Yesterday we enabled @MartinEstateai to launch a way for AI Agents to purchase a regulated product, wine, via our KYC 21 passport and @mpp, @tempo and @stripe rails. What actually changed is pretty simple: - user verifies once (21 ) - you set limits on spend / scope - the agent completes the purchase Wine is just the example, but this applies to: subscriptions, services, inventory, bookings, etc.

Excited to enable a new generation of American businesses as the agentic economy takes off.

We just recorded the first fully autonomous agent wine purchase. @agentscoredev enabled the policies including KYC to verify 21 to purchase wine from @martinestate. The agent handled the rest — payment over @Tempo USDC, @MPP @stripe rails, order confirmed. Agentscore is the trust layer for agentic commerce - identity, compliance, and risk, so agents can transact autonomously.

Every AgentScore grade is derived from on-chain transaction history. No self-reported data, no staking, no social signals. 49,655 wallets scored using only what the blockchain can verify. Turns out most reputation systems skip the part where you check what actually happened.

Looked at ERC-8004 registrations by chain. 47,243 agents across 7 networks, but remove Base and the total drops by more than half. Five of the seven chains have fewer registered agents than a single Base contract factory has deployed.

6.1% of x402 wallets score a B. 66.1% score a D. There's almost no middle class in agent reputation — you're either a consistent power user or a drive-by. The C tier at 22.2% is the closest thing to a healthy middle, and even that skews toward the bottom of the range.

Indexed ERC-8004 agents across 7 chains. Base has more registered agents with actual transaction history than the next three chains combined. The multi-chain agent economy exists on paper. In practice, it's mostly Base.

Cross-referenced 49,655 x402 wallets against 47,243 ERC-8004 registered agents. 17 appear in both datasets. That's 0.12% overlap. Two ecosystems building agent identity on the same chain, almost entirely non-overlapping populations.

7.1M x402 transactions across 49,655 wallets. Average of ~143 transactions per wallet, but the median is closer to 1. A handful of wallets are doing almost all the talking.

122 facilitator wallets handle the routing for the entire x402 payment graph on Base. Out of 49,655 scored wallets, that's 0.25% acting as infrastructure. The plumbing of the agent economy fits in a spreadsheet.

47,243 ERC-8004 registered agents across 7 chains. Pulled their on-chain transaction histories. The vast majority have never made a single x402 payment. Registration is cheap. Activity is rare.

Top 10 x402 wallets by volume account for over 40% of all payment activity on Base. 49,655 wallets in the dataset. The Pareto distribution in agent payments is even steeper than web2 API usage patterns.

All scoring data is available via API — no key, no auth. curl agentscore.sh/api/v1/x402/wa…{address} and you get a grade, transaction count, volume, and a breakdown of how the score was calculated. Felt wrong to gate reputation data behind a paywall.

Grading methodology question we keep circling: should a wallet with 10,000 small transactions score higher than one with 50 large ones? Volume vs frequency. Right now we weight both. 49,655 wallets scored, and the grade distribution is steep — 66% land at D.

Broke down ERC-8004 registrations by chain. Base has more registered agents than the next three chains combined. The agent economy isn't multi-chain yet — it's Base with a long tail.

25,536 x402 wallets. 47,243 ERC-8004 registered agents. Cross-referenced them — only 17 appear in both datasets. 0.12% overlap. Agents that pay and agents that register are almost entirely separate populations.

122 facilitator wallets route the majority of x402 payment traffic on Base. Out of 25,536 scored wallets, that's 0.48% of the network doing most of the work. Infrastructure centralizes fast, even in decentralized protocols.