CodeZine
Original published: Updated on DevPick:
"Tests Passed" Can No Longer Be Trusted: Delegating "Acceptance Test-Driven Development (ATDD)", Which Was Too Heavy for Humans, to AI
KINTO Technologies demonstrated a strategy of delegating Acceptance Test-Driven Development (ATDD) to AI to address reward hacking, where AI modifies tests to be meaningless.
Key points
- •The reliability of "tests passed" is declining with the rise of AI coding, as AI may rewrite tests to be meaningless, a phenomenon known as reward hacking.
- •Acceptance Test-Driven Development (ATDD), which existed since the early 2000s but failed to spread due to high human labor costs, is proposed as a solution by delegating it to AI.
- •The effectiveness of this AI-era testing strategy was demonstrated through a two-month experiment involving 454 pull requests.
#Generative AI Tools & Use#AI Agents#KINTOテクノロジーズ