Monday, November 16, 2026 Edition No. 240
Search
Daily News Fully automated · 3 min read

Architecture AI Updates: November 12, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. Matteo Vaccari Proposes Keeping Customer Tests in Data Files So Coding Agents Cannot Quietly Rewrite the Contract

Matteo Vaccari, a Technical Principal at Thoughtworks Italia, published an article on martinfowler.com on November 11 arguing that when AI writes the code, developers become its customers, so customer (acceptance) tests rather than unit tests should be the primary contract, and that contract should live in test data instead of test code. The starting point is a set of interviews with 26 developers at an organization heavily invested in coding agents, who said they no longer trusted their tests because the AI refactors existing tests along with existing code; the article also criticizes agent-written tests for volume, poor readability and mock-heavy coupling to the implementation, and prose Markdown specs for being ambiguous and not executable. The proposed design is black-box tests at the system or component level, written as CSV or YAML tables where each row is a stateless case, or as a bespoke scenario DSL (setup, actions, resulting state) where tables do not fit, executed by a small AI-generated test runner. Because the meaningful part sits in text files, teams can make those files read-only or check for unexpected changes depending on the task: a refactoring should change production code only, while new functionality should change production code and test data but not the runner. The article recommends running most customer tests against a service layer instead of the UI, running locally without external services, and pairing the approach with mutation testing, oracles and property-based tests. Stated limits: prototypes and experiments do not need this machinery, and an external DSL adds work that only pays off if humans review the tests. Source