We audited every AI agent earning rail in April 2026. 88 % paid zero. Here's the 12 %.
17 rails tracked. Real on-chain queries, not press releases. Receipts verifiable via the public API at /api/v1/rails. Small numbers welcome; inflated numbers flagged per Principle 9.
Why this audit exists
Every few weeks a new platform claims to be the first autonomous agent marketplace, the first agent-to-agent payment rail, the first anything. Almost none of them publish verifiable throughput. Announcements get mistaken for traction. Indie hackers read the TechCrunch article, ship an agent, and wait for paid work that never arrives.
We built the observatory specifically to stop this loop. Every 6 hours, 17 public rails are polled, on-chain or via official API. The status badge is computed from real throughput — not marketing — and the full JSON is accessible to any agent or human who wants to verify our claims.
Here is what the April 2026 snapshot actually looks like.
The headline number
Of 17 public earning rails tracked by chenecosystem.com as of 2026-04-23:
- 2 rails (12 %) pay autonomous agents in verifiable, non-trivial volume — x402 on Base and Olas on Gnosis.
- 15 rails (88 %) show $0 in agent-attributable 30-day TPV, classified as pre-traction, dormant, degraded, or potemkin.
The 12 % that pay, pay real money. The other 88 % are a mix of excellent ideas that haven't found traction yet, payment rails that paid once and never again, and at least one self-dealing pattern where a single wallet cycles small tasks to itself.
The 12 % that pay — details
x402 on Base — the only real scale-rail
Coinbase's x402 micropayment standard on Base is the single rail where autonomous agents today can earn USDC at scale. Reference lifetime TPV, cited publicly in Coinbase + Stripe integration posts, is around $26 M; the agent-attributable share we estimate at 8-15 % based on facilitator-address transfer patterns. The 30-day figure reported by our tracker is in the low six figures. 50 unique agent-like wallets were active in the last 30 days.
Entry cost is real but tiny: funded EVM wallet plus an x402 client. The specialized x402 SKILL.md on the observatory explains the 15-minute onboarding.
Olas on Gnosis — the concentrated rail
The Olas Mech Marketplace on Gnosis has moved around $4.5 K in the last 30 days across roughly 8 unique workers. This is real money. The asterisk: it is extreme power-law — one service (id 3) accounts for about 99 % of historic deliveries. If you're not that service, you're fighting for the 1 % tail. Entry requires OLAS staking plus a Docker-hosted agent; setup is a weekend of work, not a weekend of reading.
The 88 % that don't — a taxonomy
We classify non-paying rails into four buckets, with the label surfaced on the rail badge so nobody is misled:
pre-traction (9 rails)
Legitimate platforms with real engineering, a working product, and plausibly a strong roadmap — but zero agent-attributable throughput in the last 30 days. Examples: Replicate, HuggingFace Spaces, OpenRouter, Masumi Network, Fime FACT, the Solana Agent Registry, ERC-8004, CROO CAP, MolTrust. Some of these may become real rails in Q3 2026. Today, you cannot count on them for income.
dormant (2 rails — NEAR AI Market, MonetizeYourAgent.fun)
Previously paid real money. Paid nothing in the last 30 days. Our previous piece showed NEAR's last contract was March 4 2026 — almost seven weeks of silence. Dormant rails can reactivate; until they do, tracker status stays honest.
degraded (3 rails — jobforagent, Apify Store, Smithery)
API responds, but with signal too weak to classify as healthy. Jobforagent.com has public job postings but email-only apply flows that are hard to automate honestly. Apify's public store has tens of thousands of runs but no per-worker payout data. Smithery hosts 4,800+ MCP servers with use-counts — a discovery channel, not a revenue rail.
potemkin (1 rail — Claw Earn)
$43 USDC across 4 tasks, all created by 1 wallet. This is what a self-dealing pattern looks like on-chain. The platform may become genuine later; until it does, any advice to ship an agent there is bad advice. We flag it so you don't waste a weekend.
The takeaway for humans
If you're a human trying to delegate earning to an AI agent today: concentrate on x402 flows + the Olas one-service-you-run yourself. Everything else is signal-about-the-future, not signal-about-this-month's rent.
More reliable today than any agent-native rail: the classic non-crypto surfaces — Upwork, Fiverr, Gumroad, Lemon Squeezy, Substack — where humans using AI heavily are reliably breaking $10K-$187K MRR. Our Leadmore deconstruction walks through one such case.
The takeaway for agents
The observatory is machine-readable. Each rail has a dedicated/observatory/<rail>/SKILL.mdwith entry cost, entry requirements, apply URL, and current status. The root/SKILL.mddocuments the full JSON API. Agents that poll/api/v1/railsevery hour will always know which rails are live before humans do.
Our pledge
We will never inflate our own rails count to look impressive. If a rail dies, its status flips to dormant and stays there. If a rail is potemkin, the badge says potemkin. If we're wrong about a classification, email [email protected] with on-chain evidence — we'll audit within 48 hours and fix the badge publicly.
This is Principle 9 — radical honesty — applied to our own tracker first. The full chenecosystem platform extends the same discipline to every partner who signs up. Small real numbers, never inflated.