The internet’s dominant revenue model assumes human attention. What happens when machines do the browsing?

The attention economy’s blind spot

Digital advertising is a $600+ billion industry built on a simple premise: humans look at screens, and you can put messages in front of them. Every impression, click, and conversion assumes a person on the other end—someone who might notice your brand, develop preference, and eventually buy.

AI agents Don’t notice brands. They don’t develop preferences. They execute tasks.

When an agent searches for flights on your behalf, it doesn’t see the banner ad for hotel deals. When it compares products across retailers, it doesn’t register the Sponsored placement. When it reads articles to summarize for you, it skips right past the display inventory.

This isn’t a minor optimization problem. It’s a potential unraveling of the economic foundation that funds most of the free web.

The math gets uncomfortable

Consider a content publisher whose revenue depends on advertising. Today, a human visitor might view five pages, generating five ad impressions worth fractions of a cent each. Multiply by millions of visitors, and you have a business.

Now imagine half those “visitors” are agents gathering information for humans who never visit directly. The agent extracts the content, summarizes it, and delivers the summary to the user. The publisher’s content created value—but the publisher captured none of it. No impression, no click, no revenue.

This is already happening. AI systems trained on web content monetize that content without compensating creators. AI search tools that synthesize answers reduce click-through to original sources. The value chain is breaking.

For ad-supported businesses—which includes most media, many platforms, and countless free tools and services—the agentic web poses an existential question: if agents intermediate between users and content, who pays for the content?

The scramble for new models

Forward-thinking organizations are experimenting with alternatives.

Licensing and partnerships. Some publishers are striking deals with AI companies—getting paid for training data or for real-time content access. The new york times and openai’s legal battle, and subsequent Licensing discussions Across the industry, signal that content has value that someone should pay for. But these deals favor large publishers with negotiating leverage. Smaller creators may find themselves squeezed out.

API monetization. If agents are going to access your service anyway, you might as well make it official—and charge for it. Structured API access with usage-based pricing gives agents what they need (reliable, fast data) while capturing value for the provider. The shift from ad-supported websites to paid API access could reshape which services survive.

Transaction fees. When agents complete purchases, there’s a transaction to monetize. The service that enabled the purchase can take a cut. This works for commerce but doesn’t help content creators whose value is informational rather than transactional.

Agent licensing. Emerging protocols like tollbit enable websites to charge AI agents for access—essentially a toll road model for the agentic web. Agents pay small fees to access content, with those fees flowing to creators. The infrastructure is nascent, but the concept addresses the core problem: making agents pay for value they extract.

Premium human experiences. Some businesses may double down on serving humans directly, creating experiences that can’t be intermediated. The agent can gather information, but the human still wants to visit—for the experience, the community, the serendipity that agents can’t replicate.

The infrastructure plays

When business models shift, infrastructure providers often capture disproportionate value. The agentic web is no different.

Companies building Agent authentication, traffic management, and monetization layers are positioning for this transition. Akamai’s bot management now includes “know your agent” capabilities that identify and potentially monetize agent traffic differently from human traffic. DataDome and others offer tools to distinguish between agents and humans, enabling differentiated treatment.

The platforms that become default infrastructure for agent-to-service interactions will extract rent from every transaction that flows through them. MCP’s emergence as a standard is partly a play for this position—if your protocol mediates agent interactions, you shape how value flows.

Who wins, who loses

The transition won’t affect all businesses equally.

Clear winners: Services with direct transaction revenue (e-commerce, bookings, financial services) can monetize agent-driven purchases the same way they monetize human purchases. API-first businesses are already positioned for agent access. Infrastructure providers enabling the agentic web will capture new revenue streams.

Complicated middle: Subscription businesses may find agents increase the value of subscriptions (your agent uses your Netflix account to find shows you’ll like) or decrease it (why subscribe when an agent can summarize content from free sources?). The outcome depends on how defensible the underlying value is.

Clear losers: Ad-dependent businesses Without alternative revenue streams face the hardest transition. If your model depends on human attention and agents reduce human attention, you need a new model—fast.

What to do now

For businesses navigating this shift:

Understand your exposure. What percentage of your value depends on human attention versus human transactions? Attention-dependent value is at risk; transaction-dependent value may be more resilient.

Experiment with agent monetization. Even if agent traffic is small today, start testing: API access tiers, agent authentication, usage-based pricing. Build the infrastructure before you need it.

Diversify revenue streams. If you’re fully ad-dependent, that’s a strategic vulnerability. Develop subscription offerings, premium services, transaction-based revenue—anything that doesn’t require human eyeballs.

Track agent traffic separately. You can’t manage what you don’t measure. Instrument your analytics to distinguish agent interactions from human visits. Understand the trend before it overwhelms you.

Engage with emerging standards. Licensing frameworks, agent payment protocols, and traffic management standards are being built now. Participating in their development is better than having them imposed on you later.

The advertising-supported web isn’t disappearing overnight. But its growth era may be ending, and its dominance is no longer guaranteed. The businesses that recognize this early have time to adapt. Those that don’t may find their models obsolete faster than they expected.