Notes
Letting AI lead the way
In February I recorded a conversation about a provocative idea: stop treating an AI tool like an intern, and let it help plan the day.
The useful part was not “demote yourself” or “make the tool your CEO.” It was this: the exhausting part of a small business is often not the work itself. It is deciding, over and over, what to do next.
That is still worth talking about. The version of the idea that is useful now is quieter, and it keeps you in charge.
The bottleneck is often the next decision
If you run a small business, you already wear several hats in the same morning. One moment you are checking a price. The next you are answering a message. Then you are trying to decide whether today is for inventory, outreach, or a repair.
By mid-afternoon it can be hard to choose even a simple next step. That is ordinary decision fatigue, not a character flaw.
A lot of people then open a chat window and still do all of that deciding themselves. They ask a tool to draft an email or summarize a page. That can help. It does not remove the feeling of being the operations manager of your own day.
It can be useful to ask a tool for a proposed order of work, a short list for the morning, so you spend less energy negotiating with yourself. The list is a draft. You still decide what is safe and sensible to do.
“Orders” are a poor place to start
That February conversation used a strong metaphor: treat the tool as the boss, and treat its daily list as orders.
I would not start there.
A language model can sound confident while missing something obvious. In the original conversation we used a hypothetical: an assistant looks at last winter’s sales, sees that heavy coats sold well, and tells you to put parkas out on a hot day in July. The numbers in a spreadsheet are not the same thing as standing outside in the heat.
You need veto power. You need it before the first experiment, not after something has already gone wrong.
A tool can help you start. It should not send a message, change a price, move money, or publish something unless you have reviewed it and decided that is allowed. If the suggestion fails a common-sense check, you stop it. That is not being difficult. That is the job.
The tool matters less than the limits
You do not need a particular platform to try this.
These days I usually start with Grok. Hermes is still one of several agents in the mix. Other tools come in when they fit a specific job. I would not begin by picking a stack and then looking for a use for it.
Whatever you use, write down a few limits first:
- What job is this list for?
- What information may the tool see?
- What may it only draft, never do on its own?
- How do you stop it?
Keep private records and payment access out of a first trial. A friendly chat is not a reason to connect everything you own.
A cautious look, not a surrender
If you want to test the idea, keep the trial small.
Pick one familiar area of work, not your whole business. Ask a tool you already use to propose a daily order of tasks for that area, based only on information you are comfortable sharing. In the morning, read the list. Do the items that are clear, safe, and useful. Veto anything that does not make sense. Write down what you changed and why.
At the end of a short trial, look at whether the list actually saved you decision time, or whether checking and correcting it took as long as planning the day yourself. If it did not help, stop. You do not owe a method your loyalty.
The useful prize is not speed for its own sake. It is learning sooner whether a small idea is worth another afternoon, instead of sitting on it for months. That only works if you are still the person who can say no.
Keep the human part
The point of letting a tool lead a little is not to disappear from your own business.
It is to spend less of the morning deciding, and more of it doing the work only you can do: talking with a person, judging a situation, noticing that it is too hot for parkas.
Freedom, in this sense, is not infinite choice. It is knowing a sensible next step, and still being the one who can change course.