BA158, six hours from Heathrow - Friday 20/Saturday 21 February, depending on your point of view.

I’ve long been skeptical about (Generative) AI, watching with wry detachment as the tidal wave of hype and BS sloshes through every aspect of modern life.

However.

I am starting to see opinions from people who I respect that suggest it has its place, and can be a win for productivity if used responsibly.

With this in mind, and feeling ever shorter of time to get things done in all of my personal, voluntary and professional lives, I decided it was high time to give it a go.

PC Pro magazine published a very well timed group test of different AI providers, and off the back of that, I decided to give Claude a shot.

As regular readers will know, I run an invoicing system for my church. It’s a Python/Django project with around 3,500 lines of code, so it presents an ideal non-trivial test bed to see what Claude can do.

The project also contains a markdown file with my list of features I’d love to have, big and small. And under normal circumstances, making a dent in a list largely untouched since 2019/20 would be an unlikely activity to while away the jet-lagged hours in a hotel room half way across the world.

Claude Code and I knocked more items off that list in an hour than I’d done in five years before that.

I was particularly pleased with the CLAUDE.md file the AI generated to describe the project and guide its work: it provided a summary of the architecture and key choices which reminded me, the original author of the project, of several things I’d decided and promptly forgotten.

It’s not a silver bullet - more like having a pretty capable junior developer. Sometimes it gets things wrong, and you have to tell it to try again and try harder. It worked really well on this project because it’s a relatively small codebase and I have no difficulty in reviewing changes and confirming if they make any sense.

However, in the space of a jet-lagged hour, it cracked performance problems, implemented features, and generally got stuff done which I could have done by hand, given an entire weekend to devote to the cause.

The manner in which it can use tools is especially nifty, e.g. when asking it to make changes to the appearance of a web page, it can fire up a headless browser and “look” at the results of its CSS changes.

Colour me impressed. More to come on this as I push it further.