The Age of Token Abundance
There’s a way of measuring effort that sounds ridiculous at first but gets more reasonable the longer you think about it. The unit is tokens.
I don’t mean cryptocurrency tokens. I mean the tokens that large language models consume and produce — the atomic units of AI computation. Today they’re a billing abstraction. Tomorrow they’ll be the closest thing we have to a universal currency of effort.
Consider the King of England. How much effort goes into his morning? Not the effort he personally expends — the effort expended on his behalf. There’s a team that prepares his schedule. A staff that manages his breakfast. An entire room of physicians monitoring his health. A security detail that has already swept every hallway he’ll walk through. Thousands of person-hours per year are devoted to making one man’s life run smoothly.
Now imagine translating all of that into tokens. His daily briefing: maybe a million tokens of AI labor — gathering, summarizing, prioritizing, formatting. His morning coffee, orchestrated by agents coordinating schedules, dietary requirements, and preferences: another million. By noon, perhaps ten million tokens have been spent on his behalf. His weekly medical checkup, where an AI synthesizes bloodwork, imaging, genomic data, and a century of medical literature to produce a personalized health report — call it ten million more.
Now consider a common man. He shares a doctor with two thousand other people in his municipality. He gets fifteen minutes once a year. If you converted that interaction into tokens, it might be five thousand. Five thousand tokens for your annual health checkup. The king gets ten million for his weekly one.
That ratio — roughly 2,000 to 1 on a weekly basis, 100,000 to 1 annually — isn’t a new observation. We all know inequality exists. What’s new is that we finally have a unit that makes the gap legible across domains. Tokens are a common denominator for care, attention, expertise, and service. And the interesting thing about tokens, unlike doctor-hours, is that their cost is collapsing.
This is the age we’re entering. Not the age of artificial intelligence, exactly. The age of token abundance.
When tokens are expensive, they get rationed the way everything scarce gets rationed: the rich get most of them. When tokens become cheap — really cheap, cheap the way electricity became cheap — something strange happens. Suddenly the common man doesn’t have to share a doctor with two thousand people. He has his own AI doctor, on call every minute of the day, running ten million tokens on his behalf for what it once cost to run five thousand. The gap doesn’t disappear, but the floor rises so dramatically that the floor starts to look like what the ceiling used to be.
This changes things in ways we’re not yet equipped to think about.
Take education. Today a teacher assigns a 500-word essay. In five years, I suspect the assignment will be: write an essay in under 500 tokens. Not because tokens are a gimmick, but because token efficiency will be a genuinely important skill. The ability to get what you want from an AI system with minimal waste — to prompt precisely, to structure your thinking so that it translates efficiently into machine-readable intent — will be as fundamental as writing clearly is today. Maybe more fundamental, because it will be the bottleneck on how much intelligence you can deploy on any given problem.
We already see the early signs of this. Some people burn through millions of tokens fumbling toward basic outputs. Others get extraordinary results from a few hundred well-chosen words. The difference isn’t just technical fluency. It’s a kind of clarity of thought that has always mattered but never had such a direct, measurable payoff.
But here’s the thing I keep coming back to: when tokens are abundant enough, even the inefficient spenders have plenty. And that’s when society gets weird.
I was at a talk recently where a well-known tech founder came to speak at IIT. Afterward, the usual scene: a crowd pressing in for selfies. People jostling to get next to him, holding up phones, hoping for a decent angle. Half the photos come out blurry or awkward.
That entire ritual is going to evaporate. When you can prompt an AI to generate a photorealistic image of you standing next to anyone — your hair perfect, the lighting flattering, the composition exactly what you’d want — why would you elbow through a crowd for a blurry snapshot? The selfie-with-a-celebrity is a social behavior that exists because of scarcity: scarcity of access, scarcity of image quality, scarcity of the celebrity’s time. Remove the scarcity and the behavior dissolves.
This is one example. There will be thousands. Every social behavior that exists primarily because some resource was scarce — and that resource can now be synthesized by spending tokens — is on the extinction list. The question that should be keeping people up at night is: which behaviors are we not seeing yet?
Think about how much of daily life is shaped by rationing. You listen to a podcast made for a million people because it’s not economical to make one just for you. In the age of token abundance, an AI agent spends millions of tokens producing a podcast tailored exactly to your interests, your knowledge level, your preferred style of humor, your commute length. Every piece of media becomes bespoke. The generic version becomes the weird one — like how mass-produced bread would seem weird to someone who’d only ever eaten bread baked for them personally.
I mostly listen to podcasts during my commute.
— Harshil Mathur (@harshilmathur) March 7, 2026
But I kept hitting two problems:
1.I bookmark great long-form articles every day and never get to read them.
2.Spotify podcasts often feel dated or not tailored to exactly what I want.
So built an @openclaw skill.
Every morning…
Or consider decisions. Today you make most small decisions with almost no information, because the cost of gathering information exceeds the value of making a slightly better choice. Which restaurant for lunch. Which route to drive. Which brand of toothpaste. In a token-abundant world, you can throw a million tokens at each of these. An AI that has ingested every review, every traffic pattern, every dental study, gives you a recommendation calibrated to your exact preferences and circumstances. The quality of mundane decisions goes up by an order of magnitude.
At some point this tips over into something qualitatively different. It’s not just that each decision is better. It’s that the texture of life changes when every interaction is attended to with the kind of care that was previously reserved for royalty. That’s what token abundance really means: the democratization of attentiveness.
I should note that this is the optimistic scenario. The one where the technology develops and distributes roughly the way electricity did — unevenly at first, then broadly, lifting the floor for everyone. There is another scenario, a grimmer one, where the abundance flows upward instead of outward, where tokens become another axis of concentration rather than liberation.
I won’t dwell on that scenario. Not because it’s impossible — it’s disturbingly plausible — but because the optimistic version is the one worth building toward, and the one that’s more useful to think about clearly. If you spend all your time staring at the worst case, you tend to build defensively rather than ambitiously, and defensive building has a way of producing exactly the mediocre, captured systems you were trying to avoid.
So here’s the question I’d leave you with. If you accept that tokens are becoming the universal unit of effort, and that their cost is plummeting, what are you doing that will still matter when everyone has a billion tokens to spend? What skills, what instincts, what kinds of judgment become more valuable when intelligence is cheap?
My guess is that it’s something close to taste. Not taste in the aesthetic sense — though that too — but taste in the deeper sense of knowing what’s worth doing. When you can do anything, the hard part is choosing what to do. When you can generate anything, the hard part is knowing what’s good. When every decision can be optimized, the hard part is deciding which decisions matter.
Tokens will be abundant. Judgment won’t be. That’s the arbitrage of the next century.
Notes
This essay explores one likely trajectory among several. The “optimistic scenario” described here assumes broad distribution of AI capabilities and continued cost deflation. The alternative — a concentration scenario — is, as the saying goes, the one we must not name, lest we will it into existence. But it would be dishonest not to acknowledge it exists.
Thanks to conversations with friends at IIT for sparking several of these ideas.