The Next $100B Market: Selling To AI Agents

AI agents are replacing human beings as the primary end users and customers of software and services. Where the old web was built around humans searching, reading, comparing, clicking, and buying, the emerging agent web involves agents discovering, evaluating, invoking tools, paying, and renewing autonomously. Human customers want persuasion while agent customers want structured capability, permission, and trust. The speakers frame this as the foundational shift driving the next hundred-billion-dollar market.

The scale prediction is significant. Over the next ten years, the speakers expect billions of agent customers with millions of wallets to emerge, with agent traffic eventually outnumbering human traffic on the internet. Personal and business agents will be everywhere. Despite this trajectory, the speakers claim almost nobody is currently building for the machine-to-machine economy, which they treat as a large and largely uncontested gap.

Existing internet infrastructure was built assuming a person was doing the work, and it fails agents in specific ways. Agents need identity, an inbox, memory, spending capability, and receipts, and current infrastructure does not reliably provide any of these. The speakers argue that every SaaS category, including tools like Notion, Slack, and Stripe, will need to be rebuilt in agent-native versions. Stripe recently launched a mechanism to give agents a wallet, and Agentmail, an email inbox API for AI agents backed by Y Combinator, is cited as an early example of this infrastructure forming and reportedly performing well.

A website that agents cannot understand or safely interact with is effectively invisible to them. The speakers argue that businesses need an agent-readable layer covering documentation, schemas, policies, examples, endpoints, MCP tools, SDKs, OAuth, checkout, sandbox environments, and receipts. A slash-agents entry point is described as a significant discoverability opportunity. Rather than forms, sites should offer tool calls as action endpoints. Rather than support documentation, they should offer executable support where agents can perform refunds, returns, rescheduling, and escalation directly. Rather than landing pages, businesses should publish capability manifests. Rather than sales calls, they should build agent procurement processes where buyers send agents to shortlist vendors before a human is ever involved.

The speakers identify a wide range of startup opportunities inside this shift. Agent SEO agencies, agent identity and permissions technology, agent receipts and audit trails, agent-ready documentation generators, agent inbox security, agent-readable pricing pages as a service, MCP servers built for franchise businesses, agent support desks, and sandboxes that let agents test SaaS products before purchase are all named as viable categories. The underlying logic is that every layer of the existing internet stack has an agent-native equivalent waiting to be built, and most of those equivalents do not yet exist.

The long-run structural prediction is that the internet bifurcates into a human internet and an agent internet. Businesses that build only for human visitors will lose discoverability and revenue as agent-driven procurement and usage grows. The speakers treat this not as a distant scenario but as something already beginning, with the current moment representing the earliest window to establish position before the infrastructure layer becomes crowded or consolidated.