Tokenmaxxing: Why AI Keeps Getting More Expensive — and What It Means for Leaders.

Back in the days of the Roman Empire, in the year 49 AD, same year as the Council of Jerusalem and the Bosphoran War, the Roman philosopher Seneca, one of the great masters of Stoicism, sat down to write a letter to his father-in-law that eventually became his masterpiece. Yes, in "On the Shortness of Life," Seneca wrote something that the first time I read it hit me like a slap in the face. He said: It is not that we have a short time to live, but that we waste much of it. Life is long enough, and a sufficiently generous amount has been given to us for the highest achievements, if it were all well invested. End quote. Now, pause here for a second. He's saying the point isn't having more time available. It's how we use the time we already have. Because I always assumed, and maybe you did too, that the real problem in our working lives is simply not having enough time, not having enough hours. And supposedly AI was going to fix that age-old problem. AI, the most powerful productivity tool ever invented, was meant to be the thing that finally hands us our time back and our money back by making things cheaper, by making work cheaper, automating the boring stuff, freeing up the human, watching the costs fall. That was the deal. Except, a couple of weeks ago, the president and COO of Uber, Andrew MacDonald, sat down for an interview and said something that shows us that the deal is not really materializing. Talking about the billions of dollars companies are now pouring into AI, he said, and I'm quoting, "The link is not there yet." The link between how much AI you're burning and how much value you're actually producing is not just there yet. And why not? Well, part of the answer lies in Seneca's words. See, 2,000 years before a single AI token was ever processed by a single chip, Seneca had already diagnosed a paradox that AI has only made more obvious. His point wasn't that people are short on time; it's that we waste much of it filling our days with unnecessary activities. And why do we do that? Because activity feels like meaning. And staying busy lets us dodge the much harder, much lonelier work of doing what really matters. If that sounds familiar to you, it's because a Georgetown computer science professor named Cal Newport made it popular and gave the same idea a 21st-century name. He called it the difference between deep work and shallow work. Between the small handful of things that genuinely move your life forward and the avalanche of busy, reactive noise that just makes you feel productive. Seneca and Newport, 2,000 years apart, two different continents, but describing the exact same human bug. And that bug now has a 2026 name. That is token maxing. And that's a trap we're going to unpack in today's episode of Between You and AI. Here's your host, Andrea Iorio. I'm an Italian keynote speaker and USA Today bestselling author. I spent 5 years as head of Tinder in Latin America. I was a chief digital officer at L'Oréal Brazil, and I teach in the MBA at Fundação Dom Cabral. I'm also a columnist for the MIT Technology Review and Wired magazine, and my latest book is Between You and AI, published by Wired. You can get to know more about me and my keynotes at my website, andreaiorio.com. And to start off, let me take you back to 1865, to Victorian England, where lived a British economist named William Stanley Jevons, England at the time was obsessed with a single question: were they about to run out of coal? Because something was happening. With the newly invented steam engine, the common sense assumption was simple: if you made the engine more efficient, you could squeeze more work out of less coal, then the country would obviously burn less coal. Right? More efficiency, less consumption. Except Jevons looked at the data and found The exact opposite. Every single time engineers made engines more efficient, England burned more coal, not less. And why? Because when you make something cheaper and more useful, people don't just pocket the savings and go home early. They invent a thousand new uses for it. Cheaper coal didn't give Victorian England leisure. It gave them factories running around the clock and humans who started behaving like the machines they ran. Today we call this the Jevons paradox, and it's one of the most reliable patterns in all economic history. Every efficiency gain triggers a demand explosion, never a leisure dividend. Hold on to that because we're going to get back to it. If in 1865 the resource was coal, well, in 2026, the resource is cognition. Thinking itself. And how do you measure thinking? Well, in the age of AI, it's through tokens, a concept every leader, especially in HR, should get comfortable with. See, here's what you really need to know. A token is a tiny unit of text that an AI reads or writes, roughly a word. Sometimes it's a piece of a word, and every single one is metered. Think of it as a taxi meter for thinking. The model never stops the meter while it works. And so, which makes tokens two things at once. First, your AI bill on the one side, and more importantly, on the other side, a live proxy for how much your employees are actually leaning on AI. And the second part is exactly where token maxing comes into the picture. Because until recently, token maxing was all the rage inside all companies. The logic went like this: if you wanted to know which employees are the most innovative with AI, well, just track their token usage. The more token they burn, the more productive their AI agents must be, or at least the more AI-forward that person is. That was the hypothesis. And that's when everybody started tracking tokens. At Meta, an employee built an internal leaderboard, nicknamed "Clodonomics," ranking the company's 85,000 employees by how many tokens they burned, handing out badges like "Token Legend" and "Cash Wizard." At Amazon, there was a similar one called Hero Rank, and employees later admitted they'd assign AI agents to do pointless busywork just to climb up the rankings. Just to look productive. And that's why it backfired. Meta took its leaderboard down. Microsoft canceled Claude Code subscriptions across several product divisions, according to The Verge. Uber burned through its entire 2026 token budget in the first 4 months of the year, largely on Claude Code. And Salesforce CEO Marc Benioff said his company's Anthropic bill alone will run about $300 million. Not cheap at all. So why did token maxing fail? Two reasons. The first is obvious, and it's the one McDonald from Uber named. There was no real correlation between token spend and business impact, especially once employees realized they could game the metric. The second reason is less obvious and takes us straight back to a man we already met. Remember Jevons and the coal? Well, same law. Except now they call these tokens. Even as the price per token falls, total spend keeps climbing because cheaper tokens simply unleash far more usage. Since the start of 2022, the number of tokens, according to Exponential View, processed per quarter has risen roughly 17,000 times, an exponential growth in an order of magnitude steeper than Moore's Law. It's also due to AI agents because they pour fuel on the fire. They turn one human request, one human prompt, into dozens of autonomous steps running in the background, and sometimes we just lose track. The worker no longer types every request, but the agent just keeps going and the meter keeps running. And here we land back right to Seneca. To feel busy, we fill the time AI hands us back with more activity. And so we overspend, which raises the real question: is AI actually delivering the cost savings it promised? Not actually. Arvind Jain, the CEO of an enterprise AI company called Glean, went on CNBC and said something I never heard put this plainly. He said this is the first time he can remember that technology costs the same as people. Companies, he said, are now literally choosing between tokens and humans, and the AI budget is growing in place of future headcount. See, the money you're spending to feel productive with AI is in some companies the exact money that could have hired the next person on your team. Think about the Salesforce budget. And the kicker from that same reporting is that roughly 95% of enterprise AI usage runs on the most expensive top-tier models, even for tasks a far cheaper one could do in its sleep. We're not just overusing. We're overusing in a Ferrari just to drive to the corner store. And that's exactly the trap I spent the last year writing about in Between You and AI. See, one of the 9 skills I argue we need in this era is what I call augmentation. And people always assume augmentation means use AI as much as humanly possible. No, it actually means the opposite. Augmentation is automating the routine so you can elevate the human. Not automating the routine so you can manufacture more routine. And token maxing is what augmentation looks like when you forget the second half of the sentence. It's the machine running the human instead of the human running the machine. So here's the message I want to leave you with. For all of human history, the volume of your output was a decent signal of your work. More words, more hours, more output, more value. AI just broke that signal forever. When anyone can generate infinite output at zero cost, the volume of what you produce tells us nothing about you anymore. What tells us everything now is your selectivity. What you choose not to do. The email that you drafted yourself and not through AI. The meeting you killed. The slide you deleted. The token you refused to spend. AI didn't make us addicted to busyness. It just revealed at infinite scale how addicted we already were. Exactly like Seneca saw in a Roman senator 2,000 years ago and exactly like Cal Newport saw in a knowledge worker 10 years ago. The disease was always human. AI just stripped away the last excuse we had for hiding it. So the rarest, most valuable, most competitive skill of the next decade won't be using AI as much as possible. It will be the discipline to use it better. Sometimes that means using it less. Choosing better. Seneca told us life is long enough, generous even, if it's well invested, right? Well, what he couldn't have imagined is a world where investing it well would sometimes mean having the courage to leave the most powerful tool ever built switched off. And with that, I want to thank you for your attention in this episode of Between You and AI. Next week we'll come back with the new one. And in the meantime, if you're more interested about my work or my keynotes, you can check it online at andreaiaro.com, or you can check my book Between You and AI on Amazon and on all platforms. See you next week with yet another episode of Between You and AI.