Is AI making agile development less "costly"?

As the cost of code generation goes down because AI can do it faster and less expensively than a human, the cost to benefit ratio of adopting agile/xp practices favors adopting them.

We know that AI doesn't generate the best code sometimes but using tests/types as a specification mechanism reduces the risk that the generated code doesn't meet the requirements.

Being agile is all about fast feedback loops so why slow down the process by forcing a human to validate the generated code afterwards to assure it meets the spec?

Don't like the code it generated? Write another test/type spec to guide the AI in the direction you want it to go.

Code is too slow? Write a test to specify the behavior and let AI fix it.

Found a security vulnerability? Write a test to reproduce it and let AI fix it.

What is stopping your teams from using this flow to more effectively use AI to generate value generating code?