The cognitive surrender study from UPenn highlights the risks of agents producing all of the code - eventually you give up verifying the result. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646

There’s going to be a bottleneck on what is verified because over time we will realize how much tail risk we are creating by simply surrendering our own agency to the agents - https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6298838