It's interesting to see people who have been working as software engineers worrying that codegen LLMs will take their job. In the past week the amount of work that consisted of writing well-defined code was pretty much zero, and the actual core logic I have written lately has been almost mindlessly trivial. I spent
much more time discussing requirements, arriving at consensus on where components should live, determining whether change requests should go forward, determining how changes affected latency or disk space, and debugging nuanced issues that required multiple engineers to slowly reveal the problem.
If your programming job consists of simply receiving a greenfield spec and banging out the code for it over and over, LLMs might be an issue. But most jobs are not like this. The bulk of the work, as always, is in dealing with the people at either end of the process. And knowing what the computer is doing and how it works will continue to be invaluable when an LLM writes logically perfect code that happens to run slowly or do the wrong thing once it makes contact with the real world.