Mental BoundTokens: the new utility bill
As AI becomes infrastructure, token consumption becomes the metered cost of intelligence — a new line item with new divides.
Your electricity bill measures kilowatt-hours. Your water bill tracks gallons. Your intelligence bill counts tokens. Every transformative utility follows the same arc: factories once ran their own power plants until the grid made electricity something you plug into; businesses once racked their own servers until cloud made compute something you rent. AI is at that inflection now — the "on-premise" equivalent was hiring analysts, researchers, and specialists; the grid equivalent is metered intelligence delivered through an API.
Strip the jargon and a token is a unit of thought: roughly 750 words per thousand tokens, counted in and out. A short email costs a couple hundred; a full market research report can run fifty thousand. What matters for business is that tokens are predictable, measurable, and instantly scalable — no hiring, no training, no severance. They also don't understand nuance without careful prompting and won't question a flawed assumption, which is exactly why the point isn't replacement; it's that intelligence becomes a resource you can turn on and off like a faucet.
When intelligence meters, the economics move. Enterprises get an "intelligence spend" line item that CFOs forecast like cloud costs, with budgeting shifting from headcount planning to usage prediction. Startups get the barriers knocked down — a solo founder ships AI features that needed a twenty-person team five years ago. Individuals get access to drafted contracts, financial analysis, and tutoring — at a price, which is the uncomfortable part. Utilities create access divides, and this one decides who can afford to be augmented. A competitor spending $50,000 a month on token-powered research and automation doesn't just outspend you; it out-thinks you. Rural electrification needed policy intervention; the intelligence divide likely will too.
For builders, tokens become a design constraint like bandwidth: optimize prompts the way you optimize payloads, design pricing for metered usage rather than "unlimited," instrument token observability next to performance monitoring, and route work across hybrid model tiers — small and cheap for routine tasks, expensive and capable when quality demands it.
Every utility was a luxury, then a convenience, then a necessity. Today AI feels optional. Soon, having no token budget will feel like having no internet access.