Interesting to read this almost a year later. I’ve been working with the latest models of Claude Opus and OpenAI Codex and I can tell you that they both can work on very complex code now and yes, they do still get things wrong but by and large they get it right more than they get it wrong.
It’s amazing what 9-10 months have done for the quality of the final result.
That being said, it still generally needs a human in the loop unless you don’t care about the final result.
I have to admit I haven't tried Opus, but I haven't found there to be that much improvement. They interact with the IDE's more cleanly, but still a lot of boneheadedness. .e.g https://blog.nappisite.com/p/coding-assistance-case-study
Interesting to read this almost a year later. I’ve been working with the latest models of Claude Opus and OpenAI Codex and I can tell you that they both can work on very complex code now and yes, they do still get things wrong but by and large they get it right more than they get it wrong.
It’s amazing what 9-10 months have done for the quality of the final result.
That being said, it still generally needs a human in the loop unless you don’t care about the final result.
I have to admit I haven't tried Opus, but I haven't found there to be that much improvement. They interact with the IDE's more cleanly, but still a lot of boneheadedness. .e.g https://blog.nappisite.com/p/coding-assistance-case-study
Great article!