I usually try to be polite to everyone I meet, which probably is a good idea in general. However, as frontier models advanced and were capable of doing work across multiple surfaces for multiple minutes, or even hours - they started feeling like… entities. Especially if they have persistence - CLAUDE.md, AGENTS.md, dedicated memory features like in consumer web interfaces. It also helps (or doesn’t help?) that they’re sometimes smarter than me now, especially on topics that I have thinner knowledge on (i.e. Claude Code making decisions on behalf of me when managing servers, or me asking for its “opinion”).

This led to me talking conversationally to them more than like “talking” to a tool - which, in itself, is a weird concept. Usually tools are deterministic. LLMs aren’t, either by design or due to various processing noise, and even then we don’t exactly know what’ll happen before we give them a specific prompt. They also talk back, like a human… somewhat. I can tell a text “looks like” Claude, Gemini or ChatGPT - but also, that could very well pass as a human. A sometimes overly confident, too kind human, but still a thing that sounds human.

I’m not sure if this is a good or a bad thing, per se. It’s likely to become more common as AI gets ever more capable. I’m probably going to keep on doing this. This entire blog shows me doing it routinely. Let’s see what path this takes the next… half year, if I’m honest. Things have been changing too quickly for it to really feel real!