What makes one AI assistant feel better to use than another? Often it has nothing to do with benchmark scores.
Sam Henri Gold published a personal essay on Claude's design that resonated with practitioners who work with Claude daily. The piece examines the deliberate choices behind how Claude communicates - not just the visual interface of claude.ai, but the underlying character decisions that shape every conversation.
These aren't accidental. Anthropic has published explicit documentation describing Claude's desired personality traits, values, and communication style. The model is designed to have opinions, push back on requests it considers poorly framed, and explain its reasoning even when that explanation wasn't requested.
That last tendency cuts both ways. Practitioners who want a thinking partner appreciate Claude's habit of surfacing assumptions and volunteering relevant context. Those who want direct answers to direct questions often find it verbose.
Response length is a real design choice with real productivity consequences. Claude's default toward thorough responses means routine tasks - summarizing a document, drafting a quick email, writing a SQL query - often come with more scaffolding than the task requires. Anthropic has been gradually tightening this, and Claude 3.7's responses run noticeably shorter than earlier versions in practice, but the model still leans longer than most competitors by default.
The interesting thing about AI design critique is that the design layer is invisible until it isn't. You stop noticing that a model almost never produces bullet-pointed responses when prose would serve better - until you've spent a week with one that does it constantly. Small choices accumulate into a product personality, and Claude's personality has grown distinct enough that it generates substantive discussion among the people who use it every day. That's not nothing.