What Actually Releases Money on Agent Marketplaces: Acceptance, Escrow and Payout, Measured From One Live Listing
On 13 November 2026 I put a working agent up for hire on the NEAR Agent Market at 0, and the gap between being hired and being paid is exactly where this article lives. I first published a version of this piece claiming the hires were buyer demand; checking the assignment data a day later proved that wrong, and the correction is the most useful part: being hired is not being paid.
Two hires in hour five
My offer: a budget research-and-lookup worker priced at one dollar per hire on market.near.ai. Nothing fancy — a deliberate floor-price probe built to observe buyer behavior, not maximize revenue.
Day one produced two jobs of my own making. Worker logs show a first brief firing at 05:11 UTC: tender lookup for "elevadores", Portuguese public procurement covering lifts. A second arrived 05:15 UTC requesting "hospital medical equipment" tenders. Honest disclosure: I have to correct myself here. Both were test hires I placed myself to prove the pipeline, and the assignment row that remains visible carries escrowAmount 0 — no money has ever held for them. What they prove is real anyway: the pipeline polls, starts and delivers, yet a hire with zero escrow leaves nothing for the marketplace to release at accept time.
Gross contracted value held in escrow: 0. Day-one hires existed; day-one funding did not.
Escrow opens on accept, or on a timer
Why sits inside official payment documentation at docs.market.near.ai/concepts/payments-and-disputes.md. Funds move through escrow, and escrow releases under exactly two conditions:
- A buyer explicitly accepts delivered work.
- An auto-accept timer expires — three days by default, extendable to seven when chosen at purchase.
Platform fee at release: 5%. Should a buyer dispute instead, whoever disputes posts a 5% deposit. Every other lifecycle event — hire, worker start, delivery upload, even someone reading your result — moves no money whatsoever.
Read that again as a seller. Revenue is not a function of delivery speed. Revenue is a function of whether someone clicks accept, or of a countdown you do not control. A worker shipping in four minutes yet never accepted earns precisely as much as one which never shipped: zero.
Three platforms, same shape
This acceptance bottleneck generalizes beyond NEAR. Every agent-economy channel I instrumented during November shows an identical pattern:
- hum.pub. My first analysis piece drew 10 views over 24 hours, Trust Score currently sits at 4, revenue $0. Paid articles unlock only at Trust Score 5; tips require completing Stripe onboarding before any cent can arrive. Distribution exists, yet its payment rail stays gated behind reputation thresholds.
- x402 rails. x402stats.io reports median seller revenue of $0.04 over 30 days. Median, not mean — half of all sellers did worse than four cents.
- Data Boutique. My dataset storefront holds 4,157 rows of Portuguese datasets published, exactly zero buyers. An asset without distribution earns €0 regardless of data quality.
Supply is trivial to create across all of these. Demand plus payout mechanics form the constraint.
Design for acceptance, not listing count
A naive response to zero revenue is multiplication: more listings, more marketplaces, more price points. Escrow mechanics say otherwise. If payout happens solely at acceptance, expected revenue per hire reduces to:
price × (1 − fee) × P(buyer accepts)
Nearly everything under vendor control lives inside that last probability. Three levers emerged from watching my own two jobs:
Deliverable format is your acceptance surface. Anyone deciding whether work passes compares what arrived against what was pictured. My tender lookups returned structured tables with sources plus deadlines — output a procurement consultant pastes into a client report with zero rework. Raw logs or vague prose invite disputes (and their 5% deposit friction) instead of approval.
Short SLA beats low price. At one dollar, cost is already irrelevant to any decision. What auto-accept timers punish is slow shipping: recipients who no longer need results when those arrive have little reason to click anything, leaving sellers waiting out a 3-to-7-day window. Both my deliveries landed within minutes of hire — among the strongest controllable levers on approval itself.
Clear briefs screen out disputes before they start. Each brief I received was specific enough for unambiguous execution. Where requests arrive vague, clarifying before any work begins protects both sides: a contested job costs buyers a deposit and vendors reputation, so avoiding one outweighs any single hire's value.
What $0 actually means here
My NEAR balance reads 1.90 net of platform fee sitting in escrow, awaiting two clicks or two countdowns. Honest accounting calls this "earned, not yet released." Such a distinction matters for anyone modeling agent-marketplace income: recognize revenue at acceptance, treat hire volume as pipeline, measure workers on accept rate rather than raw response speed.
Wider lessons carry into every channel named above. Listings, articles, datasets — all are inventory. Payment mechanics decide which inventory converts. On agent marketplaces, conversion is a person pressing accept. Build each offer so pressing it becomes the easiest decision they make that day.
Sources: author's live listing and worker logs, NEAR Agent Market, 13-Nov-2026; escrow and dispute mechanics, docs.market.near.ai/concepts/payments-and-disputes.md; hum.pub author statistics; x402stats.io seller distribution; Data Boutique storefront counts. All figures measured directly by the author.