Your opinion of AI is three years old

When I ask a business owner whether they use AI, I almost always get one of two answers.

The first is an enthusiastic yes. And when I ask what for, the answer is writing help and looking things up. Drafting an email, cleaning up a proposal, asking a question they would have typed into Google. Real value, honestly. Also the ground floor of a much taller building.

The second answer is a wave-off. It's hype, it's slop, it makes things up, it has nothing to do with a business like mine. Also honest. Also earned, because the early tools overpromised constantly.

Here's what both camps have in common: they made up their minds when AI meant a chatbot. A box you type questions into that answers like a search engine with better manners. That verdict was reasonable when it was formed. The problem is that the technology changed underneath it, and the verdict didn't get updated.

A chatbot answers. An agent does.

The tools I put to work for clients today are not chatbots. They're agents. The difference is simple to state and hard to overstate: a chatbot answers questions, an agent does the work.

An agent can open the files, read the account history, cross-reference the invoices, notice the price increase buried in a vendor email, draft the reply, and follow through. It's the difference between asking a consultant a question and handing an employee a job.

That's why conversations about AI go sideways so fast. When I describe what an agent did on a real engagement, someone who's still picturing the chatbot hears something that sounds impossible, because it genuinely is impossible for the tool they have in mind. They're not rejecting the evidence. They're rejecting a kind of tool they haven't met yet.

If a story about AI doing real work sounds like a lie to you, first check whether you're picturing the right tool.

"That doesn't apply to my business"

The second thing that keeps the verdict frozen is deeply human: we are bad at transferring a principle from someone else's situation to our own. Unless an example is our industry, our software, our exact annoyance, we file it under things that work for other people.

So the accountant hears a story about a contractor and thinks, I don't do field quotes. The contractor hears a story about a publisher and thinks, I don't have a catalog. Everyone is hunting for the one example that matches their business exactly, and dismissing the whole thing when it doesn't arrive.

But that's how you evaluate a narrow tool. A plasma cutter is narrow: if you don't cut steel, you genuinely don't need one. An agent is not narrow. The honest question is not "do I have that guy's problem." It's "do I have recurring work that happens on a computer." Every business I have ever walked into does.

Try this instead. When you hear a story about AI doing something for someone else, strip out the specifics and look at the shape of the task underneath:

• Reading a pile of documents and surfacing the three things that matter.
• Noticing a change in an account before it costs money.
• Building the small internal tool that was never worth a developer's quote.
• Chasing the paperwork that follows every job.

None of those belong to any one industry. I have yet to meet a business that doesn't have all four.

The 90/10 flip

Here's why I think this matters more for the businesses I work with than for anyone else.

If you run a client-serving business, a contracting firm, a practice, a studio, a shop, you know the ugly ratio: most of your week goes to everything around the work, and a fraction goes to the work itself. Quoting, scheduling, invoicing, chasing, filing, following up. You didn't start the business to do any of that. It's just the tax you pay to do the part you're actually good at.

Agents are the first technology I've seen that genuinely attacks that ratio. Not the craft. The craft is the last thing anyone wants automated; nobody is asking for machine-made memories or machine-judged workmanship. What can be carried is the administrative layer wrapped around the craft, and that layer is exactly the kind of recurring, computer-bound, rule-following work agents are built for.

Which leads me to a claim I'll stand behind: there has rarely been a better time to run a small client-serving business. The thing that used to force you to hire early, or stay small, or work nights, is the thing that's now becoming cheap. The owners who see that first get to spend the difference on the work, and on their clients, while their competitors are still typing questions into a box.

The verdict is the risk

Notice what I'm not saying. I'm not saying buy anything, and I'm not saying believe me. I argued last month that the right order is think first, AI second: start from the business problem, not the tool. The same order applies to being convinced. Judge AI by other people's stories and you'll acquit or convict based on whether their life resembles yours. Judge it against your own most annoying process and it gets a fair trial.

The real risk isn't buying the wrong subscription. It's running your business on a three-year-old opinion.

Every month that opinion goes unexamined, the owners who updated theirs pull a little further ahead: an hour saved here, a missed charge caught there, a quote out the door a day faster. None of it is dramatic. All of it adds up.

What to do this week

One exercise, five minutes, no purchase required.

1. Date your verdict. Write down when you actually formed your opinion of AI, and which tool you were using at the time. Be honest. For most people it's a free chatbot, a long time ago, tried twice.
2. List three tasks. Pick three recurring tasks in your business that happen entirely on a computer and that nobody enjoys.
3. Price them. For each one, write what it would be worth, in hours or dollars, if it ran mostly on its own.

You now have something most owners never build: a current question to replace an old verdict. Not "is AI hype," but "what are these three things worth to me." That's a question with an actual answer, and if you want a second set of eyes on your list, that's exactly the conversation we like to have.