AI agents for small businesses

A useful AI agent takes an input, does one defined job, and hands off to a person when it is not confident. A chat window bolted onto your documents is not that; it is a search box that is sometimes wrong. This page explains what a working agent looks like, so you can build one in our class or have us build it.

Learn it, or have it built

In our AI classes, owners work in Claude Code on custom apps for their own businesses, the same skills an agent like this needs. If your job is one repetitive office task, a small AI tool is usually the right size. The rest of this page applies either way, whoever does the building.

What a working agent looks like

Four properties. Miss one and it stays a demo.

  1. A bounded job. “Read the new inquiry email, pull out the name, the service and the address, and flag what is missing.” Not “help with sales.”
  2. A real input and a real output, in software you already use. If you have to open a new tab to get the value, you will skip it on a busy day.
  3. A confidence boundary. It says what it does not know and hands the item to a person, with the reason. An agent that guesses silently is a liability.
  4. A record of what it did. When something looks wrong later, you need to see what it did and why.

Where agents work well

Where they do not

The failure to plan for: getting quietly worse

An ordinary automation breaks loudly: an error message, a step that stops. A step driven by an AI model can instead get slightly worse without stopping, and nobody notices until a batch of records is subtly wrong.

So build in a check. Decide how you will spot-check the output (for example, review a handful of items each week), and give the agent a rule for when to stop and ask a person. Our guide to AI agents that do work goes deeper.

Examples of bounded jobs

These are examples of jobs sized correctly for an agent in a small business:

Inside the software you already run

An agent should live inside your existing software (CRM, inbox, calendar, accounting) under accounts you own. There should be no new platform to adopt. Connecting those systems is covered on the system integration page.

Who owns the agent

You do. The accounts, the logins, the instructions and the code stay yours, and what the agent does is logged where you can read it. The policy is written down in who owns the system.

Learn to build one in class → · Small AI tools → · All services →

FAQ

What does an AI agent cost to run?

There are two costs: building it, and running it. The running cost is the AI usage billed by the AI provider to your own account, plus any software fees. Ask for both in writing before you start, and check the usage bill yourself each month.

What happens when the agent is not sure?

It should stop and hand the item to a named person, with the reason. Decide who that person is before the agent goes live.

Can I build an agent myself?

Often, yes. Owners in our AI classes work in Claude Code on custom apps for their own businesses, and no coding background is needed to start.

Can we start small?

You should. One job, one agent, one result you can check. Add a second only after the first one has earned it.

AI classes for business owners · AI tools built for one task · Workflow automation · All services


Get in touch

Email solutions@ikonic303.dev to get the next AI class date, to talk about marketing, or about one AI tool or job you want built. ikonic303 is based in Colorado, and classes are held in the Denver metro.

ikonic303 home · AI classes · what we built · free Anthropic courses · contact