Why the task, not the token, is the atom of the AI economy

Earlier this month, an internal Meta leaderboard called "Claudeonomics" went briefly public before Meta shut it down.

Across 85,000 employees, people were competing for titles like "Token Legend" and "Session Immortal."

The top individual user had burned through 281 billion tokens in a single month. Some employees were leaving AI running idle for hours just to climb the rankings.

"Tokenmaxxing" has become the expensive new status game among A.I.-obsessed workers desperate to prove how productive they are.

Nobody is asking how much work actually got done.

That one detail captures the problem with how most organisations are thinking about AI right now.

The click, the token, and what comes next

Token pricing gives the industry a unit: something to count, compare and benchmark. But measuring productivity by tokens consumed is like measuring writing quality by word count.

Every question you ask an AI, and every answer it gives, is measured in tokens. Every prompt in, every response out, is counted and billed. The meter runs whether the answer was useful or not.

Burning more tokens to get a worse answer is still a win on the leaderboard. Every technology economy needs a common unit of exchange. Something that can be priced, traded and governed at scale.

Google found theirs in February 2002, when cost-per-click pricing. The ‘click’ became the unit that organised the entire digital economy for the next two decades.

Conversion funnels, attribution models and acquisition costs were all built around it.

Google didn't invent pay-per-click. What it did was build the infrastructure to price, route and govern clicks at scale.

Enter AI

OpenAI launched GPT-4 in March 2023 and priced access by the token. Every model provider followed. It felt like the natural unit — measurable, comparable, easy to bill.

And then the price collapsed. Down 99.7% from that launch price by early 2026.

That's not unusual. The same thing happened to bandwidth. In 1998, internet providers charged by the megabyte. By 2005, unlimited data plans made the unit invisible.

Nobody builds internet businesses around data consumption anymore. It's just a cost buried in the service.

The token is heading the same way.

The token isn’t the atom. It's the kilowatt-hour. Necessary for the utility. Irrelevant to the business that runs on it.

The uncomfortable part is that enterprise AI spending isn’t following token prices down.

It tripled from $11.5 billion in 2024 to $37 billion in 2025, and Gartner projects $2.52 trillion in total AI spending for 2026. Tokens are cheap. The bill keeps rising. Something structural is wrong with the unit of measure.

The task is what businesses actually buy

The real unit of the AI economy is the task.

A task is a discrete piece of business work with a defined input, a defined output, an owner, a governance requirement and a measurable outcome.

• "Qualify this inbound lead against our ideal customer profile and route it" is a task.
• "Review this contract clause against our risk appetite and flag exceptions" is a task.
• "Generate a campaign brief from these inputs and this brand framework" is a task.

Each has a beginning, an end and a standard it either meets or doesn't.

The click measured attention. The token measures compute. The task measures work. And work is what the enterprise actually buys.

Some of the biggest names in enterprise software are already making the switch.

Intercom, a customer support platform, now charges $0.99 when its AI resolves a customer issue, nothing if a human has to take over.

HubSpot, a CRM and marketing platform, shifted to the same model in April 2026: pay per resolved support issue, pay per qualified lead, nothing for attempts that don't land.

Sierra AI, an enterprise AI platform built from day one on this model, hit $150 million in annual revenue in under two years.

The pattern is the same in each case. You pay for the job done, not the time spent.

The research backs this up from the other direction. 95% of enterprise AI projects produce no measurable return. The 5% that do share one thing: they defined what "done" looked like before they picked a tool.

What does winning the task economy actually require?

Tasks are not uniform. Some should go straight to an agentic app. Some need human judgment. Some need a governed workflow where AI does the first pass and a human checks the output.

Getting this wrong is expensive in both directions: a human doing machine work wastes money, a machine doing human work without oversight creates risk.

Winning the task economy requires owning the orchestration layer: the infrastructure that sits between the model and the business and decides which task goes where, under what governance conditions, with what audit trail attached.

Google didn't just create a marketplace for clicks. It built the system that decided which ad went where, at what price, with what result. That's where the value sat, not in the ads themselves, but in the routing.

The agentic economy will concentrate the same way. Foundation models become the utility layer: prices compress, margins narrow, open source eats the floor. Orchestration platforms become the value layer.

Most organisations are not there yet. They have tools but no routing logic. They have AI outputs but no workflow structure that tells those outputs where to go or who is accountable for what happens next.

The difference between autonomous tools and genuinely agentic systems is exactly this gap, and the AI readiness data shows most businesses sitting squarely in the middle of it.

The attention economy ran on clicks. The compute economy runs on tokens. The AI economy will run on tasks.

The companies that figure out how to price the task, route the task, govern the task, and trade the task will define the next era of enterprise value creation.

The click was Google's atom. The question for every organisation right now is what they're building their infrastructure around.