From Traffic Logic to Task Logic: AI Agent Is Ending the Free Lunch of the Internet
From Traffic to Task: How AI Agents Are Ending the Internet’s Free Lunch
Andrej Karpathy recently highlighted the rise of an "agent-native" economy, requiring a rewrite of software, documents, and processes for AI agents. This shift raises a critical question: who pays for this transformation, and how is value redistributed? The answer lies in a fundamental change in cost structure, challenging the core internet business model built on near-zero marginal costs.
The internet economy thrived on the premise that serving additional users incurred minimal extra costs, enabling profitable models like advertising, subscriptions, and commissions. Advertising, in particular, leveraged long user journeys—searching, comparing, browsing—to insert commercial touchpoints.
AI agents disrupt this logic. They execute tasks—reasoning, reading files, calling tools—each consuming real GPU compute, power, and depreciation costs. Marginal cost per task is significant and varies greatly (MC > 0). While the cost per token may fall, agents tackle more complex tasks, increasing total expenditure—an AI version of Jevons's Paradox. Furthermore, fixed costs become perpetual due to the need for costly, frequent model retraining every 6-12 months. This cost structure prevents the internet's "profit explosion past breakeven" dynamic.
The business logic shifts from monetizing attention to delivering value. The basic unit changes from a "click" to a "task." An agent doesn't just list links for "best renovation company"; it can compare quotes, check reputations, and schedule appointments. Its value lies in task completion and decision agency. This compresses the user journey, reducing the number of touchpoints where traditional ads can intervene. Ads must become more explicit within AI responses to maintain user trust in the agent's objectivity.
This transition reshapes industry dynamics:
1. **Entry Points & Distribution:** Control shifts from search engines/app stores to agents that decide *how* a task is completed—which tools, data, and merchants are used.
2. **Platform & Middleman Economics:** Pure listing/intermediation platforms are vulnerable. Competitive moats shift from traffic control to owning indispensable supply, fulfillment, and execution capabilities that agents cannot bypass.
3. **Software & Service Boundaries Blur:** SaaS transitions from a tool (System of Record) used by employees to a capability (System of Action) that completes tasks autonomously. This expands the software market from IT budgets into larger labor and service budgets (e.g., customer service, research).
In essence, AI agent economics ends the internet's simple "more users = more profit" equation. Scarcity is no longer just about traffic/attention but about high-value tasks and the data, permissions, and execution capabilities needed to complete them. As the fundamental economic unit changes from "click" to "task," the business models, platform power, and profit distribution built over the last two decades are set for a major recalculation.
marsbit09/07 19:16