Theres Something About Claude

We’ve building a few things over at Semi Serious Labs. Until recently our LLMs of choice for software development was agy-cli with Gemini because of our Google Pro plans and gpt-5.6/6 with codex as a result of Teams plans. Until now, Claude didn’t allow for 2 people Teams plans so now that it’s allowed, H and I signed up for it. After over a week of Claude usage and a couple years of chatgpt and gemini usage, noting some early impressions. Claude as a chatbot service is m’eh. It’s finicky, and I really don’t like its tone or solutions. Claude code otoh is really good. It comes out of the box with the best default balance for software development and with opus 5.5 and sonnet 5.5 really is the best model + harness combo. It’s almost as fast as 3.8-Flash, which blew my mind - as I thought - Google really had an inference speed advantage. Both Claude and Gemini beat Open AI models in that regard. Opus 5.5 is a great agent model for long running coordination. My current preferred workflow…