Then your “code” is riddled with bugs. Coding by Statistics will only end in tears.

I didn't say I blindly trust generated code, it needs to be read through and covered by tests, but tasks like generate boilerplate code for some lib I never used before and which would take time to locate and read through documentation now are 10x faster.

There is huge utility in LLM outside generating complete working code.

This is just cope. We've rolled out cursor company wide with no noticeable uptick in bugs.

Most "AI" startups aren't building coding tools, and the utility of this tech goes down dramatically in industries that are less legible on the open internet than software is.

> industries that are less legible on the open internet

I think there will be also wave of private LLMs fine-tuned on corporate data, and it will be also good tools.