex users alike.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Guide
- Cost
- Free
- Duration
- ~2-3 hours to read all chapters; ongoing as new chapters are added
- Provider
- Simon Willison
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Distinguish agentic engineering from vibe coding and apply the difference to professional work
- ✓Understand how coding agents work: tokens, system prompts, tool-call loops and replayed context
- ✓Use red/green test-driven development so agents write tests first and verify real fixes
- ✓Start agent sessions by running the existing test suite to ground the agent in the project
- ✓Use Git and subagents effectively to keep agent work reviewable and the main context clean
- ✓Generate linear walkthroughs and interactive explanations to understand code an agent wrote
- ✓Build a personal hoard of working examples that agents can search and recombine
- ✓Avoid anti-patterns such as filing unreviewed agent-generated pull requests on collaborators
Highlights
- •Written by Simon Willison (Django co-creator, Datasette and LLM CLI author), a practitioner who ships code with agents daily
- •Strong practitioner reception: 543 points and 310 comments on the Hacker News launch thread (2026-03-04)
- •Includes annotated real-world prompts and an appendix of the author's own prompts, not just principles
- •Tool-agnostic patterns that apply equally to Claude Code, OpenAI Codex and similar agents
- •Free, short, and still growing: new chapters such as interactive explanations were added after launch
Who It’s For
Best For
- ✓Professional software engineers adopting Claude Code, Codex or similar coding agents
- ✓Tech leads setting team norms for reviewing and testing agent-generated code
- ✓Developers worried about skill atrophy or cognitive debt from agent-written code
- ✓AI engineers who want concrete, battle-tested workflows instead of hype
Prerequisites
- •Working software engineering experience including Git and automated testing
- •Access to a coding agent such as Claude Code or OpenAI Codex to try the patterns
FAQ
What is Agentic Engineering Patterns?
Agentic Engineering Patterns is a free online guide by Simon Willison for professional software engineers who use coding agents such as Claude Code and OpenAI Codex. It collects concrete practices, including red/green TDD, running the tests first, subagents, Git workflows and agent-written code walkthroughs, so you can ship agent-generated code you actually understand and can defend in review.
Is Agentic Engineering Patterns free?
Agentic Engineering Patterns is free to access.
What level is Agentic Engineering Patterns for?
Agentic Engineering Patterns is aimed at a intermediate audience. Recommended background: Working software engineering experience including Git and automated testing, Access to a coding agent such as Claude Code or OpenAI Codex to try the patterns.
How long does Agentic Engineering Patterns take?
Expect roughly ~2-3 hours to read all chapters; ongoing as new chapters are added. Most learners work through it at their own pace.
What will I learn from Agentic Engineering Patterns?
You'll learn: Distinguish agentic engineering from vibe coding and apply the difference to professional work; Understand how coding agents work: tokens, system prompts, tool-call loops and replayed context; Use red/green test-driven development so agents write tests first and verify real fixes; Start agent sessions by running the existing test suite to ground the agent in the project; Use Git and subagents effectively to keep agent work reviewable and the main context clean; Generate linear walkthroughs and interactive explanations to understand code an agent wrote; Build a personal hoard of working examples that agents can search and recombine; Avoid anti-patterns such as filing unreviewed agent-generated pull requests on collaborators.
Topics
Sources
This page was written from 5 sources, 2 on domains other than simonwillison.net.