I completely disagree. I am completely confident that an Opus 5.5 will code circles around the average developer.
I’ve also had to rewrite AI slop, but I did it with AI. Came out great. Turns out, knowing how to use a tool makes a real difference. It’s like saying knifes are shit at cutting because a lot of dumb fucks keep holding it the wrong way.
Yes, the AI is not great at the task without intense supervision and direction. Hence, it’s not good at the task.
Example I’m currently working on. Unqualified junior dev yelled at AI over and over to resolve a bug without looking at the code. The AI noticed there was a discrepency between “user_name”, “username” and “host_name” fields in the database. The discrepancy applied to two users, resulting from a poorly thought out decision to use “user_name” for a new version of this app. Instead of recommending correcting the two records, it created an additional backwards compatibility layer to pull the entire schema of a table, itemize every possible combination of columns, and write a generic lookup system for every possible combination of columns. This was not even a production system that needed backwards compatibility, it was still in development (6 months of development by this guy, I’m his replacement and finished it in about 6 weeks). And it was not a generic system to deal with that table in general, this was only inside the context of a single auth API method. Without basic intuition, it completely failed to diagnose the problems of its own creation, and just kept endlessly expanding the system to account for its own mistakes that it forgot hours later. The codebase that resulted from this process is absolutely disgusting.
There’s a reason for this. It’s not designed to do symbolic reasoning. It’s designed to passively infer symbolic reasoning from linguistic analysis. It doesn’t know what the hell it’s doing.
I completely disagree. I am completely confident that an Opus 5.5 will code circles around the average developer. I’ve also had to rewrite AI slop, but I did it with AI. Came out great. Turns out, knowing how to use a tool makes a real difference. It’s like saying knifes are shit at cutting because a lot of dumb fucks keep holding it the wrong way.
Yes, the AI is not great at the task without intense supervision and direction. Hence, it’s not good at the task.
Example I’m currently working on. Unqualified junior dev yelled at AI over and over to resolve a bug without looking at the code. The AI noticed there was a discrepency between “user_name”, “username” and “host_name” fields in the database. The discrepancy applied to two users, resulting from a poorly thought out decision to use “user_name” for a new version of this app. Instead of recommending correcting the two records, it created an additional backwards compatibility layer to pull the entire schema of a table, itemize every possible combination of columns, and write a generic lookup system for every possible combination of columns. This was not even a production system that needed backwards compatibility, it was still in development (6 months of development by this guy, I’m his replacement and finished it in about 6 weeks). And it was not a generic system to deal with that table in general, this was only inside the context of a single auth API method. Without basic intuition, it completely failed to diagnose the problems of its own creation, and just kept endlessly expanding the system to account for its own mistakes that it forgot hours later. The codebase that resulted from this process is absolutely disgusting.
There’s a reason for this. It’s not designed to do symbolic reasoning. It’s designed to passively infer symbolic reasoning from linguistic analysis. It doesn’t know what the hell it’s doing.