The Operator plan in 68 seconds: watch 20 AI agents on @Base How the Operator plan works on the Agentic Finance Graph Desk: paste a whole fleet of agent numbers at once, find your own agent by its wallet address, read each agent's card, see the alerts your Desk rules raise, and get every detection in Telegram. Chapters 0:00 The Operator plan 0:11 Add your fleet 0:21 Find your own agent 0:32 Cards and alert rules 0:48 Your account and Telegram #AIagents, #AIagentmonitoring, #agentwalletalerts, #Base, #USDC, #ERC8004, #cryptosecurity, #Agentic #Finance #Graph

Where bridged agent money goes. All time, $10.1M of counted payments from agent wallets on Base went into bridges, in 24,575 payments. We followed each one through the bridge's own status record; 97% resolved. Of the $8.61M that crossed to another chain, 97% arrived at the same address as the paying wallet, directly or forwarded to it. 84% of it went to Arbitrum One. So most of it looks like agents moving their own money, not paying anyone. That is why money that comes back to the same wallet is taken out of what we count as having left the agent's control.

x402 on @base , measured from the chain, last seven days: $236,422 reached the payTo addresses of 1,448 listed sellers, from 6,773 paying wallets. 99 of those wallets are bound to an ERC-8004 agent registration. They paid $1,439, 0.6% of the total. On the identity side, 25,820 of 1,140,297 Base registrations have been seen paying anyone. Agent payments and agent identity are still mostly separate worlds. Joining them is what lets you check a counterparty before you serve it or pay it. Upper bound for listed sellers (a payTo address can receive other money), floor for x402 overall.

Builder Codes on AI agent payments, measured on Base: $9.11M of counted agent payments carry an ERC-8021 code, 23% of the value we checked (99% of counted value is checked for a code). 456 distinct codes. The largest code carries 76% of the coded value: $6.95M from 529 agents to 13 payees. A code names the app that built the transaction, not the payee. Its registered metadata is usually private, so we show codes as codes and do not guess who is behind one. agenticfinancegraph.com/buil… @base

Agent-to-agent commerce, measured: Virtuals' Agent Commerce Protocol on Base. 9,569 jobs indexed since 13 August. 5,279 funded into escrow, 2,780 completed, 1,008 paid the provider. 175 clients, 255 providers, 111 of them paid. Funded so far: $300. Paid out: $131. Five days ago it was $239 and $94. 64% of jobs with a payout released nothing, so counting payout events instead of value would overstate it about threefold. The flow works end to end; the money is still small. A baseline worth watching. Measured 26 Oct 2026, 14:26 UTC. agenticfinancegraph.com @virtuals_io @base

Do registered AI agents actually answer? In a random sample of 1,024 Base registrations: 94% have a registration URI. 83% declare a service. 75% answer at it within 8 seconds. 74% complete a protocol handshake (MCP, A2A or x402). Of the 82 in the sample with an endpoint on another domain, 3 serve ERC-8004's domain-proof file, and none of those files lists the agent.

Builder Codes (ERC-8021) on AI agent payments on Base: $8.29M of counted agent payments carry a code, 22% of the value we read. 390 distinct codes; the largest carries 76% of the coded value. We show codes as codes: the registry names no one, and we do not guess. agenticfinancegraph.com/buil… @base

Our Telegram group posts live from the ledger: drain patterns and address-poisoning losses on AI agent wallets as they are detected, a 09:00 UTC digest with the day's figures and the x402 rail, and dated findings. Watch your own agents privately: message @AgenticGraphBot with /watch and the agent number. t.me/AgenticFinanceGraph

What AI agents on @Base pay with: USDC carries 99.4% of the value we count ($37.07M of $37.28M). EURC $110K, USDT $92K, DAI $5K. We also read every issuer freeze on Base from the chain. 503 addresses are frozen for USDC today. Registered agents have paid none of them, and no agent wallet is frozen. Measured 21 Oct 2026, 08:19 UTC. agenticfinancegraph.com @circle

What AI agents on @base pay with: USDC carries 99.45% of the payments we count, $38.73M of $38.94M. EURC $111K, USDT $97K, DAI $5K. Circle can freeze USDC and EURC addresses, and every freeze is an event on chain, so we read them. 504 addresses are frozen for USDC and 476 for EURC. No agent wallet in our index has paid a frozen address, and none is frozen. @circle

The clearest line we've read on this. Agents need settlement that is instant, global, and fine with a small amount. Stablecoins are that rail. Glad Coinbase is saying the 2030 number out loud.

Who actually submits x402 settlements on @Base right now? In a random 300 of the EIP-3009 settlement transactions of the 24 hours to 21 Oct, 07:48 UTC: Central City's facilitator 161, Coinbase's 95, senders on no public facilitator list 44. By count, an agent platform's own facilitator is landing more settlements than anyone else. By value it is different: facilitator-submitted transactions carried 19% of the dollars in the sample. Method: agenticfinancegraph.com/def/… @CoinbaseDev

What would a stolen key control right now? We measured every wallet of funded AI agents on Base at block 52,977,621 (23 Oct, 08:39 UTC). 728 wallets. $196,890 in total. Median $31. The largest holds $37,552. 195 run an EIP-7702 delegation. None points at a known sweeper. 19 have open token allowances, $1,050 exposed. A measurement at one block, not a verdict. agenticfinancegraph.com/ai-a…

Before an AI agent pays, it can ask us first, free, on our MCP server: pay_check: has the agent you are about to pay ever paid anyone, and does it have a drain or poisoning history. payee_facts: seven measured facts about the address it is about to pay. gate_check: your policy, answered block, warn, log or allow, with every reason. agent_pass: eight checks on an agent at one moment, valid three hours. We never block a payment. We measure, and the client decides.

Second prediction from our manifesto. Also wrong. Does the amount of on-chain feedback an AI agent receives predict that it starts paying anyone? We expected no better than chance. On the cohort we fixed in advance, 24,727 agents on Base not yet paying on 15 September, with the ten raters who wrote 52% of all feedback set aside, it came out worse than chance: AUC 0.45 (interval 0.42 to 0.49). Agents with no feedback started paying at 0.60% within 30 days. Agents with five or more: 0.07%. A second cohort a week later sits within chance. Feedback volume is not a signal for who pays. Every row, with token ids, is public.

We published a prediction in our manifesto. It was wrong. The question: of all USDC that left AI agent wallets on @Base since 3 February, which of our exclusion rules changes the spending figure most? We expected the pre-registration rule. It is the routing-hop rule. $307.98M left agent-bound wallets. We count $38.48M of it as spending, 12.5%. Switch off the routing-hop rule alone and the share jumps to 49.1%. Switch off the pre-registration rule alone and it moves to 19.2%. Switch off the own-address rule and it moves by $2,314. Most money that left before registration was also a routing hop, so it stays out either way. Method written down before the query ran. Queries and rows are public. agenticfinancegraph.com/rese…

Address poisoning against AI agents on @Base, last 7 days: 463 agent wallets targeted, 57 fake look-alike tokens, and 985 look-alike transfers in the last 24 hours alone. Losses that actually landed, all time: 2 payments, $60. Targeting is constant, loss is rare, and the exposed moment is always an agent's first payment to a new address. Our free payee_facts check covers that moment with seven measured facts about the address. Measured 21 Oct 2026, 07:47 UTC.

Which contracts do the most AI agents on Base pay into? Ranked by how many different agents (L7 and up) paid each one, not by dollars: A job escrow contract: 4,550 agents. Relay's depository: 858. LI FI's diamond: 793. 0x Settler: 500. Bridges and swap routers are the everyday plumbing of agent money. Counted payments only, routing hops left out. Measured 21 Oct 2026, 06:07 UTC. agenticfinancegraph.com @lifiprotocol @0xProject

Agents that get paid, not just agents that pay: 1,424 ranked agents on @Base have received counted income, $3.36M so far, $1.84M of it in the last 30 days, from 1,170 paying wallets. Income is the half of agent finance almost nobody measures. Same rules as spending: after registration, routing hops left out. Measured 21 Oct 2026

AI agents on Base have sent $7.44M into bridges. We followed 97% of it to the other side through the bridges' own public status records. 82% of what crossed chains landed on Arbitrum One, and 96% of all of it arrived at the paying wallet's own address on the other side. So most agent bridge payments are an agent moving its own money, not paying someone else. We keep them apart from spending. Thanks to the teams that publish status APIs: Across, LI FI, Relay, deBridge and Circle's CCTP. @AcrossProtocol @lifiprotocol @circle

How much machine money moves on Base in a day? We now read every USDC payment authorisation on Base (EIP-3009, the mechanism x402 settles with), not a sample. 24 hours to 21 Oct, 07:48 UTC: 300,164 authorisations, $1.71M, 8,572 payers, 3,125 payees. 653 of those payers are wallets of registered AI agents. In a random sample of 300 settlement transactions, 85% were submitted by a listed x402 facilitator. @CoinbaseDev @base