I don't see a binary answer to my question: Agents have proven that they are quite capable in various programming tasks such as reviewing code, typical bug fixes, refactoring, testing, and front-end development. For other tasks that require higher order thinking such as building abstractions, the delivery via agents can be quite effortful. But then again, precision is what we will continue to need no matter how capable agents will be in delivering the writing code part.
It would be interesting to see your own answer to the question you posed in the first part:
“why can’t we delegate to AI the thinking effort that goes into writing the actual programs?”
I don't see a binary answer to my question: Agents have proven that they are quite capable in various programming tasks such as reviewing code, typical bug fixes, refactoring, testing, and front-end development. For other tasks that require higher order thinking such as building abstractions, the delivery via agents can be quite effortful. But then again, precision is what we will continue to need no matter how capable agents will be in delivering the writing code part.