The Next $100B Market: Selling to AI Agents
Greg Isenberg · 2026-06-02
💡 Quick Take
1. Build startups for agents.
2. Rebuild every SaaS category for agents (payments, communication, memory).
3. Design products to be machine-usable, not just human-readable.
4. Cater to agent needs: structured capability, permission, and trust.
5. Map the agent buying journey: finding, evaluating, transacting, using tools, recommending.
6. Agents will need identity, tools, inboxes, memory, wallets, and receipts.
7. Develop agent-native infrastructure for email inboxes (Agent Mail is an example).
8. Create agent wallets with spend caps, approval rules, and audit trails (Stripe is innovating here).
9. Build agent-native solutions for support, procurement, and business operations.
10. Make your website/app agent-readable with structured docs, schemas, policies, examples, and endpoints.
11. Optimize for AEO (Agent Experience Optimization) instead of just SEO.
12. Shift from forms to tool calls and from support docs to executable support.
13. Replace landing pages with capability manifests and sales calls with agent procurement.
14. Implement agent analytics to understand agent behavior and optimize conversions.
15. Explore startup ideas like agent SEO agencies, identity solutions, receipt/audit trail services, and agent-ready doc generators.
📊 Detailed Explanation
1. Build startups for agents. The core message is that the internet's primary users are shifting from humans to AI agents. This presents a massive, untapped market of billions of agent customers with millions of wallets looking for services. The advice is to pivot your business strategy to serve this emerging demographic.
2. Rebuild every SaaS category for agents (payments, communication, memory). Just like how current SaaS tools are built for humans, future tools need to be "agent-native." This means rethinking how payments are handled (agent-native payments), how agents communicate with each other and with services (agent-native communication), and how agents store and recall information (agent-native memory).
3. Design products to be machine-usable, not just human-readable. The old web focused on human attention through beautiful websites. The new "agent web" requires machine usability. Your products and services need to be accessible and understandable by AI agents, allowing them to discover, evaluate, invoke tools, pay, and renew without human intervention.
4. Cater to agent needs: structured capability, permission, and trust. While human customers respond to persuasion, agents require clear, structured information. They need to know what capabilities your service offers, what permissions they have, and they need to trust your system. This means providing unambiguous APIs, clear access controls, and robust security.
5. Map the agent buying journey: finding, evaluating, transacting, using tools, recommending. Understanding how agents will interact with your business is crucial. They'll find services (e.g., "find a payroll tool for 40 contractors"), evaluate them (reading docs, pricing, APIs), transact (pay, book, subscribe), use tools within your service, and even recommend other tools to other agents. You need to build infrastructure that supports each of these stages.
6. Agents will need identity, tools, inboxes, memory, wallets, and receipts. These are fundamental requirements for agents that humans don't typically need in the same way. Identity ensures accountability, tools allow them to perform actions, inboxes are for communication, memory helps them retain context, wallets manage spending, and receipts provide an audit trail of their actions.
7. Develop agent-native infrastructure for email inboxes (Agent Mail is an example). Agent Mail is highlighted as a prime example of building essential infrastructure for agents. It provides AI agents with their own email inboxes, acting as the "email inbox API for AI agents." This shows a clear need for specialized communication channels for agents.
8. Create agent wallets with spend caps, approval rules, and audit trails (Stripe is innovating here). Fintech is a key area for agent adoption. Imagine purchasing agents that can buy software autonomously but within defined limits. Stripe's innovation in giving agents wallets with spend caps and approval rules is a significant step towards enabling agent-driven transactions securely.
9. Build agent-native solutions for support, procurement, and business operations. The transcript provides concrete examples: a support agent can file tickets, request refunds, and follow up; a CFO agent can compare vendors and negotiate terms; a SAS app can give agents tools to search customers, create invoices, and update tickets without UI scraping.
10. Make your website/app agent-readable with structured docs, schemas, policies, examples, and endpoints. To be discoverable and usable by agents, your online presence needs to be machine-readable. This means going beyond human-friendly content to include structured data, clear API documentation, defined schemas, security policies, and accessible endpoints. Think of a `/agents` entry point on your website.
11. Optimize for AEO (Agent Experience Optimization) instead of just SEO. Just as SEO helped human users find websites, AEO will help agents find and trust services. Agents will rely on other agents and AI search tools to discover what they need. Being discoverable and trustworthy in this new landscape is paramount.
12. Shift from forms to tool calls and from support docs to executable support. The user interface for agents is different. Instead of filling out forms, they'll make direct tool calls. Instead of reading support documentation, they'll be able to execute support actions like refunds or reschedules directly through your service.
13. Replace landing pages with capability manifests and sales calls with agent procurement. Agents are less interested in marketing fluff and more in what your service can *do*. Capability manifests clearly outline what actions agents can perform. Similarly, procurement processes will increasingly involve agents evaluating and shortlisting vendors before human involvement.
14. Implement agent analytics to understand agent behavior and optimize conversions. Just as businesses track human user behavior, they'll need to track agent interactions. This includes knowing which agents visited, what they asked for, where they encountered issues, and why they left. This data is crucial for optimizing the agent experience and conversion rates.
15. Explore startup ideas like agent SEO agencies, identity solutions, receipt/audit trail services, and agent-ready doc generators. The transcript explicitly lists several promising startup opportunities: agencies specializing in AEO, technologies for agent identity and permissions, services for agent receipts and audit trails, tools to automatically generate agent-readable documentation, and secure agent inboxes.
🎯 Tech Expert Opinion
Wow, this is exactly the kind of paradigm shift that gets me excited! The idea that AI agents are becoming the primary internet users isn't just a trend; it's a fundamental rewrite of how we interact with the digital world. The transcript nails it: we're moving from a human-centric internet to a machine-to-machine economy, and the infrastructure for this is largely missing.
The concept of "agent-native" is key here. We've spent decades optimizing for human cognition, intuition, and even our biases. Now, we need to build for logic, structured data, and deterministic actions. Think about it: a human might browse a website for an hour, get distracted, and come back. An agent will execute a task, hit an API, get a response, and move on. This requires a completely different approach to UX/UI, which for agents, is really about API design, data schemas, and robust error handling.
The opportunities in building infrastructure are immense. We're talking about the foundational layers of this new internet. Agent identity and secure wallets are huge. Imagine a world where your personal AI agent can securely manage all your subscriptions, make purchases within your budget, and even negotiate contracts on your behalf. This isn't science fiction anymore; it's the next logical step, and companies like Stripe are already paving the way.
The shift from SEO to AEO is also fascinating. It means that not only do you need to be discoverable, but your service needs to be *understandable* and *actionable* by AI. This will push companies to create incredibly well-documented APIs, clear capability manifests, and perhaps even standardized agent interaction protocols. I predict we'll see a rise in "API-first" product development, where the API is treated as the primary product, with a human-friendly interface as a secondary layer.
The idea of agent analytics is particularly compelling. For years, we've been trying to understand the "voice of the customer" through surveys and indirect metrics. Agent analytics will provide direct, objective data on how agents interact with our systems. This will allow for hyper-optimization of workflows, identification of friction points that humans might miss, and ultimately, a more efficient and effective digital ecosystem.
Looking ahead, I see a bifurcation of the internet. There will still be a human internet, catering to our creative, emotional, and social needs. But running parallel will be the agent internet, a highly efficient, automated layer focused on task completion, data processing, and transactions. The companies that successfully bridge these two worlds, or build robustly within the agent internet, are going to be the giants of the next decade. The transcript's call to action to "build for agents" is not just a suggestion; it's a strategic imperative for survival and growth.
Kanal: Greg Isenberg