A Look at Agents · Lesson 3 of 3
When to bring in help
The last lesson is the honest one about where do-it-yourself ends. Short version: further than you think, with three rules for the far edge.
What you can now do alone
Everything in this course, which is more than most businesses ever set up. You can hold a real working conversation, write prompts from scratch and fix broken ones, run six saved workflows, and operate Projects that know your business. You know why prompts work, which means you can build your own. That was the promise on the front page, and if you did the exercises, you've kept it.
Where the stakes change
Three situations raise the cost of getting it wrong, and they're worth naming plainly:
- Systems talking to systems. When AI connects to your calendar, your invoicing, your website, or your phone system, a mistake doesn't sit in a chat waiting for review. It happens.
- AI talking to customers unsupervised. A draft you review is safe. A bot that answers customers alone carries your reputation with no editor.
- Anything touching money. Prices, invoices, payments. The [CHECK] habit exists because this category punishes guessing hardest.
None of these are off limits. Businesses run all three well. But they're the difference between a saved workflow and a built system, and built systems deserve three rules.
The three rules of boring automation
- Boring first builds. The first automation should be one you review before anything leaves the building. Nothing that talks to customers or moves money unsupervised on day one.
- You own everything. Whoever builds it, the accounts, the logins, and the know-how stay with your business. If your helper disappears tomorrow, nothing should break.
- Measure it or it didn't happen. An automation earns its keep in counted hours saved, not in how impressive the demo looked.
If you hire anyone to build AI into your business, me or anybody else, hold them to those three. A good builder won't flinch.
And because every lesson in this course hands you a prompt, here's the last one. It runs any automation idea through the three rules before you spend a dollar on it:
I'm considering this automation for my business: {describe it in a few sentences}
Run it through three rules and answer each one plainly:
1. Boring first version: what's the smallest version where a person reviews everything before it reaches a customer or touches money?
2. Ownership: what accounts, logins, and know-how would I need to keep in my own hands so nothing breaks if my helper disappears?
3. Measurement: what number would prove it's saving hours, and how would I count it?
Finish with the honest part: what could go wrong if this ran unsupervised on day one?