NEAR AI Agent Market Lets Builders Keep Agents and Earn on Rehires

What happened

NEAR AI is making a specific pitch to developers who build AI agents: sell your agent's work one job at a time, and keep owning the agent when the job is done. The company's Agent Market lets builders list task-specific agents, get paid per hire, and earn again whenever a buyer comes back.1 Coverage this week describes a two-track pricing model. A builder sets a fixed price per hire on a listing. Separately, buyers can post a job and let agents bid on it, and an agent can quote a price for that bid that differs from its listed rate.1

The news hook came from NEAR AI itself. On October 16 the company published a long explainer that reads like a recruiting pitch, opening with a builder whose working agent "earns nothing" because selling it would mean building billing, refunds, a storefront and a track record first.3 NEAR Protocol shared a shorter version the next day. Trade outlets then picked up the same points, with one framing them as new features that let creators earn recurring revenue.5

One caveat applies to how this is being covered. The core marketplace is not new. NEAR AI launched Agent Market on February 4, 2026, as an extension of NEAR Intents, its system for stating a desired outcome and having it carried out across blockchains.111 The October material is better read as a repositioning, aimed at a wider pool of agent builders, than as a fresh product release.

How the mechanics work

All the reports describe the same transaction flow. When a buyer hires an agent, the marketplace holds the buyer's money. Delivering the work does not release it; only the buyer's acceptance does. If the buyer never responds, the work is accepted automatically once the acceptance window closes.31 Payouts arrive minus a marketplace fee. Builders can withdraw US dollars to a bank account through Stripe Connect or take USDC to a crypto wallet.3

Refunds go to the buyer if a job is cancelled before work starts, if the agent misses its delivery window, or if a dispute goes against the agent, and NEAR AI says it takes no fee on refunds.3 Disputes are the most unusual part of the design. Either side can open one between delivery and acceptance by posting a deposit equal to a share of the job price. An automated resolver checks open disputes roughly every 12 hours and rules one of four ways: the agent wins, the buyer wins, the payment is split, or the work is sent back for a redo. A human operator steps in when the resolver isn't confident, and either party can ask for human review after any ruling.3

This is a quieter version of what NEAR promised in February. The launch post said dispute agents would hear all sides and issue rulings, and suggested that AI coordination would eventually replace legal contracts and courts.11 The October version keeps the automated resolver but adds human operators and appeals. In practice, NEAR AI seems to have learned that buyers paying real money want a person available when the algorithm can't decide.

"Your agent stays yours"

The ownership claim is mostly about architecture. Builders run their agents' code and models on their own infrastructure. The marketplace sends the agent a job brief and gets back the finished work.1 NEAR AI says listing takes about ten minutes and four HTTP calls on top of an agent that already exists, and works with any programming language or host. Agents poll for work by default, so they don't need a public URL and can run from a laptop or behind a firewall.3 The company also says the marketplace never sees the builder's code or model.3

This is the main difference from many agent platforms, which host the agent and own the runtime. Agent Market works more like a dispatcher plus a payment processor. For builders, that means less lock-in, because the same agent can be listed elsewhere or sold to clients directly. The trade-off is that they handle uptime, inference costs and operations themselves. As one report puts it, the builder is still responsible for running the agent, and the marketplace handles the hire, the payment, refunds and disputes.1

The second part of the ownership pitch is paid assignments. To fill its catalog, NEAR AI posts jobs asking builders to create a specific agent from a detailed spec. Builders are hand-picked from bids, paid in USDC if their agent passes testing, and then keep the agent and earn from any later hires.31 It works like a bounty program, except the builder keeps the result. For NEAR AI, it is a cheap way to cover categories buyers want without hiring staff to build them. For builders, the upfront payment lowers the risk of building something nobody may hire.

Distribution is the real bet

The strategically important details are about where buyers come from. Agent Market connects to Anthropic's Claude through the Model Context Protocol, so people can hire agents from inside the assistant without visiting the site.31 The marketplace also publishes a SKILL.md file and an llms.txt for other agents to read, and supports agent-to-agent (A2A) calls with x402 per-call payments.3 The homepage invites outside agents to read the skill file and either hire agents or do paid work themselves.2

This is the most interesting part for anyone following agent frameworks. Instead of building a single walled-off client, NEAR AI is relying on emerging interoperability standards (MCP, A2A, x402) to put its supply in front of demand from other ecosystems. The February launch already listed compatibility with OpenClaw, Claude, Codex and other major frameworks.11 If agent discovery ends up happening inside general-purpose assistants rather than on separate storefronts, a marketplace that can be reached through those protocols is in a strong position.

The homepage adds a concierge called Axi that matches a plain-English brief to a specialist agent, along with connected accounts. Buyers link Gmail, Calendar or Notion once, and hired agents get scoped, audited, revocable access without handling API keys.2 NEAR AI's one detailed example is Geo, a research agent that checks whether AI assistants recommend a buyer's product. It costs $1 per call, returns a report in about 10 to 20 minutes, and runs on its builder's own hardware.3

Where the coverage diverges

The reporting is consistent on mechanics but differs on chronology and payment rails. A May exchange blog described the Agent Market as part of a May 9 product bundle that moved NEAR's token price, and said agents are paid instantly in NEAR tokens with on-chain settlement.7 NEAR AI's own blog dates the launch to February 4.11 Its IronClaw assistant and confidential GPU marketplace were announced at NEARCON in late February.15 So the May write-up looks like a delayed bundling of earlier announcements, written mainly for traders.

The payment description has also shifted. The February launch said agents would receive instant payment in NEAR.11 By May, NEAR AI had added USDC through its Confidential Intents system to support the marketplace. The October explainer offers USD through Stripe or USDC, with no mention of the NEAR token as the payment currency.3 This is a meaningful change. A marketplace that wants mainstream developers and buyers can't ask them to hold a volatile token, and the move toward dollars and stablecoins suggests NEAR AI is putting usability ahead of token demand.

One trade outlet presented the October update as a strategic push to boost engagement on NEAR.5 That piece offers little detail beyond the company's own statements. The more grounded reporting sticks to how pricing, escrow and ownership work.1

Why it matters, and what to watch

Agent marketplaces have mostly struggled with a cold-start problem: buyers don't show up without capable agents, and builders don't list without buyers. NEAR AI is attacking both sides. It pays builders to fill the catalog and routes demand through Claude and agent-to-agent protocols. Its portable reputation system is meant to keep builders around. Scores are based on delivery, timeliness, responsiveness, dispute-free jobs and repeat hires, weighted toward recent work, and they unlock after a handful of scored jobs.3

The traction claims come entirely from NEAR AI. Its homepage advertises $1.2 million paid to builders, a 96.4% median success rate and a median response time of about 14 seconds.2 None of the coverage independently verifies these numbers, and they should be treated as marketing until outside data appears. The parent network has made larger claims about the underlying rails, saying NEAR Intents had settled more than $14 billion in cross-chain volume by March.6

The October pitch is aimed at freelance developers more than crypto speculators. It promises the payment, dispute and reputation systems they would otherwise have to build themselves, and lets them keep ownership of what they build. Whether that holds up will depend on things NEAR AI doesn't yet disclose: the size of the marketplace fee, how often the automated resolver needs a human, and whether repeat hires turn into real income rather than a talking point.