Evaluation-Driven Development: Why QA Is About to Own the AI Stack

Evaluation-Driven Development (EDD) is emerging as a critical discipline in AI engineering, as frontier models become commoditized and the challenge shifts to verifying agent reliability. QA professionals, with their testing expertise, are poised to lead this shift, replacing traditional deterministic testing with evaluation-based methods suited to non-deterministic LLM outputs.

Member-only story Evaluation-Driven Development: Why QA Is About to Own the AI Stack Prompt engineering was the hype cycle. Eval engineering is the job Models are commodities now. Every lab ships a frontier model within weeks of the last one. Claude, GPT, Gemini, Kimi, Qwen — pick your poison, they’re all good enough for most tasks. What isn’t commodity: knowing whether your agent actually works. That’s the whole thesis of Evaluation-Driven Development EDD — and it’s quietly becoming the most important discipline in AI engineering. If you spent 15 years in QA like I have, this is the moment the industry finally caught up to what we’ve been saying since forever: you can’t ship what you haven’t tested, and “it looked right in the demo” is not a test. TDD is dead. Long live EDD. Test-Driven Development gave software engineering a contract: write the test, watch it fail, write the code, watch it pass. Deterministic. Binary. Green or red. LLM agents break that contract completely. Ask the same agent the same question twice and you can get two different — both defensible — answers. There’s no assert equals. There’s only was this good enough, on this axis, for this use case.