Vibe coding would be having the LLM generate most of the code for him.

In this case, he's not doing that. This is mostly him using the LLM in ways to write up reports on xfwm4, scan for issues, and help with pseudo-code.

It’s easily saved me months of time, not in writing code (for non-trivial things, I can usually write it just as fast or faster), but in tracking down problems and helping me figure out solutions, whether that’s in Claude reading and reasoning about my code to find something that’s wrong, or in adding tons of targeted debugging prints to gather the information we need, and then sifting through the log file for me to do analysis.