Two lessons from a course for small business owners. Every interaction works, so go ahead and fill them in.
StartI designed and built this course for a client's small business program. This is a portfolio rebuild of the first two lessons, with the client's branding and materials taken out.
Learners leave with a scoped problem for their first AI agent, a north star they can download, and an audit of the data they already have.
Most business owners have experimented with AI at some point, usually to draft an email or summarize a document.
That kind of tool is great for getting a single task done fast. An AI agent puts that same intelligence to work on your business every day, using your data and aimed at your goals. This lesson explains what an AI agent is, why it’s different from what you’ve already used, and what it makes possible for a business like yours.
Right now, your business knows things you don’t. The patterns costing you money, the opportunities you keep walking past, the clients most likely to renew are all in the data. It lives in your invoices, your calendar, your inbox, your booking history. It’s everywhere, and it’s just buried across multiple apps, gathering digital dust, doing nothing for your business beyond existing.
AI agents are how you change that. They put your data to work and translate it into something actionable.
Most of us have used AI to draft an email, answer a question, or summarize something. That kind of AI is a large language model, and it works like a very smart conversation: you open it, you type, and it responds.
Every time you close the tab, it forgets everything. You’re the one providing context every session, and the moment you stop prompting it, it stops working.
An agent maintains context, connects to your data, and can take action between sessions without you having to prompt it each time.
With a large language model, you’re the operator in the loop every time, keeping it pointed in the right direction. With an agent, you’re the architect. You design and direct the system, and then the system runs.
The tasks that used to consume entire working days, including analysis, follow-up, and pattern-spotting, are now ones that AI agents can handle a lot of. That doesn’t mean bandwidth stops being a constraint altogether. If you’re a chef, your agent still can’t cook the food.
What it means is that for the first time, you can carve out real time to work on your business: time to look at the numbers, spot what’s working, and make strategic decisions instead of constantly firefighting. Most business owners never get to that layer. AI agents open that door.
It helps to picture your agent the way you’d picture a new hire. They need a desk, a briefing, the tools to do the job, and time to get good at it. Your agent goes through the same four stages. Select each one to see what it involves.
Treat your agent like a Director or COO, not as an intern. The more seriously you take the briefing, the better it performs.
The first two steps of building an agent, scope and surface, are about knowing exactly what you’re building and what you have to build it with. Neither one is glamorous, but skipping either one is the most common reason a first agent fails to deliver.
Not every problem is worth solving with AI right now, and not every opportunity is worth chasing. Scoping is how you decide which one deserves your attention first. Without a clear scope, you risk trying to fix everything at once and ending up with nothing. With one, everything that follows has direction.
The most common mistake at this step is treating a wish like a scope. A wish describes how you want to feel. A scope describes exactly what you want the agent to do.
A real scope has three parts: a specific outcome, a specific mechanism, and a specific metric, and you need all three. Complete the questions below to lock yours in.
Learn more about identifying your north star: North Star Metric, Amplitude
Every problem worth solving sits somewhere on two questions. If you have more than one problem you’d like an agent to take on, the value matrix helps you pick where to start.
It sorts problems by two things: who benefits (your own operations or your customers) and what kind of work it is (improving something you already do or creating something new).
Your first agent belongs in Internal and Improve. It’s the lowest-risk place to learn the process, because if something goes wrong, your customers never see it. Once you have an agent you trust, you can move into the other quadrants.
Already know which problem you want to solve? Find where it sits anyway. If it lands outside Internal and Improve, look for a smaller version that stays within your own operations for your first build. Select each quadrant to see an example and when to pick it.
Most first agents land in Internal and Improve. It’s the lowest-risk place to learn before you touch anything customer-facing.
AI has been trained on almost everything humanity has ever written, so it knows an enormous amount about the world, including industries, markets, trends, and best practices. But it knows nothing about you.
This step is about surfacing what you already have and getting it ready, so your agent can point to the right sources.
Before you look at anything yourself, test your thinking against your AI assistant first, because it often surfaces an angle you hadn’t considered. Open it, describe the problem you scoped, and ask:
Let it tell you what to look for. This single conversation can save you hours of guesswork and may surface data sources you hadn’t considered.
Example: You describe a scope focused on catching overdue invoices before they reach 60 days. Your AI assistant comes back with your QuickBooks invoices, payment history, and client contact list. Then it suggests something you hadn’t listed: your email history. Your past reminders show which clients pay after one nudge and which ones go quiet. You hadn’t thought of that as data.Once you have your list, work through three questions for each source before granting access to anything.
Before you connect any data source to your agent, you need to know it’s ready to use. Not every source on your list will clear all three questions, and that’s fine. The audit tells you what you’re working with before you build anything.
Start with the example below. It follows an invoice follow-up agent across four data sources, so you can see what earns a Go, what earns a No-go, and what to do about each.
When you’re ready, select Start your own audit, add each data source to your list, and answer the three questions to get a verdict for each.
Know what you want your agent to do, what data you have to work with, and start there.