The Operator's AI Brief
Which AI Model Does A Small Shop Actually Need
Two creator videos from October 8 raise two questions to put to whoever sets up your AI: which model size, and how many rules.
Listen to this brief (3 min)
The audio is AI-produced: an AI clone of Jon's voice and an AI co-host, reading a script built from the sources below.
Two creators posted videos on October 8 that give a small business owner something to act on. One covers which size of AI model to pay for. The other covers how long your assistant’s instruction list should be.
Which model runs your phone line
Nick Saraev posted a video titled “I Spent $700 Testing Haiku 5.5 vs Sonnet 5.5 vs Opus 5.5”. I worked from the title and have not seen his results. The three names are small, medium, and large versions of Claude. Anthropic lists a different per-token API price for each, so the larger model costs more every time it runs.
That matters when a vendor builds your phone line or inbox sorting. Answering “what time do you close” is routine work with the same questions every day. Reading a messy contract is harder, and a larger model may earn its price there. Ask your vendor which model handles your calls, what each call costs, and whether they tested a smaller model on your real calls. A vendor who won’t answer has told you something.
Shorter instructions work better
Nate Herk posted a video titled “Anthropic Engineers Just 10x’d Everyone’s Claude Code”. His description says more instructions do not always mean better results. It covers what to keep in the standing instructions and what to load only when needed. I worked from the description and have not watched the video.
That fits how I build these systems. People keep adding rules until the assistant slows down and follows the important ones less, like a new hire handed a thick binder. Keep the rules that apply on most days short. Put the rest in a reference sheet the assistant opens on demand.
Neither video settles anything yet. Each hands you a question to ask whoever sets up your AI.