I had an issue with my internet provider today and the Hotline didn't really work. The real person chat was ok and solved my problem, but after using Claude code all day, I think a good AI agent with proper rights and scaffolding, would have probably been a nicer experience.

I just read www.anthropic.com/engineering/..., which is a long and well written blog post about AI agent evaluation. A lot of stuff is not really new, but it's nice to see it written down like this with some personal experience. Kind of motivates me to experiment with writing my own evals.

I've been mostly observing bluesky in the last months and I feel like it's getting to a stage with more and more interesting and cool discussions and posts for me. I might start posting and commenting more, let's see.