More musing about the future of software engineering

I’ve been spending more time writing code with Claude on one of my github projects. I didn’t have access to something like this in my most recent role so I am coming in a little late on this.

It would be so easy to just vibe away and give it the reins, fully trusting it with what I want to do. As I’m trying to do an analysis of housing development in my city though, I’d really prefer to be closely familiar with what I’ve ‘written’ because the point is to be able to confidently answer questions about the result to be persuasive. The agent acts like an overly enthusiastic intern or junior engineer and I don’t think it would be responsible to just trust it completely, mostly because I am not sure how complete my own instructions are! This isn’t something that I could ever see improved by UX or data or model training.

I don’t think AI could possibly replace humans fully. Maybe it’s my workflow or Claude settings, but I’m surprised how long some of the replies take to come back sometimes.

Is it really faster or am I more productive? I’m still not sure.