My favourite use-case for Claude so far is the “search engine on steroids” one. It’s proven particularly good - although not perfect - at answering questions on the wide variety of open source libraries and tools I use every day at work. Of course, it ought to be good at that since it will doubtless have consumed the entire source code and documentation of these things during its training.

Actually writing code with it is definitely a skill in its own right, and not one learned overnight, but personally I’m finding it a breath of fresh air as it cuts out the bit I was never very patient with - trawling through API documentation and Stack Overflow trying to work out what prior art existed for what I’m trying to do. If it’s quicker to verify the solution than come up with it yourself, then using Generative AI to write code is a sure-fire productivity boost.

My next venture, sure to be written up here, is to go properly “agentic” and try and get it running my part of the village magazine for me. Don’t worry, it won’t be running amok without human verification of the actions it’s proposing to take, but I fancy my chances of cutting four hours a month down to about 15 minutes, and it’s a nice low-stakes test bed for doing more elsewhere later.