If you can gather requirements and lay out a detailed and perfectly accurate flowchart, then congratulations: you’ve just programmed. It’s done. The first difficult part is over. Translating that flowchart into machine code is easy, tons of tools already do that visually or however you want it and LLMs are just an additional tool for this.
Then there’s the second difficult part of a project’s lifecycle: Debugging, maintenance, and support. Where again AI can help punctually as part of the tool box, but most of those tasks don’t require writing (a lot of) code.
All the senior software engineers I know spend, optimistically, 20 % of their time actually “writing code”. That’s your upper limit on the efficiency gains of LLMs for higher level software engineering. Saying LLMs will replace programmers is like saying CAD software will replace architects.
If you can gather requirements and lay out a detailed and perfectly accurate flowchart, then congratulations: you’ve just programmed. It’s done. The first difficult part is over. Translating that flowchart into machine code is easy, tons of tools already do that visually or however you want it and LLMs are just an additional tool for this.
Then there’s the second difficult part of a project’s lifecycle: Debugging, maintenance, and support. Where again AI can help punctually as part of the tool box, but most of those tasks don’t require writing (a lot of) code.
All the senior software engineers I know spend, optimistically, 20 % of their time actually “writing code”. That’s your upper limit on the efficiency gains of LLMs for higher level software engineering. Saying LLMs will replace programmers is like saying CAD software will replace architects.