Sign in Subscribe

The AI Agent Gold Rush Is Here. Are You Selling Picks or Digging?

The AI Agent Gold Rush Is Here. Are You Selling Picks or Digging?

In the last week:

  • Meta acquired Moltbook, a social network for AI agents
  • AgentMail raised $6M to give AI agents email inboxes
  • Lyzr AI hit $250M valuation building enterprise agent infrastructure
  • OpenAI acquired Promptfoo to secure AI agent outputs

Something is happening. Most people are looking at it wrong.

Everyone Wants to Build the Agent

The default startup pitch in 2026: "We're building an AI agent that does [X]. It's like having a [job title] that works 24/7."

Cool. You and 10,000 other teams.

Here's the problem: agents are commoditizing faster than anyone expected. Claude, GPT, Gemini—they all become "agents" with a weekend of prompt engineering and some API glue.

The moat isn't the agent. The moat is everything around it.

The Real Money Is in the Plumbing

Think about the internet gold rush. The winners weren't websites. They were AWS, Stripe, Cloudflare, Twilio.

The pattern is repeating.

AI agents need:

  • Identity — How does an agent prove it's authorized to act?
  • Communication — Agents talking to agents, humans, APIs
  • Memory — Where does context live? Who controls the data?
  • Trust — How do you verify what an agent actually did?

That last one—systems of record—is the sleeper. Every agent needs a source of truth. CRM for agents. ERP for agents. The companies that become the canonical data layer for AI workflows will be massive.

Meta didn't buy Moltbook because they care about bots chatting. They bought it because agent-to-agent communication is the next protocol layer. Whoever owns that owns the network effects.

The Unsexy Billion-Dollar Bets

AgentMail raised $6M to give AI agents email addresses.

Sounds ridiculous until you realize: every AI agent operating in the real world needs to send emails, receive confirmations, handle bounces, manage threads. There's no "Log in with Google" for AI. No OAuth. No established identity layer.

This is infrastructure. It's boring. It's going to be everywhere in 18 months.

The same logic applies to:

  • Agent auth (who is this agent acting on behalf of?)
  • Agent observability (what did this agent actually do?)
  • Agent-to-agent protocols (how do agents coordinate?)

None of this is sexy. All of it is essential.

What This Means for You

If you're a founder: Ask yourself—what happens when GPT-6 makes your agent obsolete overnight? If the answer is "we're screwed," you're building the wrong thing.

Build proprietary data loops. Build network effects. Build systems of record. The model is a commodity. The context isn't.

If you're an investor: I'm more excited about agent infrastructure than agents themselves. The API layer for AI agents is being written right now. Most people don't see it because they're too busy arguing about which chatbot is smarter.

The infrastructure companies will be the Stripes and Twilios of the AI era. They'll be boring. They'll be everywhere. They'll be worth tens of billions.

The AI agent gold rush is real.

But like every gold rush, the winners won't be the miners.

They'll be the ones selling mining equipment.

Your move,

Blaine