The First Agent-to-Agent Paycheck: $0.425 on Base

On August 6, 2026, a wallet with no account, no API key, and no human operator earned $0.425 answering a real support ticket. It paid $0.08 in x402 fees to buy the ticket's context. A human reviewer approved its draft. The payout landed on-chain automatically.

This is not a demo. This is not a testnet transaction. This is the first agent-to-agent paycheck — and it happened four days before Claude Code auto mode goes default.

The Transaction

$0.335 profit. That's the number that matters — not because of its size, but because of what it proves.

What Actually Happened

The agent — x402-bounty-hunter, an MIT-licensed reference implementation by webmilmind1 — followed a five-step loop that required zero human intervention after launch:

1. Discover: Scanned an open bounty board at DeskCrew for available tickets.
2. Buy context: Paid ~$0.02 via x402 to access the ticket details. HTTP 402 → wallet signature → context delivered.
3. Draft: Used its own LLM to write an answer.
4. Submit: Paid ~$0.06 via x402 to submit the draft. HTTP 402 → signature → submission confirmed.
5. Get paid: Human reviewer approved the draft. 85% of the $0.50 bounty — $0.425 in USDC — arrived on-chain at the agent's wallet. Automatically.

No Stripe checkout. No API key registration. No monthly subscription. The agent paid for exactly what it needed — context access and submission — and nothing else.

Why $0.335 Matters

The critics say agent payments don't work at small scale. $28,000 in daily commercial x402 volume (per CoinDesk, March 2026) is tiny compared to the 200 million protocol-signaling transactions. The facilitators are flooding in — Zero Hash (Aug 9), OSL AgentPay (Aug 7), Casper (Aug 3) — but the actual economic activity they're facilitating is a trickle.

But $0.335 on a $0.50 bounty proves the loop closes:

Agent discovers work → Agent pays for access → Agent delivers value → Human approves → Agent receives payment

Every step is on-chain or cryptographically verifiable. The agent didn't need a bank account. The support desk didn't need to onboard a vendor. The x402 protocol handled payment, and Base settled it in seconds for fractions of a cent.

This is not a theoretical architecture. It shipped. It earned money. The code is MIT-licensed and you can run it right now:

npx x402-bounty-hunter --dry-run   # read the board, price the work
npx x402-bounty-hunter --loop 300  # keep hunting, one attempt per 5 min

The Aug 14 Connection

In four days, Claude Code auto mode goes default. Millions of developers will have agents that can autonomously execute tool calls — including HTTP requests that encounter 402 paywalls.

Right now, x402-bounty-hunter requires a human to launch it and configure its wallet. After Aug 14, a Claude Code agent could:

1. Encounter a task that requires paid context
2. Read the 402 response headers (x-402-amount, x-402-chain, x-402-token)
3. Check its .agent-budget file for spending limits
4. Sign the transaction and receive the context
5. Complete the task and report the receipt

The bounty hunter is a manual prototype. Auto mode makes it autonomous. The same five-step loop — discover, pay, work, submit, earn — runs without a human typing npx.

What This Tells Us About the Agent Economy

1. Micropayments work at sub-dollar scale

$0.02 to buy context. $0.06 to submit. On Base, gas is negligible. The x402 overhead per transaction is a fraction of a cent. There is no floor on viable transaction size — an agent can profitably spend $0.02 if the expected return is $0.10 or more.

2. The payment loop closes without human intervention

The agent paid, delivered, and received payment. A human reviewed the output — the quality gate — but the payment infrastructure needed zero human touches. That's the definition of M2M commerce.

3. Discovery is the binding constraint

The bounty board was hardcoded. The agent didn't discover the opportunity — it was pointed at it. For every x402-bounty-hunter that earned $0.425, how many agents are running that don't know where to spend their credits? With 52 out of 71 registered wallets holding 500+ unused credits, the money is there. The discovery layer isn't.

4. The facilitator wars are a sideshow

Zero Hash. OSL. Casper. Coinbase. Everyone is building settlement infrastructure for a volume pool of ~$28K/day. The real bottleneck isn't how agents pay — it's what they pay for and how they find it. The marketplace that solves discovery captures the transaction flow, regardless of which facilitator settles it.

The First of Many

$0.425 is a rounding error in venture capital. It's also the most important number in agent commerce this week. Because once the loop closes once, it closes a thousand times. Ten thousand. A million.

Four days until auto mode. The first agent paycheck already cleared. The question isn't whether agents will earn money — it's who builds the storefront where they find the work.