What an agent costs to run, how it might earn, and who could capture the value
Personal AI agents that book, buy, research, write, and operate other software on a person's behalf moved from demos to products in 2025 and 2026, faster than anyone settled how they make money. They also break the arithmetic of the previous internet: search and social media served each extra user at almost no cost, so advertising could pay for everything, while an agent that reads dozens of pages, calls tools, and checks its own work spends real compute on every task. Across ten modules and 40 sessions, this collection works through the economics from the ground up: the capital stack underneath, what serving an agent actually costs, the price of everyday tasks, and the candidate revenue streams of subscriptions, usage pricing, commerce, advertising, enterprise, and devices. It then compares agents with the ad-funded web, social networks, utilities, and professional services, and lays out competing scenarios for who captures the value. Prices change fast, so every figure is a dated estimate and the method matters more than the number. The aim is a reader who can estimate what an agent task costs, judge whether a pricing model holds up, and test any claim about who will win.
Software that acts instead of answering, the end of near-zero marginal cost, and a market best grouped by strategy.
From chips to applications: where the money is invested, where revenue flows, and why the price of intelligence keeps falling.
Tokens, tool calls, infrastructure, and failures: the full bill behind one completed task.
A cost ladder from a quick answer to a multi-hour project, and the gap between what a task costs and what it is worth.
Monthly tiers, rate limits, bundles, and the open question of how many people will pay.
Agents that buy, the fees and ads that might fund them, and what happens to a web built for human visitors.
Agents at work, models on the device, agent marketplaces, and the value of accumulated personal context.
Agents set beside ad-funded search, social networks, utilities, and professional services.
Margins across the stack, distribution as the decisive asset, open weights, and the scenarios still in play.
Unit economics on one page, the open risks, the wider effects, and a toolkit for judging any agent forecast.
“A clear, even-handed ledger for the business of AI that acts on your behalf.”
It starts from what a task costs to serve and builds up to revenue, market structure, and scenarios. Bull and bear cases sit side by side, every price is dated, and the method lets readers redo the math as costs change, which matters in a market moving this quickly.
Search and social served each new user for almost nothing; an agent spends real compute every time it works.
Advertising could fund the earlier internet because the marginal cost of one more search or feed refresh was tiny. An agent that reads pages, calls tools, and checks its own output consumes compute on every task, so its business model has to cover a cost that rises with use.
Econ Beyond is a LearningFirst channel that explains the economics behind everyday and emerging systems for general readers — the prices, incentives, and trade-offs hidden inside things people use without a second thought. This collection applies that lens to personal AI agents: software that acts on a person's behalf, and the business still being built underneath it.
Ranked by semantic similarity
The smallest billable unit of machine cognition, and the economy being built on it
Free inboxes, business suites, senders, gatekeepers, and the money that keeps email running
The portable toolkit for seeing what a choice really costs and what happens next
Idols, superstars, founders, and collectibles, read as one economic system
