So instead of paying a human that "commands an engineer salary" then they'll be forced to pay whatever Anthropic or OpenAI commands to use their LLMs? I don't see how that's a better proposition: the LLM generates a huge volume of code that the product team (or whoever) cannot maintain themselves. Therefore, they're locked-in and need to hope the LLM can solve whatever issues they have, and if it can't, hope that whatever mess it generated can be fixed by an actual engineer without costing too much money.

Also, code is only a small piece and you still need to handle your hosting environment, permissions, deployment pipelines, etc. which LLMs / agentic workflows will never be able to handle IMO. Security would be a nightmare with teams putting all their faith into the LLM and not being able to audit anything themselves.

I don't doubt that some businesses will try this, but on paper it sounds like a money pit and you'd be better off just hiring a person.