The copy-paste tax

Here is the shape of a normal AI session for most business owners. Open a tab. Paste in the email. Paste in the quote it refers to. Explain who the customer is. Wait. Read the answer. Copy it out. Paste it into the reply. Fix the two things it got wrong because it could not see last month’s invoice.

Then tomorrow, do all of that again from the start.

That is the tax. Not the twenty dollars a month — the ninety seconds of context you re-supply every single time, because the thing you are talking to cannot see anything you did not hand it.

Why nobody notices

Because it never arrives as a bill. It arrives as thirty seconds here and a minute there, in the gaps of an already-busy day, and every individual instance feels efficient compared to writing the email yourself. It is efficient compared to that. It is enormously inefficient compared to not re-explaining your own business twice a day.

The scepticism this produces is reasonable, and it shows up in the way owners actually talk about AI:

“Most of the AI tools released this year are, quite frankly, a waste of your money. They are built for tech nerds, not for people who have to actually go out and quote jobs or manage a shop floor.”

That is a real quote from a real trade audience, and the diagnosis underneath it is correct even though the conclusion is wrong. The tools genuinely are demonstrated by people whose work is already inside the computer. If your work is in a van, a workshop or a WhatsApp thread, a chat box that cannot see any of it is a toy.

What actually changes

Not a better model. Access.

The difference between an AI that you paste context into and an AI that can read the folder is not a difference of degree. It is the difference between an assistant who has to be briefed and an assistant who already read the file. Everything people find impressive about AI at work is downstream of that one property, and nothing else on the first rung of the ladder gets you there.

So the question worth asking this week is not “which AI should I use”. It is:

How many times today did I paste something in so that the AI could see it?

Count it honestly for one day. That number is the size of the problem, and it is also the argument for doing something about it.

The next step is smaller than it sounds

It is not a platform, a project or a consultant. It is downloading the desktop version of the assistant you already pay for, and letting it read one folder. That is rung 1, it takes about ten minutes, and it is the only step on this ladder that changes what the AI can see rather than how well it writes.