TDD inside AI agents does not improve code quality

Research by Birgitta Böckeler on MartinFowler.com showed: enforcing TDD in AI agents does not provide code quality benefits, but consumes 3–8.5 times more tokens.

Compiled by Sergey KostenchukPublished 2026-08-14Updated 2026-09-18

2026-08-14 Coding Anthropic

💻 TDD inside AI agents does not improve code quality

Birgitta Böckeler (Thoughtworks) on MartinFowler.com: research with Claude Sonnet 4.6 and Opus 4.8 showed — enforced TDD in AI agents does not provide measurable benefits, solutions without TDD consistently lead, and 3–8.5 times more tokens are consumed.

🌍 The research questions TDD as a QA methodology in agentic development.

👤 Abandoning the ritual will save tokens. Caveat: tests only on Anthropic.

Source: https://martinfowler.com/articles/exploring-gen-ai/tdd-in-the-agent-loop.html

Sources