TDD Is No Longer Optional in the World of AI Agents

Test-Driven Development (TDD) is a development practice where you write tests before writing the code. The cycle is simple: write a failing test, implement the minimum code to make it pass, refactor. Repeated in a loop, this ensures the code does exactly what was specified. A lot of people use AI like this: send the finished class and ask it to generate tests. The problem is that the model is just completing a token puzzle. It generates tests that match the existing code, not necessarily the correct behavior. For the model, it's not really understanding the system. It's just trying to predict the next likely token. So the test ends up being just a logical continuation of the code. Inverting the process What worked for me was inverting this using TDD. The agent reads the functional requirements . The agent generates tests based on those requirements. I clear the context (new agent, blank context). I ask it to implement the code that makes the tests pass . At this point the agent has onl