A while back I wrote that AI was going to bring back some old habits. The ones where creative work gets treated as casual, as something that doesn’t really count as labour because the effort isn’t visible. I said it in the context of Caribbean designers, watching a technology arrive that made our work look even easier than clients already assumed it was.
I didn’t expect the same thing to happen to the AI itself, on a much bigger stage, in under two years.
Here’s the short version. Companies spent 2023 and 2024 replacing people with AI because it looked cheap. Layoffs rolled through the big tech firms, call centres swapped agents for chatbots, and the pitch to every boardroom was simple: why pay a person a salary when a model can do the work for a fraction of the cost. It looked effortless from the outside, the same way a finished logo looks effortless to someone who’s never sat with a blank canvas trying to make an idea hold together. Type a prompt, and out comes the work. Must be nice.
Except somebody was always going to have to pay for what that prompt actually costs to run, and it wasn’t going to be free forever.
The invoice arrives
Token prices, the units companies get billed on for AI usage, more than doubled between 2025 and mid-2026. Running models like Claude at real scale turned out to be expensive enough that Microsoft had to put limits on how much its own staff could use Claude Code internally. Uber, Nvidia and Amazon all ran into budget overruns nobody had planned for. AT&T alone is now processing close to eight billion tokens a day, and enterprise AI bills across the board have climbed into the tens or hundreds of millions a year for a single company.
None of that made sense if AI really was the cheap replacement everyone was sold. What actually happened is the early pricing was never real pricing. Venture capital was covering the difference behind the scenes so the tools could look unbeatable next to a human salary, which is the same trick printing companies have used for years, throwing in a free logo because they’re making their money on the print run instead. The cost doesn’t disappear when someone else eats it for a while. It just waits for whoever’s left holding the tab when the subsidy runs out.
Nobody asked what it was actually worth
There’s a strange twist inside all of this that any designer will recognise instantly. Engineers inside these companies started treating heavy token usage as a badge of honour, a way to look productive to leadership. It got a name: “token maxing”. People ran AI on trivial tasks they didn’t need it for, because using more of it made them look busy and important, and tech leaders openly encouraged it.
That’s the exact same energy as a client who wants a full brand identity done overnight because they saw someone else’s designer turn one around fast, without asking whether the job actually needed that speed or whether the price reflected doing it properly. Volume got mistaken for value. Nobody stopped to ask what the work was actually worth; they just wanted more of it, faster, because more looked impressive.
Turns out that habit is expensive no matter who’s doing it.
The part that should sound familiar
I’ve said before that underpaying for design doesn’t save money; it just delays the cost and makes it bigger. Cheap logos get redone within a year because they never worked the first time properly. The business owner who tried to save money on the front end usually pays twice.
Enterprise AI spend is following the same arc, just with more zeroes. Nearly half of the US data centre builds planned for 2026 are facing delays or cancellations because the hardware supply chain couldn’t keep up with how fast everyone wanted to scale. Gartner is now telling executives to expect AI to make up more than a fifth of enterprise tech spending within a decade. Some companies are burning through a full year’s AI budget in a matter of weeks. And in areas like customer service, running an AI agent is starting to cost more than the human role it replaced, which means the ROI conversation that never happened at the start is happening now, under pressure, after the layoffs already went out.
That’s not a technology failing. That’s a cost getting revealed after it was hidden long enough for everyone to make decisions as if it didn’t exist.
What to expect from your artist, and from your AI
I’ve written before about what a healthy working relationship with a designer actually looks like. Clear brief, fair pricing agreed up front, proper delivery, respect on both sides for what the work costs to do well. None of that was ever really about designers specifically. It was about what happens when you skip the process because you assume the work is easier and cheaper than it is.
AI didn’t get a pass on any of that just because it doesn’t send an invoice with a face attached to it. The process still exists. The cost still exists. It was just sitting off to the side, subsidised, waiting for the moment it couldn’t be subsidised anymore. That moment has arrived, with AI companies heading toward public listings and the venture money that made the early pricing look magical starting to dry up. Prices are going up from here, not down.
So here’s the question worth sitting with if you swapped a person for a model somewhere in your business over the last two years. Did you actually work out what it costs to keep running, long after the launch excitement wore off, or did you just assume cheap was permanent because it looked that way at the start?
Because if there’s one thing three years of writing about this has taught me, it’s that nothing skilled stays cheap once the truth of what it costs to do properly finally catches up with it. Doesn’t matter if it’s a person or a machine. The bill always finds its way home.
