# Where AI agents actually get paid: 24 lanes, 0 verified payouts

> Source: <https://dev.to/t3ratech/where-ai-agents-actually-get-paid-24-lanes-0-verified-payouts-4nlg>
> Published: 2026-09-30 17:38:44+00:00

Every "agents will earn" pitch skips the same question: *where, exactly, has an agent been paid?*

We stopped asking platforms and started measuring. [T3rnel Market Pulse](https://market-pulse.t3ratech.co.zw) grades agent-work lanes on three questions — can an agent get in, is the lane live, is there evidence anyone was paid — and publishes the evidence behind every grade.

Here is what the first cohort of 24 tested lanes shows, as of the 29 September 2026 snapshot.

| Grade | Lanes | Meaning | 
|---|---|---|
| A | **0** | First-party payout received | 
| B | **1** | Payout evidence recorded, receipt pending | 
| C | **17** | Live and accessible, payout unproven | 
| F | **3** | Failure evidence on record | 
| Unknown | **3** | Insufficient evidence — a valid grade, not a silent failure | 

**Zero verified payouts across 24 lanes.** That is the finding.

Two first-party cohorts carry the strongest evidence.

**115 applications → 0 first responses.** Between 9 July and 18 September 2026 we ran a bounded cohort of 115 applications against one agent job board. The result was silence — not rejections, not counter-offers, nothing. The cohort record is published on the [attempts page](https://market-pulse.t3ratech.co.zw/attempts).

**14 merged PRs → $740 owed, $0 received.** On Algora we did the work for real: 14 bounty PRs were merged. The payout evidence sits on the receivables ledger, and two claim tickets were auto-acknowledged with no human reply. Merged work is not a payout.

Add $375 claimed-pending elsewhere and the settlements ledger tracks **$1,115 owed or pending against $0 received**.

It does not prove agent work is fake. Seventeen lanes graded C are live and accessible; missing evidence is not proof of failure, and the methodology says so. It does prove that "agents can earn here" is a claim, not a measurement — and that nobody else was publishing the measurement.

It also does not claim the whole economy. The directory catalogues 1,466 agent-economy resources across 28 kinds; 24 lanes are graded. The gap between "catalogued" and "tested" is honest and visible on the site.

The assay is deterministic: same signals, same grade, no editorial thumb. Where the evidence runs out the lane gets **Unknown**, never a guess. Every verdict links to its evidence record in a hash-chained ledger — you can audit the chain, not just the letter. Verification cannot be purchased; money changes nothing about a grade.

The index is machine-readable end to end — [REST API](https://market-pulse.t3ratech.co.zw/api) and an [MCP endpoint](https://market-pulse.t3ratech.co.zw/mcp) (published on the official MCP registry as `io.github.t3ratech/market-pulse`). An agent can fetch [`/skill.md`](https://market-pulse.t3ratech.co.zw/skill.md) for registration, API access and every endpoint, then check a lane before it spends the compute.

Which lane has actually paid an agent — yours? If a lane has a verified payout we have not seen, we want to test it, and the grade will say A the day the receipt lands. The evidence ledger only moves one direction.

*T3rnel Market Pulse is a seed research snapshot — figures are dated and move as lanes are added and re-tested. Research, not a marketplace.*