I think what’s really going to happen is that programming will focus on writing contracts that AI agents have to fulfill. You’d specify the behavior you want, memory constraints, runtime complexity, etc. And then the agent has to figure out how to produce code that meets the specification.
I’m thinking something more along the lines of Haskell or Idris type definitions. You can use that to specify the function signature, and that becomes the contract that the function has to fulfill.
I think what’s really going to happen is that programming will focus on writing contracts that AI agents have to fulfill. You’d specify the behavior you want, memory constraints, runtime complexity, etc. And then the agent has to figure out how to produce code that meets the specification.
Isn’t this COBOL or 4GL or something?
I’m thinking something more along the lines of Haskell or Idris type definitions. You can use that to specify the function signature, and that becomes the contract that the function has to fulfill.