The Next Marketplaces

‍

‍

In 2002, while my team and I were at Verizon Wireless, we launched something called Get It Now! - the first downloadable App Store in the U.S., five years before the iPhone and six years before Apple’s App Store. We did 100 million downloads in our first year, and we introduced the 70/30 business model that still defines digital marketplaces today.

Back then, the idea of buying apps over the air seemed wild. But once developers had a distribution platform, a billing system, and a catalog, they built. And consumers came.

Today, I’m seeing that same shift, but this time it’s not about apps. It’s about agents.

We’re entering the age of intelligent software primitives - and behind them is a quiet but powerful enabler: the MCP server. If you thought APIs unlocked SaaS, MCP will unlock the marketplace for machine intelligence.

Model Context Protocol (MCP), first introduced by Anthropic and extended through projects like MIT’s Nanda, defines how AI models and agents communicate: passing context, delegating work, and responding intelligently to requests. The MCP server is the infrastructure that makes this work at scale. 

In other words: it’s the switchboard for the agentic internet.

Marketplaces of Intelligence

With a standard (MCP) and a router (MCP server), the natural next layer is a marketplace. Not for SaaS apps. Not for plugins. But for intelligent agents; reusable, composable pieces of cognition.

These marketplaces will host:

• Specialized agents (e.g., “Insurance Claim Analyzer,” “Email Summarizer”)
• Vertical-tuned models (e.g., a marketing LLM, an commerce Q&A agent)
• Agent chains and workflows (e.g., “Contract Reviewer → Risk Tagger → CRM Logger”)
• Context Element or mini modeled data objects (which can be a modeled vertex, junction, or a context point)

Just like the early app stores unlocked mobile behavior we hadn’t imagined, MCP-driven marketplaces will unlock workflow-native intelligence - giving developers the ability to plug in agents like Lego bricks.

I recently wrote a piece that touched on the stack going away and being replaced by a loop.  

In this world, a workflow doesn’t call a service; it calls a thinking teammate. SaaS still matters, but the "feature layer" gets eaten by agent marketplaces.

Here’s what’s forming and who is poised to build this:

• MCP Infrastructure: LangChain, DSPy, LangGraph, OSS agent frameworks
• Agent Creators: Domain-specific AI startups, expert fine-tuners
• Marketplace Operators: Curators of reputation, billing, search, and ranking
• Enterprise Builders: Teams that compose vertical workflows from agent parts

We’re back in 2002 - but instead of ringtones and games, the market is for modular cognition.

Creating Pricing Liquidity for Agents

For a marketplace to thrive, pricing needs to flow. This is where MCP servers shine. They can track: agent call volume, token usage, response accuracy or override rates, and latency and availability, among other things. With this, agent marketplaces can enable dynamic, usage-based pricing - where high-performance agents earn more and get surfaced more often.

Think of it as: Twilio’s per-SMS pricing combined with App Store-like discoverability, built on a protocol-native billing layer.

This creates real pricing liquidity for intelligence - where agents are paid like gig workers, and workflows optimize cost and performance automatically.

Why Now?

Well, for starters, MCP protocols are shipping and evolving today. Developers are composing agents with real tools (LangGraph, LlamaIndex, DSPy). Enterprises are seeking modular, cost-effective AI, not black-box monoliths. And lastly, specialized models need marketplaces to find users, not just wrappers.

The conditions are here. And just like in 2002, they won’t wait.

Picture This

You visit a marketplace, pull together:

Contract Uploader → Clause Summarizer Agent → Risk Detector → CRM Note Generator

Each block is intelligent. Each is tracked, paid, and composable. Each gets better with usage. This isn’t an app. It’s a workflow of minds.

The Bottom Line

In 2002, we proved that developers would show up if you gave them tools to distribute and monetize. That same truth applies here.

The fundamentals are in place: MCP, memory, routing, identity, metering. The developers are building. The marketplaces are forming. If APIs gave us composability, and the cloud gave us scale, MCP gives us cognitive liquidity - and the ability to buy, sell, and compose intelligence itself.

It’s still early. But we’ve seen this movie before.

And we know how it ends.

‍