Home

AI - TDD with LLMs

Handwritten notes on TDD with LLMs (transcription below)

Transcription of the handwritten notes

TDD tests on LLM models

When we’re talking about LLM TDD, I find it quite challenging to use their first step as “test first”. The models I’ve been testing have more success using a small implementation as foundation.

That way we have the following:

  • Small initial implementation
  • Red
  • Green
  • Refactor (and back to red)

The small initial implementation could even be a small spike of the whole context.

Nowadays, 20/03/2026, the best optimizations are coming from Claude Code.