AI Made It Cheap to Build. It Didn’t Make It Good.

There is a tax nobody warned you about when you started using AI in your business. It is not a dollar amount and nobody is coming to collect it. It is this: the more AI does, the more your judgment about whether the result is any good actually matters. Not less. More.

That feels backwards, and it catches people out constantly. AI is supposed to make things easier, and in a lot of ways it has. Content that took a week takes an hour. A website that took two months takes a weekend. A tool that would have needed a developer gets built by somebody in accounts who is good with spreadsheets.

But output is cheap now. Anybody can produce it. The only question left is whether what comes out the other end is right, is good, and is worth putting your name on. That part is still entirely on you, and it is harder than it used to be.

Output got cheap. Attention did not.

Think about what has actually changed. Your competitor can now publish forty articles this month. They can spin up a site over a weekend that looks completely respectable. They can ship a customer portal without hiring anyone.

So can you. So can everybody.

When production stops being the constraint, the constraint moves. It is no longer "can we make enough of this?" It is "can we tell the difference between good and adequate?" And adequate is now free and infinite, which means it stopped being worth anything.

What did not become infinite is the stuff you already had. Your customers' trust. Your reputation in your town or your niche. Your sense, built over years, of what your particular buyers actually respond to. Those things are worth more now, not less, precisely because they are the only things that did not get mass produced overnight.

This plays out differently in the three places we see it most.

In content: the flood is real and it is already reversing

The first wave was volume. Businesses discovered they could publish endlessly and did. Sites that had eight pages went to eighty in a quarter.

What most of them found is that nothing happened. Traffic did not move, or it moved briefly and then went backwards. The reason is simple once you see it. Content earns attention and citations by containing something worth quoting, and a model has none of your specifics. It has the average of what everyone else has already written. Publishing that at scale does not make you prolific. It makes your site indistinguishable at exactly the moment when being distinguishable is the whole opportunity.

The tell is easy to check on your own pages. Look for a number you could only know from doing the work. Look for the exception that catches people out in your climate, your building stock, your regulatory environment. Look for a position somebody could disagree with. If a page has none of those, it is not doing anything for you, and it is diluting the pages that are.

We are now watching businesses quietly delete most of what they generated last year. Deleting it usually helps, which tells you what it was worth.

In websites: it will build exactly what you specify

AI is genuinely good at producing a website fast. What it cannot do is decide who the site is for, what it needs to say in the first two seconds, or why anyone should pick you.

That is the failure we see most. A business feeds a vague idea into a tool and gets back polished vagueness at speed. The result looks fine. It says nothing in particular. It does not convert, and the owner concludes the design is the problem and rebuilds it, which produces a second site with the same fault.

Think of it as a very capable contractor. It will build precisely what you specify, quickly and cheaply. Hand it a bad blueprint and it builds that, faster than anyone ever built a bad blueprint before. The blueprint is your positioning, and no tool supplies it.

There is a second, quieter problem. Generated sites tend to look competent and generic in the same way generated writing does, because they draw on the same averages. Fine if you compete on price. A real problem if you are trying to charge more than the cheapest option in your market, because visitors read the site as a signal of what you are worth.

In software: the first ninety percent got easy

This is the one people underestimate most, and it is the one with the largest downside.

AI will get a working tool up remarkably fast. A booking system, an internal dashboard, a customer portal, an integration between two things that never talked to each other. For a small business that previously could not afford custom software at all, this is a genuine and welcome change, and we build this way ourselves.

The trap is that working and finished are different states, and the gap between them is where the expensive part lives. What happens when two people book the same slot at once. What happens when the payment succeeds and the confirmation email fails. What happens when someone puts an apostrophe in their surname, or the internet drops halfway through, or a customer hits the back button after paying.

Generated code handles the path where everything goes right. It is much weaker on the paths where things go wrong, and it is confidently weak, which is worse than obviously weak. It will not tell you what it did not think about.

Then there is the part nobody enjoys discussing: security and data. A tool that touches customer records, payment details, or anything you are legally responsible for needs somebody who understands what they are looking at. Not because AI writes insecure code on purpose, but because it writes what it was asked for, and most people do not know what to ask for.

The rule we use internally: the faster something was built, the more carefully it gets reviewed before anyone outside touches it.

You are the quality gate now, whether you meant to be or not

Here is the shift, and it is worth naming plainly because it changes what your job is.

You used to be the maker. You wrote the page, built the thing, made the decisions inside it, and you knew it inside out because you produced it. Ownership was obvious.

Now you are the one deciding whether what came back is good enough to carry your name. That is a different skill and, honestly, a harder one. When you wrote every line, you could defend any of it because you knew why it was there. When a thing is assembled from a machine draft and your edits, ownership has to come from somewhere else. It comes from how clearly you framed the job, how well you knew what good looked like before you started, and how rigorously you actually checked rather than skimmed.

Most people skim. The output looks confident, it arrives finished, and there is nothing in its tone to suggest which parts it made up. That is the trap.

Three questions before anything goes out

You do not need a process document. You need three habits.

• What in here could only have come from us? If the answer is nothing, do not publish it, do not ship it, do not put your name on it. Add the number, the exception, the thing you learned the hard way. That addition is the entire value.
• What happens if this is wrong? If the answer is that you look silly for a moment, move fast. If a customer will act on it, or money moves, or data is involved, somebody has to check it properly and that checking time is part of the real cost.
• Would I be comfortable if the customer knew how this was made? A drafted proposal you rewrote: fine. A condolence note you never read: not fine. The line falls roughly at whether you used the tool to work faster or to appear to have done work you did not do.

Those three take a couple of minutes and they catch nearly everything.

It can read the cookbook. It cannot taste the meal.

That is the whole thing, really.

AI can generate infinite output. It cannot generate judgment, it cannot generate taste, and it cannot generate the sense you developed over years of doing this work with your customers in your market. It knows what has been written about your trade. It does not know what happens on the job.

Which means the businesses that come out of this well will not be the ones using AI the most. They will be the ones who did the harder work: knowing what good looks like, noticing the gap between good and adequate, and standing behind the result regardless of what helped build it.

That is the tax. Judgment, taste, and owning the outcome. Unlike most taxes, this one is worth paying, and paying it is most of what separates a business that stands out from one that just published more.

If you would rather have someone do the building and the judging, that is what we are for. It is also why our websites start with strategy rather than with a layout, and why we settle what you stand for before anything gets designed or written.

See how we help with AI →