Call it automation when you set the steps ahead of time. Call it an agent when the software picks the next step toward a goal. Start with automation for a shop job. Add an agent step only where the input changes from case to case.
Look at the job first. Pick the tool after that.
What is the plain difference?
Look at a recipe. Write each step down. Run the steps in order. Call that automation.
Look at a helper with a goal. Give it the goal. Give it a short list of tools. Let it pick the next step from the input. Call that an agent.
Look at the source of the choices. Is it you, ahead of time? Call it automation. Is it the software, in the moment? Call it an agent.
Read the AI automation business ideas with this test in mind. Mark each idea as a recipe or a helper.
When should you use plain automation?
Use it when the input looks alike each time. Use it for a form to a sheet. Use it for a reminder before an appointment. Use it for a note to the owner after a new lead.
Pick it when a wrong step hurts the shop. Keep the steps in plain view. Test each step by hand. Fix each break on the spot.
Start with a new client here. Show a working result fast. Keep the setup simple to explain.
When should you use an agent?
Use it when the input changes from case to case. Use it for an email with a vague question. Use it for a chat with a customer in a rush. Use it for a messy pile of notes.
Give it a narrow goal. Give it a short list of tools. Set a short list of allowed actions. Send the odd cases to a person.
Read the chatbot post for a narrow agent job with a handoff to a person.
How do you mix automation and agents?
Build the recipe around the agent. Let the recipe catch the input. Let the agent read it and pick a label. Let the recipe do the rest.
Keep the agent step small. Ask it for a label. Ask it for a short reply draft. Check its output against a short list of allowed labels. Send the rest to a person.
Keep a log of each run. Read the log with the owner. Mark each wrong label. Fix the setup.
How do you test the fit of a job?
Get a stack of real cases from the shop. Sort them by hand. Put the alike cases in a pile. Put the odd cases in a separate pile.
Look at the size of each pile. Is the alike pile the big pile? Build a recipe for it. Send the odd pile to a person.
Is the odd pile the big pile? Try a small agent step on it. Run it next to a person for a while. Compare the output of the agent with the output of the person. Keep the agent step only if it holds up.
Write down the test. Show it to the owner. Repeat it when the shop changes.
How do you explain it to a shop owner?
Skip the jargon at the start. Name the task in the words of the owner.
Show a plain before and after. Show the task by hand. Show the task with the setup. Show the spot where a person steps in.
Read the post on selling AI agents to small businesses for the rest of the talk. Read the agency versus product post for that choice. Read the checklist post on what you need for the setup side.
Frequently asked questions
Is an agent better than automation?
Pick the tool for the job. Use the recipe for fixed steps. Use the helper for messy input.
Do you need an agent to start?
Start with a recipe. Add an agent step later if the input calls for it.
Is a chatbot an agent?
Look at the job of the bot. Is it a fixed script? Call it automation. Does it pick the next step from the chat? Call it an agent.
Which should you pick for a new client?
Pick the setup with the easy explanation. Pick the setup with a fast fix. Start with a small setup.
What should you do next?
Try the scorer on your idea tonight. Try the generator if you are stuck on the trade. Use the idea board for a narrow task. Read the step by step post for the full order of things. Use the newsletter form below this post.