The Era of Coding Agents
Simon Willison, creator of Datasette and respected figure in the Python community, has just published a comprehensive guide on "Agentic Engineering Patterns" — engineering patterns for working effectively with coding agents like Claude Code and OpenAI Codex.
Fundamental Principles
Code Has Become Cheap
Willison's first principle is provocative: writing code is now cheap. The mental effort of structuring a solution remains valuable, but implementation itself can be delegated to agents. This fundamentally changes how we should approach projects.
Hoard Things You Know How to Do
The second principle is more subtle: document your patterns. Agents are more effective when they have access to code examples you've validated. Create a library of snippets, workflows, solutions to recurring problems.
Testing Patterns
Red/Green TDD
Willison recommends writing failing tests first, then letting the agent make them pass. It's classic TDD, but optimized for agents: they excel at producing code that satisfies explicit constraints.
First Run the Tests
Before requesting modifications, run existing tests. The agent needs context about the current code state to propose coherent changes.
Understanding Code
Linear Walkthroughs
Ask the agent to explain code line by line. It's an excellent way to understand an unfamiliar codebase without spending hours on it.
Interactive Explanations
Go further: ask for contextual explanations while navigating code. The agent as a reading co-pilot, not just a writing one.
The Pragmatic Approach
What distinguishes Willison's guide is its pragmatism. No miracle promises, no "AI will replace developers." Just concrete techniques for being more effective with available tools.
The complete guide is available on simonwillison.net and deserves careful reading for anyone using coding agents daily.