Here’s a number I didn’t expect to be writing about this year: my AI bill is up 15x on a single seat in thirty days. Not because we did anything reckless, but because the ground moved.   

We used to run a lot of our AI work on a flat Google AI Ultra subscription. Predictable. Budgetable. Then the subscription model went away and the pricing went to pure consumption via Google Cloud; pay per token, every call, every run. Same work, same person, fifteen times the cost. And that’s one seat. Now multiply it across a team, add the agentic tools that loop through a task on their own and burn tokens at a rate a chat window never could, and you start to see the shape of the challenge.   

“Negotiating a better contract” is not really an option in this day and age. With agentic workloads, flat-rate subscriptions  are out of the question, because no vendor can subsidize a tool that runs all night at a fixed price. Consumption pricing is everyone’s destination.   

The promise has been if we deploy AI, it would get cheaper and free up our people for higher work. The first half of that sentence is turning out to be false, and the back half depends entirely on what you do about the first.   

Recently, an AI consultant infamously told Axios that one of their enterprise clients ran up half a billion dollars on Claude in a single month. They’d handed it to the whole company and forgotten to set any usage limits. Half a billion. In thirty days. Microsoft pulled back internal Claude Code licenses after per-engineer costs hit somewhere between five hundred and two thousand dollars a month. Uber reportedly burned through its entire 2026 AI budget by April. These aren’t scrappy startups losing the plot. They’re the most sophisticated technology operations on earth, and the invoice still caught them off guard.  

Which raises the question we all need to sit with: if AI costs more, not less, how do you pay for it?   

CloudBees CEO, Anuj Kapur, gave Axios the honest answer. Workforce cuts at AI-heavy companies, he said, may simply be the only lever they can pull to offset their AI bills. Not “AI replaced the workers.” The bill got too big, and the people were the thing they could cut to cover it.  

That’s the math agencies are running right now, whether they say it out loud or not. Billions sunk into proprietary AI platforms, the tech debt of keeping them current, and now climbing token costs on top — and when that bill comes due, the lever is the same one it always is: cut the people.   

There’s a different way to run this math, and it’s available to any shop willing to be deliberate.    

First: pace the rollout to what your headcount can absorb. The temptation is to bring the bill down the fast way — cut roles, fund the tooling with the savings, ship a leaner shop. Resist it. Deploy at the speed your team can adopt without anyone losing a job. Yes, it’s slower than it could be. Slower also never becomes the agency version of that half-billion-dollar headline.   

Second: treat AI cost as an engineering challenge, not a procurement one. The efficiency work is real and worth doing. One of our internal pipelines — an automated proofing and compliance agent for a national delivery brand’s print mailers — actually got cheaper as it matured, because re-architecting how it handles documents cut token use about 20% while the agent did more, not less. That result is available to anyone willing to do the unglamorous work. But treat it as table stakes, not the strategy. Efficiency was never the point.   

Because here’s what clients actually hire an agency for, and it isn’t tokens. It’s people. They don’t call asking for more AI. They call asking for more senior attention, more time, more of the judgment that doesn’t come out of a model. AI is that leverage. Behind the people, not in place of them.   

Third: run the actual numbers, out loud, before someone runs them for you. An enterprise tier that puts this stack in every person’s hands — secured, governed, guaranteed never to train a public model on client work — runs around forty-five thousand dollars a year. That’s a junior salary. The whole decision sits in that one line item: spend it on the tool, or spend it on the person. It doesn’t have to be a zero-sum choice. You should invest in both–spend more on AI now so you never have to choose the tool over the person — because the day the math forces that trade, you’re no longer the thing your clients hired.   

AI didn’t make us cheaper. It made the choice clearer. The only mistake is pretending you don’t have one to make.