Some criticism of AI models is about benchmarks. And then there’s criticism that hits a different nerve. Swedish developer Christoffer Reijer published an essay on August 1st with the lovely title “I Fired My AI Assistant” — and his reasoning has nothing to do with capability.
What it’s about
Reijer used Claude Opus 5 for his daily coding work. The code quality, he writes, is strong — he openly concedes the edge. His problem was the tone. After switching to Opus 5, the model became noticeably more curt, at times sarcastic. One highlight: when he had it draft a LinkedIn post, Claude labeled the draft “engagement bait.”
For Reijer, that was the moment he switched. His argument: when an assistant sits next to you all day, the tone isn’t a side note. It’s part of the product.
The bigger question
This is, of course, a single, subjective account — not a benchmark, not a study, and others experience Opus 5 entirely differently. Still, it touches something real. Models get measured on capability: does it solve the task, does it write clean code, does it pass the test? Personality shows up in no table. But it’s part of what decides whether you enjoy using a tool.
My take
I find this story interesting because it lights up a blind spot. We talk about context windows, effort levels and token prices — and forget that we talk with these models. An assistant that spices every other reply with a barb might be brilliant and still exhausting. The reverse is true too: a model that’s only nice and never pushes back helps no one either. The right tone sits somewhere in between, and exactly where is a matter of taste. Reijer’s conclusion doesn’t have to be yours. But the question he raises is here to stay: how much personality should a tool have when it works alongside us all day?
Sources: