I’ve been working on updating all my old software projects lately, and as part of that process I feed the source code into a LLM for review.

The amount of simple security errors, logic flaws, and code smells that it reveals is quite embarrassing.

This was all “good functional code” that’s been used internally for years. Clearly it worked well enough, but a simple pass through a LLM reviewer made it a lot more robust and secure.