Uncle Bob: My current strategy is to not read any of the code by my agents

But what about TDD?

It’s a religion you know.

Even AI can’t stop the believers - you must write a test for every line of code before you write the code.

It may sound ridiculous crazy stupid and bananas to some but that the way TDD works.

For any sufficiently complex system, evaluating the interfaces, the constraints, the guarantees, its test, is the only way you can practically work with them IMHO.

Most developers, if not pretty much everybody, will struggle to keep anything over 100k lines (my made up arbitrary number) of code in their head.

That's actually ok IMO, if you know systems well & know lots and lots of good architectural questions about what it's built.