article --summary
The summary
The piece looks at a worry about AI coding assistants: if agents write both the production code and the tests, people can lose confidence that the tests will catch functional mistakes. Vaccari's reasoning starts from the idea that, once AI writes the code, developers effectively become its customers.
From that view, customer tests rather than developer tests become the primary agreement between people and AI. Keeping that agreement as test data instead of test code, he suggests, makes it harder for agents to alter it without notice. He also notes that custom parsers and test runners were once too costly to build and maintain, but AI has made them cheap enough to be practical.
Editorial note: This is a summary of an article first published by Martin Fowler. The full piece is on their site.