Saw a guy coding today. No cursor. No windsurf. Just typing. Like a psychopath

In a field where the programmers don't know that floating point has different dynamic range than integers, or the difference between a blocking and non-blocking call, where there's no limit to memory allocation, and where everything —including the back end— runs in a library, inside a framework, running through an interpreter, inside a virtual machine, inside an abstraction layer, inside a web browser, inside of the operating system, and where metrics put release cadences over everything else, you end up with an environment where the speed, reliability, and readability of LLM generated code is on par with the average human. It's not a high bar: https://xkcd.com/2030/.

If you're writing something that requires stability, efficiency, and capability, you're not going to benefit from the output of an LLM, but if you're constantly "refactoring", by shoving a new web interface into the codebase of the old one, which itself was created in the same manor, several layers down, all so that you can release a new user "experience" every year or two, each running slower and with fewer features than the last, than pulling in code from an LLM is practically a requirement, and trying to make something that works well is a waste of time, because there's no reliable foundation for it to run on.