I liked programming a lot, but with AI and how my company is basically requiring us to use it for AI-first development I’m not enjoying my job anymore…
It feels like forced effort on my part to write design documentation and refine it so the LLM does it the way I want and having to review the prompt continuously opposed to actually solving problems the way I would. Plus things aren’t even supposed to be reviewed until an AI review of the AI code is done… I just feel like I’m supposed to be a rubber stamp on the output. They say that dev/review skills won’t atrophy because you’re always reviewing the code, but I can’t accept that conclusion. I became a better developer by actually problem solving, not reading code to see if it makes sense. It’s all about getting more done and trusting it at a conceptual level now, but not really understanding what’s really going on underneath the AI-gen docs because it’s “so much better than human-generated code”.
I don’t understand or really handle that approach well. I don’t really program outside of work because I want to just hang out with my family after it and it feels like a chore to do my job anymore. I just don’t know how to make myself enjoy that type of workflow they are trying to force onto everyone. Any ideas on how to find something to enjoy with that?


You don’t want it writing the tests. They’ll just pass
Yeah 6 months ago maybe. It can write perfectly good tests now.
I actually find the more common failure mode is it finds something that doesn’t quite work right (like a bug in a third party component) and it just silently works around it rather than telling you so you can get it fixed. My suspicion is they benchmaxxing makes them try to make as much progress as possible so they’ll do lots of workarounds rather than asking you.
Another example is when it’s building Rust code for me, it will use a crate that is pulled in as a transitive dependency even if it actually wanted a different one, because it’s sandboxed and doesn’t want to bother me by asking me to run
cargo fetch. Quite annoying.It doesn’t write vacuous tests though.
How are you measuring how good the test is? How do you know it’s a good test?
Why do you think the tests aren’t doing the same thing?
I skim them.
What you mean like it find a bug and it just writes the test to confirm that the bug exists? In my experience Opus 5.5 will do that but only in an “expect failure” kind of way, to confirm the bug is really there, and it tells you about it.
Opus 5.5 employs much more rigorous development practices than most humans developers. It isn’t lazy so it will do things like confirming that regression tests detect the bug before it is fixed.
It’s a huge step up from earlier models, where I would have agreed with all your concerns. Honestly the downvotes and questions along these lines feel a little like someone seeing a train for the first time and asking “but how does it know which way to go?”. Just try vibe coding some Rust with Opus 5.5. It literally costs £18. You can see for yourself.