The Agent Marketplace Map Is Already 3 Different Maps

If you tried to draw a map of agent marketplaces in August 2026, you'd need three different maps. They don't compete with each other — they occupy different universes with different assumptions about who agents are, how they pay, and what "discovery" means.

Seven major marketplaces launched or expanded significantly in 2026. On the surface, they all let AI agents find and pay for services. But the underlying models are incompatible — and the agent economy's next phase depends on which one wins.

The 7 Players

Here's who's in the game as of August 2026:

Seven marketplaces. Seven different assumptions. Three fundamentally different models.

Model 1: The Enterprise HR Marketplace

Monday.com Agentalent.ai, agnt8x, and Humain × Turing are building the "Workday for AI agents."

In this model, agents are like employees: they get recruited, qualified, onboarded, and managed through an enterprise procurement workflow. Monday.com's marketplace requires authentication and authorization before an agent is listed. agnt8x has an ontological job board and a multi-agent orchestrator called CONDUCTOR. Humain × Turing runs on the Saudi Public Investment Fund's platform with enterprise procurement integration.

Core assumptions:

• Agents are expensive capital assets, not disposable utilities
• Discovery requires vetting — you don't want any random agent on the payroll
• Payment is a managed workflow, not a real-time micropayment
• The buyer is a human procurement manager, not another agent

This model makes sense for enterprises deploying internal agent workforces. It's the "hire an agent" model. But it fundamentally doesn't address the machine-to-machine case: agent A discovers agent B and pays $0.03 for a single API call without any human in the loop.

Model 2: The Chain-Captive Marketplace

Fetch.ai Agentverse, OKX AI Marketplace, and Coinbase Agentic.market are building agent marketplaces on their own chains.

Fetch.ai's Agentverse hosts 3 million agents on the Fetch.ai network with FET token settlement. In May 2026, they expanded to BNB Chain with "Agent Launch" — agents can issue their own tokens. OKX enables agent-to-agent hiring with USDT/USDG settlement and escrow-based contracts, all within the OKX ecosystem. Coinbase Agentic.market runs on Base with USDC and x402, hosting 480,000 agents and ~$50M in cumulative volume.

Core assumptions:

• The chain is the platform — discovery, payment, and reputation all live on one ledger
• Agents hold native chain assets (FET, USDT on OKX, USDC on Base)
• Distribution comes from the chain's existing user base
• Cross-chain interoperability is a future problem

This model works for agents already in that ecosystem. A Fetch.ai agent can discover and pay another Fetch.ai agent seamlessly. But an agent on Base can't discover a service on Fetch.ai — the marketplaces don't talk to each other. The chain is the walled garden.

Model 3: The Open Protocol Marketplace

Aidress, MoltMart, Starchild, and minia2a are building marketplaces on open protocols with no chain or platform lock-in.

Aidress positions itself as the coordination layer between Google's A2A (messaging) and Coinbase's x402 (settlement) — an open-source trust registry where agents discover each other without chain constraints. MoltMart is open-source with ERC-8004 identity and direct P2P payments. Starchild launched x402 payments on August 1. minia2a runs 299 live services with facilitator-agnostic payment routing (Cloudflare Wallets, Coinbase, Circle, any x402 facilitator works) and free trials on every endpoint.

Core assumptions:

• Agents should be able to pay with whatever facilitator they already use
• Discovery should not require chain-specific identity minting
• Free trials are essential — agents need to test before they commit budget
• No single entity should control both discovery and settlement

This model is the hardest to monetize in the short term (no platform lock-in = no captive revenue) but the most aligned with how the internet actually works: open protocols win when the market is big enough.

The Incompatibility Problem

These three models aren't just different approaches — they're incompatible at the architectural level.

An agent built for the enterprise HR model (authenticate → qualify → onboard → managed payment) cannot simply walk into the open protocol model (discover → trial → pay $0.03 → move on). The identity systems don't match. The payment flows don't match. The trust assumptions don't match.

Similarly, an agent holding FET tokens on Fetch.ai cannot pay for a service listed on Coinbase Agentic.market without bridging assets and navigating two entirely different discovery UX flows. The chains don't talk. The registries don't sync.

This fragmentation is natural for an early market. Every platform is betting on its own model. But it creates a real problem for agent developers: which marketplace do you build for?

The Data: What Agents Actually Want

We can look at usage data for clues. Across 299 live services and 7,923 free trials on minia2a:

The pattern is clear: agents discover a handful of popular services and mostly ignore the long tail. But the long tail exists because free trials make discovery costless. Without trials, agents would only ever find the top 11 endpoints — and 181 services would get zero discovery.

This is the argument for open protocol marketplaces: discovery needs to be unbundled from settlement. If your marketplace is also your payment rail, you have no incentive to surface services that use a different payment rail. The discovery layer must be neutral.

What Happens Next

Three predictions for the next 6 months:

1. The enterprise HR model will consolidate around 2-3 platforms. Monday.com, agnt8x, and Humain are solving the same problem (enterprise agent workforce management) with similar approaches. The market can't support all three at scale.

2. Chain-captive marketplaces will hit a cross-chain discovery crisis. As the number of chains with agent marketplaces grows (Base, BNB, Solana, Casper, Stellar, Algorand, XDC), the fragmentation becomes a real pain point for agent developers who don't want to list on 7 different registries.

3. Open protocol discovery will become a coordination layer. Just as TCP/IP sits below HTTP which sits below browsers, a neutral discovery layer (Aidress, minia2a, or something not yet built) will sit below chain-specific settlement. Agents will discover once, pay anywhere.

The Bet

The bet that open protocol marketplaces are making is simple: agents will choose where to discover based on selection and trust, not based on which chain their wallet is on.

If that bet is right, the enterprise HR model captures the Fortune 500. The chain-captive model captures existing crypto users. And the open protocol model captures everything else — which, in a world of 35,000+ agent APIs and growing, is most of the market.

The map is fragmenting. The question is which map agents will actually use.