Build Startups for Agents

The internet is splitting in two. Here is the map, and where the money is.

For the longest time, the internet had one user: humans. We searched, read, compared, clicked, and bought. That era is ending.

AI agents are becoming the customer.

Over the next 10 years you get a market of billions of customers, all of them agents, carrying millions of wallets, looking to use your services. Almost nobody is building for them yet.

This is a clear primer on what is changing, and how you can make money from it.

The old web vs the agent web

The old web was built for people.

Humans searching. Reading. Comparing. Clicking. Buying. Your beautiful website fought to win attention.

The agent web runs on different rails.

Agents discover, evaluate, invoke tools, pay, and renew. Your company has to be machine usable.

The shift in one line: a human customer wants persuasion. An agent customer wants structured capability, permission, and trust.

Agents ignore slogans. They run on schemas.

Who these agents are

Everyone gets a personal agent, in their pocket and on the web. Companies get business agents too.

You can already see it forming: Codex, Claude Code, OpenClaw, Hermes.

Agents being everywhere feels inevitable. Soon they outnumber humans online.

The agent buying journey

Map the journey and you understand the customer.

Say the job is "find a payroll tool for 40 contractors." The agent moves through stages:

• Find. It goes looking.
• Evaluate. It reads docs, pricing, APIs, reviews.
• Check trust. It checks policy, checks limits, proves its identity.
• Transact. It pays, books, signs, subscribes.
• Use the tool. It files tickets, changes settings.
• Recommend. It tells other agents what worked.

That last stage is the strange one. Agents will recommend tools to other agents. We saw a glimpse with Moldbook, a social network for agents that Facebook acquired. That was the tip of the iceberg.

Every stage above has missing infrastructure, because the old internet assumed a person was doing the work.

What agents need that a website never gave them

Agents need a whole new stack:

• Identity. Who is this agent acting for?
• Tools. What actions can it safely invoke?
• Inbox. Where do OTPs, docs, and threads land?
• Memory. What does it know about my rules and preferences?
• Wallet. What can it spend, and who approves it?
• Receipts. What did it see, decide, change, and buy?

The wallet works like an employee. As trust grows, you hand over a card, then you raise the limit.

It is already shipping

I needed the examples to make it real. Here are the ones that did it for me.

AgentMail is the email inbox API for AI agents. It gives an agent its own inbox, the way Gmail does for you. Backed by YC. Doing really well. No surprise.

Stripe just shipped a way to give your agent a wallet. Picture a purchasing agent that buys software with spend caps, approval rules, shared payment tokens, and a full audit trail. Expect more fintech moves like this.

Support. A user's agent files the ticket, attaches logs, asks for the refund, follows up, and escalates when ignored.

Procurement. A CFO agent compares 12 vendors, reads their SOC 2 docs, negotiates terms, and recommends the one that fits policy.

MCP servers. A SaaS app hands agents real actions: search customers, create invoices, refund orders, update tickets, pull reports. The agent acts directly instead of scraping your UI.

Local business. A travel agent books dinner, moves the reservation, calls the hotel, pays the deposit, and updates your calendar.

Once you see these, it lands. If agent traffic beats human traffic, of course we build software that serves agents.

Make your site agent-readable

Start with your homepage.

A human-readable homepage has a brand, a video, copy, social proof, a demo, and pricing.

An agent-readable homepage has structured docs, schemas, policies, examples, and endpoints. It exposes MCP tools, SDKs, OAuth, checkout, a sandbox, and receipts.

The rule is simple. Agents only use what they can understand and act on safely. Everything else stays invisible to them.

Build a /agents entry point. Give machines a front door.

The playbook flips

Here is what changes for anyone building a product:

• SEO becomes AEO. Optimize for agents deciding who to cite, trust, and recommend.
• Forms become tool calls. The call to action becomes an action endpoint.
• Support docs become executable support. Agents do the refund, the return, the reschedule, the troubleshooting, the escalation.
• Landing pages become capability manifests. Agents care about what they can do with you, less about the slogan.
• Sales calls become agent procurement. Buyers send agents to a short list before a human appears.
• Analytics becomes agent analytics. Which agents visited? What did they ask? Where did they fail? Where did they bounce?

I once spent years raising conversion rates from 1% to 4% by pairing voice-of-customer surveys with Google Analytics. Now the question keeps me up: how do you optimize conversion for agents?

Startups hiding in the shift

A few to get your ideas moving:

• Agent SEO agency
• Agent identity and permissions
• Agent receipts and audit trails
• Agent-ready docs generators
• Agent inbox security
• Agent-readable pricing pages as a service
• MCP servers for franchises
• Agent support desks
• Sandboxes for agents to test SaaS

The prediction

The internet splits in two. A human internet, and an agent internet.

Take any popular tool you love. Ask what the agent-native version looks like. That question is where the next decade of companies gets built.

Build startups for agents. I want to see what you make.