Dealing with undocumented legacy systems without access to the original developers is also part of the job, though. I'd estimate that at least 80% of the code I've had to maintain in my career was committed by someone no longer at the company. And while I wouldn't let an LLM loose to one-shot comprehensive documentation on a complex legacy system, I find them incredibly helpful at wayfinding in an old codebase.
It's trivial, verifiable and low-risk to have an LLM drop log statements at each step of a legacy endpoint and run it, verify that the logging statements are in order in the trace, and then read through the source, step by step. I prefer this kind of spelunking to generating a bunch of static docs that may or may not lead me astray in the app.
CodingJeebus•1h ago
It's trivial, verifiable and low-risk to have an LLM drop log statements at each step of a legacy endpoint and run it, verify that the logging statements are in order in the trace, and then read through the source, step by step. I prefer this kind of spelunking to generating a bunch of static docs that may or may not lead me astray in the app.