TDD Inside the AI Agent Loop: Exploratory Test Shows No Clear Value

rseroter · x · 2026-08-13

A Distinguished Engineer from Thoughtworks explores integrating TDD (test-driven development) into AI-assisted coding, focusing on the most common pattern: running TDD fully inside the agentic loop (where the AI writes failing tests first, then implements the code).

The author conducted an exploratory evaluation to see if this workflow actually improves outcomes. The TL;DR: based on Claude Opus's judgment of the results, there was no clearly discernable difference in quality between using the TDD workflow and not using it. In fact, Opus frequently ranked non-TDD solutions slightly higher in design. This suggests that processes beneficial for humans might be irrelevant or even counterproductive for coding agents.

Related event: Thoughtworks Experts Question Effectiveness of AI-Driven TDD(2 posts)→

Original post →

More from coding & agent

coding & agent channel →