More small business owners are being asked to set up an AI agent: a receptionist that answers calls, an assistant that books jobs, a workflow that chases quotes. Most of them have never written a prompt. Give them an empty box labelled "Instructions" and they'll write two lines, or nothing, then wonder why the agent gets things wrong.
I've spent the past year designing a conversational AI for trades and service businesses. Most of the hard problems had little to do with the AI itself. They were about getting what's in the owner's head into a shape the agent can use: which services run on which days, how pricing changes with travel time, which details are worth confirming and which aren't.
Questions in plain language that build the instructions for the owner, instead of a blank prompt. Examples of good answers from businesses like theirs. Safe defaults, so an agent with half its setup done still behaves sensibly. A preview before it goes live, so the owner can hear or read what it will actually say. And a review step, where a person checks what the agent did and corrects it, which feeds back into the setup.
The other half is deciding what the agent confirms. Confirm too little and it books the wrong job. Confirm the same details three times on one call and the caller has had enough.
I map the setup journey from sign-up to the agent's first real conversation, find where people stall or guess, and redesign it around questions they can actually answer. Then we watch real owners go through it and fix what trips them up. If you're building an agent product for people who aren't technical, get in touch.