Things LLMs are great at.

Kagi's code assistant, powered by Claude 3.5 has become quite embedded in my day-to-day Rails development and in my administration of Booko. An area where it really improves my quality of life, is pasting in a bunch of logs or an exception - or both, and having it extract meaning from it. I'm yet to have a bad result with this kind of summarisation, and it really is uncanny to get meaning extracted from big chunks of text which are tedious to parse with your own eyes. I usually include the functions or methods which are related to the logs / exception and get some suggestions for improvement.

The code suggestions are often solid, but not always - they're certainly not just copy / pasteable. Sometimes they're older Ruby / Rails code, which is understandable considering how LLMs are trained, and although it's rarer now, sometimes it's just straight up wrong.

So, after finding bugs and suggesting fixes to code, Code assistants are also useful at generating test cases - which is relatively low risk / high reward automatic software development.

To summarise:

• Great at parsing / summarising logs
• Good at suggesting improvements to exisiting code
• Great at generating test cases

Booko's not so badly written that this particular scenario comes up every day, but it's great to have a tool that helps solve it so well.