This matches my recent experiences. I can ask for things, and the output works, but it's all wrong in the details. Not in a "doesn't fit my preferences" way, but in a "makes the system vastly more complicated and leaks that out to the user" way.
I gave Opus 5.5, considered the best model these days, a medium-sized task. Even after a full adversarial review and simplify loop to convergence, it produced code that worked but was garbage, with limited use of newtypes and representable invalid states everywhere
0 reposts12 likes