Workshop

AI for Small Business and Entrepreneurs

  • Monday, November 2, 2026 2:00 pm to 4:00 pm
  • Monday, November 30, 2026 6:00 pm to 8:00 pm

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

Moderator: Ryan Kelln, Innovator in Residence, Applied AI

You have the business idea. Now come the thousand questions.

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 ↓

Use AI as a research and planning partner. We practice on a sample business: turning a rough idea into a plan, digging into a market, drafting customer-discovery questions, and setting up low-maintenance automation. The most valuable thing you will get is usually a better question to test in the real world.

What we did

We practised on a made-up business: Spoke & Chain, a one-room bike repair shop. You played Sam, the owner. You taught an AI agent how the shop works, checked the ideas it suggested, chose one small idea, and had the agent build it. Every name and number was invented, so nobody put real business details into a library computer. You can use the same method at home on your own business.

Teach it how the work happens

The most useful thing you can do is give the AI the information it needs. If you only say “help with my business”, it has a great deal of guessing to do.

Get2Know is an interviewer I built for this. It asks about your business one question at a time, saves a profile you can read and correct, suggests things worth trying, and helps you write a brief for one small improvement. A set of instructions like that is called a skill. The agent reads the skill before it does the job.

Good answers to an interview:

  • Answer the way you’d reply to a text. Short is fine.
  • “I don’t know” is a good answer. It leaves a gap to fill later.
  • Ask questions that show how the work happens. When a customer asks if their bike is ready, where do you look? When a payment arrives, how do you decide which invoice it belongs to? Which decisions do you want to keep making yourself?
  • After about eight answers, ask it to read back what it heard. Then fix whatever is wrong.

Before you build: tell the story

Tell the story of someone using the software before you describe what you want built. Who they are, why they use it, what they do first, and what they see at the end. The AI makes up whatever you leave out. You don’t need the whole story first, and you can work it out with the AI in a chat or a prototype. But have it before you start building, and don’t let the AI tell your story for you.

Build the shortest trip through the story that still helps someone, and make it something you can open today. One screen, one job done. Then add the next thing, one at a time, and look at what came out. Building is fast enough to throw out a first attempt and start again, but using what you made is the only way to find out what you should have built.

Be honest with the agent about how much coding you’ve done and how well you know the problem. It will ask for direction based on those answers. Ask it to explain the important decisions, both the ones you give it and the ones it asks of you. Say what you expect the tool to be: a prototype, an occasional tool, or a daily tool you rely on.

Stay safe

  • Read permission requests. An agent that can read, write and delete files on your computer can do damage. Read what it wants to do before you say yes. Keep it inside the folder you are working in, and keep customer messages as drafts you send yourself.
  • Watch for prompt injection. An agent can be tricked by instructions inside anything it reads, such as a web page or an email. All it sees is text, and it can’t confirm who an instruction came from. The risky mix is three things together: it can read your private files, it reads things from strangers, and it can send things out. Break one of them. For example, let it draft the email while you keep the power to send it. This is an unsolved problem. See Simon Willison’s lethal trifecta.
  • What you type goes to the AI company. Describe your work in general terms and use made-up names until you have read the provider’s terms. Records are a separate problem: Standin, in the guide below, helps there.

Keep learning by asking

The AI will explain things as many times as you want. Try:

  • “I have this problem. Is there anything you could help with?”
  • “Explain what you just did, step by step, in plain words.”
  • “How did you get to that answer?”

It can be wrong about itself too, and it won’t know the newest things people do with AI unless it searches the web.

Set up at home

Two guides help you set this up on your own computer. The first is for you. The second is for your coding agent to read.

  • Set up OpenCode: an app that lets an AI work with files on your computer. Skip it if you already have a coding agent.
  • Use Get2Know with Standin (plain text, for your agent): the interview, building one small thing, and preparing exported records safely.
  • Get2Know on GitHub: the interview skills, which the guide has your agent install. To practise as Sam first, the workshop kit has them with Sam’s made-up files.

Open the folder you want to use with AI for your business in your coding agent, then paste this message:

Read https://ryankelln.com/project/tpl_2026/w06-small-business/get2know-with-standin.md
and follow it.

First ask me: is this the folder I want to use with AI for my business?
If not, stop and tell me to open the right folder and paste this message again.
Work in this folder. Don't create sub-folders for the install or start another
coding agent session; I will start any new session myself.

Help me install Get2Know here from its GitHub repository, and the official
Standin release for my computer. Read their instructions first. If this folder
already has an AGENTS.md or other instructions, show me and ask before
changing anything; never replace them. Don't change my other folders.
Don't read this coding agent's own settings or credential files.

Start with invented examples only. Explain what I need to do myself,
including choosing a provider, entering credentials, and Standin's init,
approve and restore steps. Don't run those Standin steps for me.
Don't read real records, private mappings or restored results. Don't turn
off guards or claim protections this coding agent doesn't support.

Check that Get2Know's skill loads and Standin runs. Then start the interview,
one question at a time. If a download or step fails, tell me; don't invent it.

What you can take home

Sessions