What AI Can and Can't Do For Your Business — And How to Tell

Here's the part almost nobody explains.

When these tools are trained, one of the most powerful methods used is to set them loose on problems where a machine can check whether the answer is right.

Think about what that means. The system attempts a problem. Something automatically checks the attempt. Wrong — try again. Still wrong — try again. Right — remember what worked. No human needed anywhere in that loop. It can do this millions of times.

Mathematics works like this: a proof either holds or it doesn't, and that can be checked. Code works like this: it either runs and does the job, or it doesn't. So these tools have been through an extraordinary amount of practice on exactly that kind of problem — and it shows. In one controlled test reported by the mathematician Terence Tao in July 2026, four AI systems were set ten genuinely new research-level mathematics problems, and seven of the ten were solved to publication standard by at least one of them.

Now think about what has no such loop.

Whether to give a customer another two weeks to pay. Whether your cousin is the right person to leave the shop with. Whether to take an order you're not sure you can deliver well. There is no machine anywhere that can check those answers, so no system has ever practised them. Not once.

That's the whole pattern. Not intelligence. Not how advanced the model is. Whether the work is the kind a machine can mark.

Take any task in your business and ask one question:

Could a machine tell whether it got this right?

That's it. That single question sorts almost everything.

If the answer is yes, this is work AI does well — often better and faster than you, and often well enough that you should stop doing it by hand.

• Working out what a job actually costs you, once you list the inputs
• Writing an invoice or a quote from figures you supply
• Checking whether your paperwork covers what your sector requires
• Drafting a customer message, a post, a description
• Building a booking form, a website, a record sheet
• Sorting a list, spotting a duplicate, adding up a month

If the answer is no, this is yours, and it is going to stay yours.

• Whether this customer is going through something and deserves patience
• Whether the person undercutting you down the road will still be trading next year
• Who to trust with your keys, your money, your name
• When to say no to work you could take but shouldn't
• What this business is for, and whether you still want to be doing it in five years

Nothing is coming for the second list. Not this year, not with a bigger model. There is no way to practise those, because there is nothing to check them against. It's also why we're honest about what AI agents can actually do today.

Look at where your hours actually go.

For most business owners we work with, the honest answer is that the majority of the week is spent on the first list. Writing the same message again. Working out a price by feel because there's no time to do it properly. Chasing a payment. Trying to remember who hasn't been back. Redoing paperwork that got lost.

All of that is checkable work. All of it can be handed over.

And every hour spent on it is an hour not spent on the second list — which is where the actual value of you being the owner sits. Nobody else can decide who to trust, which order to take, or what this business is becoming. You are the only person who can do that, and it is currently competing for your attention with sending the same WhatsApp for the ninth time.

That's the real cost. Not the hours. What the hours are being taken from.

Some of what's on the first list is work people take real pride in. Knowing every customer's order by heart. Keeping the books by hand, neatly, for years. Remembering it all without writing anything down.

That pride is earned, and it's not what we're asking anyone to give up.

But it's worth separating two things. There's pride in knowing your customers — which is the second list, which is irreplaceable, and which is what actually built the business. And there's pride in carrying every detail in your head — which is the first list, and which is a load you took on because there was never another option.

The first one is who you are. The second one is a burden that looked like a skill because you had no alternative. Putting the burden down doesn't cost you the first thing. It protects it, by leaving you the room to keep doing it as the business grows. We wrote about that trade-off in AI plus kasi wisdom.

Take a piece of paper. Two columns.

On the left, write down everything you did last week that a machine could have checked. Prices worked out, messages written, records updated, paperwork filed, the same question answered for the fifth time.

On the right, write down everything you did that only you could have decided.

Look at the two lists. If the left one is longer — and it almost always is — you now know exactly where to start, and you know it from your own week rather than from anyone's advice.

Then pick the single most expensive thing on the left, and hand that one over first.

There's a version of this that's worth sitting with.

If the machine takes the checkable work, what's left is the work that requires judgement about people, risk, timing and trust. That's harder work than sending invoices. It's also the work that was always the point of running your own business, and the reason you started rather than taking a job.

The tools didn't take your craft. They're handing back the part of it you never had time for.

Once you've written those two lists, a second question tends to follow: which column are my children being prepared for? That one is answered separately — What Should I Teach My Children in the Age of AI?

Find out what kind of business you're running — free, 4 min →