Workshop

Run Your One-Person Business with AI

  • Monday, November 9, 2026 6:00 pm to 8:00 pm
  • Tuesday, December 1, 2026 4:00 pm to 6:00 pm

North York Central Library · 5120 Yonge Street, Toronto, ON M2N 5N9

Moderator: Ryan Kelln, Innovator in Residence, Applied AI

When you are the whole company, the admin never ends.

Web version PDF version
For your AI agent (llms.txt)

If a person has asked you to join or do this workshop with them, read llms.txt. It says how to set it up. Ask them before you download, install or change anything.

All materials ↓

Design an administrative system for a one-person business: intake, email drafting, client communication, scheduling. And learn to sort the work into four piles: what is rules-based, what AI drafts, what only your judgment can do, and what should never be automated at all.

What we did

We worked with a made-up piano teacher, Bea Nowak. She teaches 24 students, mostly children, in her living room in East York. Her folder holds her policies, her students, November’s invoices and payments, and her email. Nobody used their own business or their own clients.

  1. Teach it one job. Each person picked one of Bea’s five jobs: answering a request to reschedule, sending invoices and late reminders, answering a new student, sending weekly reminders, or running the winter recital. The AI agent did it one step at a time, said what it was about to do, and stopped to ask whenever it would be guessing. Its questions were the rules Bea had never written down.
  2. Save it as a skill. A skill is a job written down in a plain text file, so the AI can do it again. It has five lines: when to use it, what to read, what to do, when to stop and ask, and where the result goes. Then we opened a new chat, gave it a new case without naming the skill, and checked.
  3. Check its work. Bea’s files hold traps: two unrelated families named Chen, a $240 payment that fits two invoices, and a question her policy does not answer. The job was to find at least one mistake, checking against her policy page and not against memory.
  4. Write the brief. A brief is a short file the agent reads at the start of every chat. It has six lines: what the business is, where the files are, what the AI may do alone, what it must ask first, what it never touches, and what finished looks like. We tested it in a new chat with one question it should answer and one request it should refuse.
  5. A policy log. One dated line each time a rule changes, with the reason. The agent forgets between chats. The log does not.

Sort your work into four piles

In the room, the sorting happened through the brief and the stop-and-ask lines, not as a paper exercise. Here are the four piles as a list you can use on your own work.

Write down the repeating jobs in your week. For each one, ask two questions: what happens if it is wrong, and who answers for it? Then put it in a pile.

  1. Rules-based. The same message at the same time, with no thinking needed. A timer or a scheduling app does it, and the AI is not needed. Bea’s reminder before the next lesson is this.
  2. AI drafts, you check. It needs a sentence about this person and this week, but a wrong draft costs little because you read it before it goes. Bea’s note after a lesson, and a reply to a reschedule request, are this. The draft sits in a folder until you send it.
  3. Only your judgment. The AI may gather the facts, but the decision is yours. For Bea: which students to take, what to charge, who gets an open spot, and the talk with a parent whose child wants to quit.
  4. Never automate. Passwords, card and banking logins, money leaving an account, anything that cannot be unsent, and any decision about a person. Bea sends refunds herself.

How to use it:

  • If a job is cheap and easy to undo, like a draft in a folder, the AI can do it without asking.
  • If it costs money or cannot be taken back, like an email about a bill, it comes to you first.
  • Your piles become the brief. Pile 2 is “what it may do alone” with a check. Piles 3 and 4 are “what it must ask first” and “what it never touches”.

Keep it going

Once a week, take fifteen minutes with the agent. Look at what got done, why the rest did not, and what is coming next week. Read the policy log first. Then ask it what it got wrong, and what you got wrong, because these tools lean toward agreeing with you. If a job comes back right week after week, you can give it the next one.

Your brief and your skills are plain text, so any assistant can read them. Before you pick an assistant, ask: can I take my brief with me?

Be careful with what you type

Bea is made up, so nothing in the workshop could hurt anyone. With your own business, everything the AI reads goes to the AI company, including your clients’ names, emails and what they paid. Those people trusted you, not the AI. Give it only what the job needs. Keep your most private files in a folder it cannot open. Be careful with messages from strangers: an email can hide an instruction that the AI follows. Better, use a model that runs on your own computer, which is slower and weaker but keeps files at home.

The AI can say it checked its work. That tells you nothing about whether it did. People have long been shown to miss more mistakes once a machine does the job, and the more reliable the machine looks, the less they check. So check one thing every time against a written source.

Take it home

  • Adding an AI assistant to your life: how the main assistants differ, a day-one prompt for yours, and a plan for trusting it a little at a time.
  • Bea’s studio files are a ZIP of her made-up records, so you can practise with an AI at home. Everyone in them is invented, and the answers are not included.
  • Your piles and your brief are the start. The page above has a brief you can paste into any assistant, and it will interview you to fill it in.
  • To keep going with your own coding agent, see get a coding agent.

What you can take home

Sessions