LLM-Assisted Debugging Rollercoaster

Forget Vibe Coding, Bots Are for Bug Squashing

For the last couple of weeks I have been experiencing software development synchronicities of the most horrible kind.

Normally what happens is bug gets spotted, bug gets investigated, bug gets squashed.

Unfortunately there is a flavour of software bug that defies investigation because once you observe it the darned thing hides.

That is frustrating enough when it happens during a project but it has happened during three projects simultaneously which can only mean the fault is the observer (me).

I am failing to find the lesson I am being nudged by the universe to see, but my (overly) friendly robot programming helper has come in very handy.

Up until now I have mostly done my LLM toe-dipping using the chat or IDE interfaces, but Cursor introduced these cloud Agent thingies. That is cool, but the hidden benefit is they can, if you wish, have full virtual environments including web browser and stuff.

This means you don't have to resort to sending screenshots, pleading "See? See? It doesn't work!" but can say "Check it out for yourself if you don't believe me stupid argumentative toaster!".

Another nice thing about them being "Cloud" agents, aka running on Cursor's servers, is you can set it going and walk away to do other stuff. Pretty good for things like "Find anywhere XYZ is used and refactor". It can even ping you on Slack when it is done!

I was pretty annoyed a tool decided it was a co-author of my project on Github but you know what? I think the clanker actually deserves this one.