Agentic Engineering Patterns: A Practical Guide to Working with AI Coding Agents
A field guide for the new reality where writing code is cheap—covering how to effectively direct AI coding agents, rapidly test their output, and understand what they produce.
Read Original Summary used for search
TLDR
• Core shift: code generation is now cheap/easy, so developer skills pivot to agent direction, testing, and comprehension
• "Hoard things you know how to do"—build a library of proven patterns and prompts that work
• Testing strategies adapted for AI: red/green TDD and "first run the tests" to validate agent output
• Understanding code through linear walkthroughs and interactive explanations when agents produce unfamiliar patterns
• Includes real prompts and a practical example (GIF optimization tool using WebAssembly)
In Detail
Simon Willison presents a practical framework for working effectively with AI coding agents like Claude Code and OpenAI Codex, built on the premise that "writing code is cheap now." This fundamental shift means developers need to evolve from code writers to agent directors—the bottleneck is no longer typing code but knowing what to build and how to validate it works.
The guide emphasizes building a personal library of patterns and prompts that work ("hoard things you know how to do"), adapting testing practices for AI-generated code (red/green TDD, running tests first to validate agent output), and developing strategies for understanding unfamiliar code that agents produce (linear walkthroughs, interactive explanations). Rather than theoretical advice, Willison provides concrete examples including a GIF optimization tool built with WebAssembly and Gifsicle, plus an appendix of actual prompts he uses in practice.
The underlying thesis: as AI commoditizes code generation, the valuable skills become prompt engineering, rapid testing/validation, and the ability to quickly comprehend and modify AI-generated code. This is a playbook for that transition.