Cheap tokens lead to bigger bills. This paradox sits at the heart of enterprise AI adoption in 2026, and it catches a surprising number of well-run organizations off guard.
Tokens Are the Money That Buys AI Work
Tokens are the fundamental unit of generative AI work - the “money” that buys inference. Every prompt, every response, and every step an agent takes is metered in tokens. Their price and availability are therefore not billing details to reconcile at quarter’s end; they are the cost of getting anything done with AI at all. In that sense, tokens function as the money supply of an AI-enabled organization. Nations manage that dynamic through monetary policy. Organizations now face a strikingly similar problem.
Consider a single engineering team. It begins with a modest pilot: a few hundred prompts, fractions of a cent each, nothing that registers against the budget. Then someone wires the model into an agent that reads the codebase, proposes a change, tests it, and tries again in a loop. The task that cost a few cents as a single prompt now costs a few dollars each time it runs - and multiplied across a dozen engineers running agents all day, the rounding error becomes a line item the finance team asks about. Nothing has gone wrong. The team has simply discovered how much work a token actually buys once AI is doing the work unattended.
Abundance Drives Experimentation
When token supply is abundant and inexpensive, organizations experiment freely. They build agents, run deep research workflows, and leave AI processes running continuously. Adoption accelerates.
This effect is Jevons Paradox: as a resource gets cheaper, total consumption tends to rise because the lower cost suddenly makes it worth applying to work that wasn’t worth it before. You’ll see this idea invoked constantly right now around tokens, agentic workflows, and software jobs - and for good reason. Cheaper tokens don’t shrink the bill; they invite the organization to point AI at problems it would never have funded a year earlier. Agentic workflows routinely consume 10 to 100 times as many tokens as a simple prompt, so total spend climbs even as per-token prices fall.
The pattern is visible in the field. Organizations that roll out agentic coding tools broadly see adoption climb quickly, with the heaviest users consuming hundreds to thousands of dollars of inference per person each month. Some exhaust an annual AI budget in a fraction of the year and are forced to introduce mid-stream controls. These are not failures of the technology. They are failures of token supply management.
Scarcity Slows It Down
The reverse condition produces the reverse behavior. When demand outruns capacity, effective scarcity takes hold in the form of rate limits, overage charges when usage spills past quota, and tighter internal controls. Usage slows, and each decision to apply AI becomes a deliberate allocation of capital rather than a casual experiment.
The instinct at this point is to clamp down with hard caps, approval gates, and renewed skepticism about whether a task truly needs AI. But an organization that throttles too aggressively falls behind the competitor that learned to manage its supply rather than choke it. This does not mean abundant spending is the goal. The objective is neither maximum nor minimum consumption, but deliberate management. Abundance without management leads to waste; scarcity without strategy leads to stagnation.
Token Costs Govern Adoption Speed
For this reason, token costs are best understood as the governor of an organization's adoption speed. When tokens are cheap and plentiful, a team can afford to be wrong a hundred times while experimenting, and that freedom is how high-value uses are discovered quickly. When tokens are expensive or rate-limited, every use becomes a deliberate decision, and deliberation is slow. How an organization manages its token supply sets the ceiling on how fast the whole organization can learn to use AI well.
Key Takeaway
Token costs are not merely an expense line to be minimized. They are the governor of AI adoption velocity. How an organization chooses to manage them is its tokenary policy.
Our next post pushes the analogy further: if tokens are an organization's money supply, then every company is a small sovereign nation that must choose how to manage its currency - and the history of real monetary policy turns out to have a great deal to say about the choices available.
This series - Tokenary Policy:
Post 1: Tokens as the Monetary Supply of the AI Revolution (you are here)