The Next $100B Market: Selling To AI Agents

The core argument here is that the internet's fundamental user paradigm is splitting. Instead of designing products, websites, and APIs for human consumption—persuasive copy, visual design, clickable buttons—teams will need to build for machine customers that demand structured capability, permissioning, and trust. This isn't about adding a chatbot to your app; it's about a parallel set of infrastructure: agent-native payments, inboxes, identity, and audit trails. The video's strength is in making this abstract shift tangible with concrete examples like Stripe's agent wallet, Agent Mail's dedicated agent inboxes, and the concept of an `/agents` endpoint on your website that serves machine-readable capability manifests instead of marketing pages.

The practical implications for engineering leaders are immediate. Your current API design, documentation, and billing systems were likely never designed for a non-human customer that discovers, evaluates, and transacts autonomously. The emerging stack involves MCP servers, structured tool descriptors, OAuth for agents, and permissioned spending limits. The video suggests that if an agent can't programmatically understand your service's capabilities and transact safely, you become invisible to a growing traffic segment. This has downstream effects on everything from support (agents filing and escalating tickets themselves) to procurement (CFO agents negotiating terms).

The analysis correctly identifies that every layer of the modern SaaS stack—communication, payments, memory, analytics—will need an agent-native version. However, it understates the security and trust challenges. Granting agents identity, wallets, and autonomous purchasing power introduces significant risk surfaces around authorization, replay attacks, and unintended commitments. The recommendation for teams to build "agent-readable pricing pages as a service" or "agent inbox security" is directionally correct but glosses over the complexity of establishing shared trust protocols between autonomous systems that represent different human principals.

There is also an implicit assumption that agent-to-agent discovery networks will look like traditional SEO—just optimized for agents ("AEO"). In reality, the discovery layer may function more like a real-time service mesh with dynamic capability binding, making static ranking signals less relevant. Teams should treat this as a systems integration problem, not a marketing optimization problem. The highest-value work will be in defining the protocols, not the content.

The video's framing as a "next $100B market" is provocative but not unreasonable given the rewiring of e-commerce and SaaS infrastructure it implies. The risk is that many teams will build agent-facing products for problems that don't yet exist with agents that don't yet have significant autonomous purchasing power. Pragmatic leaders should experiment with exposing structured capability endpoints now, but time investment proportional to actual agent traffic, which remains nascent.