Questions
AI questions before you buy a clerk.
What is worth automating, what must stay human, and how to put an agent in production without unsupervised send.
- How should a small business use AI?
- Treat it as a clerk, not a strategy. Draft, tag, extract, summarise — with a human on money and sends. Start with one stuck job, not a model. If you cannot say what got faster after two weeks, you bought a demo.
- How much does it cost to hire an agency to build an AI agent?
- An agent that does a real job — intake, desk, dispatch — is scoped like software, usually in the Invent range ($10–25k) plus Pod if it has to live. A chatbot on a dead FAQ is cheaper and usually worse. Price follows the job, the lock, and what you own.
- What is agentic AI with a human lock?
- An agent that does a job in software you own, with a person required before money moves or a message sends. No unsupervised refunds, bookings the calendar does not have, or medical/legal answers. The lock is the product. The model is the clerk.
- When is AI ready for production?
- When the source of truth is yours, a person can stop it, and you can measure reopen rate — not “conversations closed.” If the site cannot take a booking and the FAQ is empty, fix that first. A model on top of a mess is a faster mess.
- When is AI the wrong tool?
- When being slightly wrong costs money, a license, a body, or a customer you cannot get back. Unsupervised refunds, diagnosis, legal answers, and invented hours are incidents. Draft, tag, extract — with a human on the send — can be right.
- Should we add AI or fix the stuck job first?
- Fix the job. Publish the answers you already type. Put the workflow in software you can keep. Then, if the hours are still real, put a clerk on that job with a kill switch. “AI transformation” without a named job is a subscription.
Next
Send the stuck job in one sentence.
We reply within one business day — including no.