Simon Willison Makes the Case for Test-Driven Development as the Key to Better AI Coding Agents
Red/green TDD turns out to be a natural fit for keeping AI-generated code honest
Willison's guide, part of his "Agentic Engineering Patterns" series, makes a compelling case that the decades-old practice of TDD becomes even more valuable when the code is being written by an AI agent rather than a human.
The key insight is that tests serve as a verifiable contract. When a coding agent writes tests first and confirms they fail (the "red" phase), then implements code until the tests pass (the "green" phase), there's a concrete, automated check that the code actually does what it claims to do.
"Every good model understands 'red/green TDD' as a shorthand," Willison writes, noting that simply including these three words in a prompt is enough to dramatically improve results from coding agents.
The approach also builds a comprehensive test suite that protects against regressions as projects grow — a critical concern when AI agents may modify existing code without fully understanding its context.
Analysis
Why This Matters
As coding agents become mainstream development tools, engineering practices need to adapt. Willison's argument that TDD isn't just compatible with AI coding but actually more important than ever provides a practical framework for developers using these tools.
Background
Simon Willison is a respected voice in the Python and web development community, known for Django and the Datasette project. His agentic engineering patterns series documents best practices for working effectively with AI coding tools.
Key Perspectives
The TDD approach addresses a real problem: AI agents can produce plausible-looking code that doesn't actually work. Tests provide an objective measure that catches these failures before they reach production.
What to Watch
Whether TDD-first approaches become standard practice in AI-assisted development workflows, and how tooling evolves to support this pattern.