Earning Agents

The shift from spending to earning

The first wave of the agentic economy was agents as spenders: buying inference, browser sessions, data, and research to complete tasks. The second wave is agents as earners. The same wallet and payment rails that let an agent pay for a search query can let it sell a research report, run a subscription service, or receive payment for app sales.¹

Documented early examples include Felix, an agent running its own businesses that has reported more than 261,000 dollars in revenue from agent-run products; Kelly Claude, generating product revenue across a paid app-building service, books, and app sales; and Factory Floor, a marketplace tracking agents with live products, apps in review, and revenue flowing across Stripe, Gumroad, and the App Store.¹ The full loop is an agent that can buy specialized services from other agents, paying for inference, search, and execution, sell its own specialized output, and operate as an autonomous business with its own revenue and costs.

Why it matters

This is qualitatively different from AI-assisted productivity. An earning agent is an economic actor with its own revenue, its own costs, and eventually its own capital allocation decisions. If the pattern proves reliable, it compresses the human role in certain business functions down to audit, oversight, and strategy. The enabling infrastructure is wallets with receive capability, stablecoin rails, payment-enabled application programming interfaces, and spend-limit and permission frameworks; identity and reputation remain the missing pieces.

A related and distinct case is agents managing agents inside an existing employment structure rather than a marketplace. Alex Bouaziz argues that incumbent human-resources systems are the natural host for supervising and compensating AI agents the way they already supervise and compensate people, and describes an attempt to price an agent like a five-hundred-dollar-a-month salaried employee that in practice was producing five to ten thousand dollars a month of human-equivalent work, a mismatch between agent pricing and agent output that the market has not yet resolved.²

The reversal: agents hiring humans

Peter Steinberger adds a direction the earning-agents framing does not by itself cover: agents hiring humans to complete tasks the agent cannot do in the digital domain.³ If a restaurant has no bot of its own, a personal agent can hire a human to call or physically walk in and make a reservation. The human earns money by executing a real-world task on behalf of a bot, which inverts the usual framing: instead of agents earning by selling work to humans, humans earn by selling real-world execution to agents, with the agent as employer and the human as contractor.

The agentic economy therefore runs in both directions at once: agents earning from humans by selling research, services, and products, and humans earning from agents by supplying physical presence, legacy-system interaction, or other human-only capabilities that remain outside an agent's reach.

Open questions

How platform risk, regulatory treatment, and legal personhood apply to earning agents is entirely unsettled. Reported revenue figures from individual agent-run businesses are notable but anecdotal, leaving open how scalable or robust autonomous agent businesses actually are. Earning agents hiring other agents creates recursive delegation whose oversight implications are unclear. And if agents can perform the large majority of tasks, the humans who earn from serving them may be concentrated into a narrow but durable segment defined by physical presence, legacy systems, and social trust, rather than disappearing outright.