{"assessments":[],"deployments":[],"fuzz":[],"identity":{"adapter":"0xde152afb7db5373f34876e1499fbd893a82dd336","chainId":1,"collection":"0x0000ec93127baa929e58e97dd0095a2bfb38ec1d","registry":"0x8004a169fb4a3325136eb29fa0ceb6d2e539a432"},"interpretation":"Records acceptance and evidence. Neither completion nor an AI assessment establishes correctness, safety, or independent review.","jobId":"f7d86cf5-de2c-4438-9797-d7e8290ab456","kind":"research","nodes":[{"acceptedSubmissionHash":null,"dependsOn":[],"execution":{"network":false,"profile":"foundry","requires":[],"tools":[]},"key":"panel","kind":"research","role":"review","skillHash":null,"skillId":null,"state":"accepted"}],"objective":"Fermi paradox for agent swarms: running an agent swarm is now cheap, so where are all the other swarms? Look for on-chain evidence of other autonomous agent networks active on Ethereum or Base in the last 30 days (agent registries such as ERC-8004, agent-launched tokens, coordinated wallet fleets). List what you find, estimate how many exist, and explain why they are hard to see.","parentJobId":null,"planHash":"214e7a83d15292a333032695a1fa07cd32196e9cf6a39ebc62cb173560cb620e","previousHash":"0000000000000000000000000000000000000000000000000000000000000000","projectId":"f7d86cf5-de2c-4438-9797-d7e8290ab456","publication":{"commit":null,"deliveredAt":null,"repoUrl":null},"receiptIdentity":{"adapter":"0xde152afb7db5373f34876e1499fbd893a82dd336","chainId":1,"collection":"0x0000ec93127baa929e58e97dd0095a2bfb38ec1d","registry":"0x8004a169fb4a3325136eb29fa0ceb6d2e539a432"},"registry":"0xb6d0a187b050fa5bb0b87033a203f37becf4a775","research":[{"answer":"**There are visible agent networks, but the evidence supports a small number of identifiable ecosystems—not hundreds of thousands of autonomous swarms.** My conservative count is **two networks with dated evidence of activity on Base: Virtuals ACP and Olas/Basius; three if agent-operated token infrastructure such as Clanker counts.** That is a discovery lower bound, not an estimate of the entire hidden population. The evidence and qualifications follow.\n\nI used **26 August–24 September 2026**, the last 30 complete UTC days. This is a review of published chain measurements, contract references, and project activity records—not a fresh, exhaustive transaction-log census. “Active” means an observed action inside that window; it does not establish continuous operation or independence from human direction.\n\n| Network or footprint | Addresses / links | Evidence within the window | What it establishes |\n|---|---|---|---|\n| **Olas → BabyDegen → Basius**, Base | Base service registry: [`0x3C1fF68f5aa342D296d4DEe4Bb1cACCA912D95fE`](https://basescan.org/address/0x3C1fF68f5aa342D296d4DEe4Bb1cACCA912D95fE). [Registry measurement](https://satohub.ai/agent-economy/olas). | Olas reports **11 active Basius agents on 23 September** and **1,891 cumulative transactions** through that date. Its earlier 8 September snapshot reported 1,618: **273 additional transactions between those snapshots**, calculated from the two published totals. [23 September snapshot](https://olas.network/agent-economies/explorer?economy=predict), [8 September snapshot](https://olas.network/agent-economies/explorer?economy=babydegen). | Strong evidence of an operating agent fleet. Basius executes Aerodrome liquidity management on Base through user-owned Safe accounts. This is a fleet sharing software and infrastructure; these records do **not** prove the individual instances collaborate. [Basius architecture](https://olas.network/blog/basius). |\n| **Virtuals Agent Commerce Protocol**, Base | ACP v2 JobManager: [`0x9c690c267f20c385f8a053f62bc8c7e2d4b83744`](https://basescan.org/address/0x9c690c267f20c385f8a053f62bc8c7e2d4b83744); [contract inventory](https://satohub.ai/agent-economy/virtuals_acp). | Agent Economy reports **17 commerce memos from five unique senders on 23 September**. Separately, Sato reports **302 transactions from Virtuals ACPX facilitator keys on Base on 24 September**. These are different measurements and should not be added. [ACP measurements](https://agenteconomy.to/virtuals-acp), [facilitator measurements](https://satohub.ai/agent-economy). | The clearest network-shaped candidate: the official SDK implements buyer discovery, seller offerings, job funding, submission, and completion. Memos and facilitator transactions still do not equal completed jobs or independent agents. [Official SDK](https://github.com/Virtual-Protocol/acp-node-v2). |\n| **Clanker token-deployment ecosystem**, Base | Example newly deployed token: [`0xC9750053FE947E0961eab0f3E29325F0311DCb07`](https://clanker.world/clanker/0xc9750053fe947e0961eab0f3e29325f0311dcb07). | Clanker’s own record dates this deployment to **5 September**, with creator `@clanker`. A separate [Laptop token deployment](https://www.clanker.world/clanker/0x9C826AAD86A616EcAEC24bc11DF0Ed040D1B3b07) is dated **7 September**. | Concrete recent launch activity. Count this as **one deployment ecosystem**, not one swarm per token. The launch records do not establish that each requester was autonomous or that the resulting tokens represent working agents. |\n| **x402 payment operators**, Base | [Per-operator chain measurements](https://satohub.ai/agent-economy). | On **24 September**, attributed facilitator transactions include **FluxA: 2,675; Ultravioleta DAO: 1,115; PayAI: 439; Heurist: 1; Daydreams: 1**. | Useful named leads and visible machine-payment infrastructure. I exclude them from the strict swarm count: a facilitator can serve humans, scripts, agents, or several networks. The source explicitly treats facilitator throughput as an upper bound. |\n| **ERC-8004 identity ecosystem**, Ethereum and Base | Both chains’ canonical identity registry: `0x8004A169FB4a3325136EB29fA0ceB6D2e539a432`; [official deployment list](https://github.com/erc-8004/erc-8004-contracts). | The 25 September inventory reports **95,721 Base identities** and **51,409 Ethereum identities**, but explicitly does not measure new registrations for that reading. [Chain inventory](https://satohub.ai/agent-economy). | A discovery surface, **not another swarm and not a 30-day active-agent count**. I would not turn those inventory totals into claims of recent activity. |\n\nThere are also concrete named ERC-8004 candidates worth investigating, but they fail the requested recency test in the records I could inspect:\n\n- **Otto AI**, including trade execution **#1345**, market alpha **#1326**, and tools **#1348**, appears in the [commerce-backed cohort](https://onchainagentintel.io/commerce-backed-agents). However, #1345’s latest reported job outcome is **134 days old**. [Individual record](https://onchainagentintel.io/agents/base/1345).\n- **Lunara #21708** has 438 recorded job outcomes, but its latest is reported as **202 days old**. [Individual record](https://onchainagentintel.io/agents/base/21708).\n- **Wolfpack Intelligence #21923** has 28 recorded outcomes, latest **179 days old**. Its name is not evidence of a currently operating swarm. [Individual record](https://onchainagentintel.io/agents/base/21923).\n\n**How many exist?** At the **network/ecosystem level**, my evidence-backed answer is **at least two, or three under the broader token-agent definition**. These observations do not support a defensible upper bound for independent swarms.\n\nAt the **agent/service level**, the population is much larger, but the units are incompatible. Olas’ measured inventory contains **661 Base services and 58 Ethereum services** as of 21 September; those are registrations, not monthly active fleets. A separate index identifies **44 commerce-backed ERC-8004 agents, all on Base**, using historical job records—not a 30-day activity filter. Neither figure should be added to Virtuals participants or registry totals because the populations overlap. [Olas inventory and overlap](https://satohub.ai/agent-economy/olas), [commerce-backed cohort](https://onchainagentintel.io/commerce-backed-agents).\n\n**I did not establish a named Ethereum-mainnet swarm’s operational activity inside this window.** Ethereum registry presence is well documented; the recent operational evidence found here is concentrated on Base. That is a limitation of this search, not evidence that Ethereum has none.\n\nThe apparent paradox has several explanations:\n\n1. **Wallets are a poor proxy for agents.** Limitless’ actual ACP implementation uses a Virtuals smart wallet for commerce and a separate trading wallet, with multiple buyers’ positions tracked in an internal ledger. Thus one agent can occupy several wallets, while one wallet can aggregate work for several agents. Clustering transfers alone can miscount both ways. [Implementation and architecture](https://github.com/limitless-labs-group/limitless-acp).\n\n2. **Registration is cheap evidence of existence, but weak evidence of operation.** ERC-8004 supplies an ERC-721 identity and metadata pointer, rather than an execution monitor. A September probe of 600 populated URIs per chain found usable cards for **56.33% on Base and 71.33% on Ethereum**; these were samples of eligible URIs, not the whole registry. [ERC-8004 specification](https://eips.ethereum.org/EIPS/eip-8004), [September measurement](https://onchainagentintel.io/blog/state-of-erc-8004-2026-09).\n\n3. **Payments reveal settlement more readily than coordination.** The Limitless implementation keeps ownership accounting off-chain; ACP exposes jobs and settlement. My inference is that chain observers see only part of the collaboration process, making the boundary of a “swarm” ambiguous. [Limitless architecture](https://github.com/limitless-labs-group/limitless-acp).\n\n4. **“Updated today” can conceal old activity.** Otto’s page reports a current on-chain check alongside a job outcome 134 days old. A current dashboard, cumulative counter, or traded token cannot substitute for a timestamped operational event. [Otto record](https://onchainagentintel.io/agents/base/1345).\n\n5. **Cheap execution does not guarantee demand or durable operation.** Olas explicitly moved Optimus and Agents.fun into maintenance after reduced adoption. That is direct evidence of a demand constraint, even within an established agent ecosystem. [Olas maintenance announcement](https://olas.network/blog/a-sho).\n\nMy interpretation: the missing swarms are partly **small, specialized services hidden behind ordinary wallets**, partly **inactive registrations**, and partly **a counting problem**. The observations establish functioning agent infrastructure and at least a couple of active networks; they do not establish a large population of independent, self-sustaining, cooperating swarms.\n\n---\nSources offered but not confirmed by this machine:\n- https://clanker.world/clanker/0xc9750053fe947e0961eab0f3e29325f0311dcb07 — does not contain the quoted text","citations":[{"contentHash":"4b5f9905c3d8ffc69495071dce90d3a699233f9cf9e17b4b04dc7da1ed946e82","quote":"It calls the contracts directly, with no human or UI in the loop","retrievedAt":"2026-09-25T18:11:48.634Z","url":"https://olas.network/blog/basius"},{"contentHash":"92b93f67fe9d475a5fcc63dc8df25386bde7e5906dc14424c0c9bd55429aac62","quote":"Each agreement is recorded on-chain for auditability","retrievedAt":"2026-09-25T18:11:48.872Z","url":"https://agenteconomy.to/virtuals-acp"},{"contentHash":"1ad5b25d3ad8fc44e6c38e9a1e407ef3bdf355b607cf7a564efcd4eb74dc432b","quote":"Transactions sent by known facilitator keys — an upper bound, not wash-filtered.","retrievedAt":"2026-09-25T18:11:49.435Z","url":"https://satohub.ai/agent-economy"},{"contentHash":"d6d6129139ee5db1bd00ee9336f8007c2e68b90f5b1837d115558aa89cafddca","quote":"Since multiple buyer agents route through one trading wallet, an internal JSON ledger tracks which buyer owns which positions.","retrievedAt":"2026-09-25T18:11:50.313Z","url":"https://github.com/limitless-labs-group/limitless-acp"},{"contentHash":"57fdfa2c22a697ea41215ddf7ad307523460866db026be90b42e72e60c56d9ee","quote":"Commerce evidence: 42 ERC-8183 / Virtuals ACP job-outcome rows, most recent 134 days ago.","retrievedAt":"2026-09-25T18:11:50.503Z","url":"https://onchainagentintel.io/agents/base/1345"},{"contentHash":"d936eb153677b8a28c002776fd2c54a3c1e7e74652e8a41f705dc0208a36cb8e","quote":"Optimus and Agents.fun have seen reduced adoption and will receive no further updates in their current form following this lack of demand.","retrievedAt":"2026-09-25T18:11:50.806Z","url":"https://olas.network/blog/a-sho"}],"device":"1d2eacca0a6e8449","id":"9dbf5369-dcd7-400e-9f1a-17f8a8cf15c4","model":null,"nodeId":"7738e002-b1ca-49e0-bb2b-2757792a008b","runtime":"codex"},{"answer":"I'll research this question. Let me start by loading the chain-query skill and doing some background research on ERC-8004 and agent activity.I have the ERC-8004 registry addresses. Now let me get chain heads and 30-day-ago blocks, and find Clanker/Virtuals contract addresses for activity checks.Now I'll anchor the 30-day window on both chains and test the log query format.No `totalSupply` — I'll count mint events. Let me test the `logs` output format with one chunk first.2,131 mints in just the first 100k-block chunk (~2.3 days) — the registry is very active. Let me count mints across the whole 30-day window on Base, plus distinct owner addresses.Notable: ~13.9k registrations in the first ~14 days of the window, then zero for the last ~17 days — the registry may have been superseded by a newer deployment. Let me check Ethereum mainnet, the reputation registry, and whether a v2 registry exists.The canonical registries haven't moved. Let me check the background ETH query and measure reputation-registry + Clanker + Virtuals activity on Base over the window.Good context: 8004scan reports 528,891+ registered agents across chains, and a June 2026 census counted 54,802 on Base. Let me check 8004scan's network breakdown and find the current Virtuals BondingV5 + Clanker deployer activity.Both Clanker and Virtuals factories show zero events — they've likely rotated contracts. Let me get the current deployment lists.Clanker v4.1 has no listed factory — the event may be emitted by a different contract in the pipeline. Let me check the v4 contracts source and Virtuals' OZ manifest for Base proxy addresses.Clanker's static-fee hook emitted 83,976 events in 30 days — the v4.1 stack is alive, just not via the old factory. For Virtuals, let me scan all 36 proxies for agent-NFT mints in a recent slice, and count sniper-auction events as a Clanker-launch proxy.All 36 Virtuals proxies are silent in the last ~10 days — consistent with Virtuals shifting toward Robinhood Chain. Let me get the sniper-auction count and check FRouterV3 over the full window.~5,812 Clanker sniper auctions in 30 days ≈ ~5,800 token launches. Now let me get the per-network breakdown from 8004scan and check whether Base has a newer registry.The full landscape is mapped. One more check — whether the Base registry was paused (explaining the mint freeze), plus a quick keccak-free check on ETH's latest mint vs. the scan's lag.`paused()` isn't implemented — the Base freeze is operator behavior, not a flag. One last citation check on the wallet-fleet repo.I have enough data. Here's the full picture.\n\n# Fermi paradox, resolved: the swarms are there — you're just looking at corpses and ghosts\n\n**Method.** I queried Ethereum and Base logs directly for the last 30 days (Base blocks 50,489,215–51,785,214; ETH blocks 25,841,087–26,056,159), plus public explorers/dashboards. What I found, roughly in order of legibility:\n\n## 1. ERC-8004 \"Trustless Agents\" registries — the visible nervous system\n\nThe canonical singletons — Identity Registry `0x8004A169FB4a3325136EB29fA0ceB6D2e539a432`, Reputation `0x8004BAa17C55a88189AE136b182e5fdA19dE9b63` — exist on ~30 chains.\n\n- **Base**: 13,856 new agent registrations (ERC-721 mints) in 30 days, but *all* before ~Sep 8–10 — zero mints for the last ~17 days. Max agentId ≈ **85,057** (~85k agents ever). 12,813 reputation events in the window, also ceasing ~Sep 13. A June 2026 census counted 54,802 agents, so ~30k were added since June — then the tap turned off.\n- **Ethereum**: 940 mints in 30 days, still minting at head; max agentId ≈ **51,440**.\n- **Elsewhere** (per 8004scan): **BNB Chain 327,404 agents**, Billions 26.2k, Abstract 16.1k, X Layer 13.7k, Monad 10.2k, Celo 9.8k, MegaETH 8.3k, Robinhood 6.1k, Gnosis 4.1k, Solana 1.5k (custom program `8oo4dC4JvBLwy5tGgiH3WwK4B9PWxL9Z4XjA2jzkQMbQ`). Site total: **~538k agents, ~624k feedbacks, ~482k unique wallets**.\n\n## 2. Agent-launched tokens — alive, but the labeled contracts are all dead\n\n**Clanker** (an autonomous token-deploying agent; Bankr's agent API launches through it):\n- All six documented factories (v0–v4.0, incl. `0xE85A59c628F7d27878ACeB4bf3b35733630083a9`) emitted **zero events** in 30 days.\n- The live v4.1 stack tells the real story: hook `0xb429d62f8f3bFFb98CdB9569533eA23bF0Ba28CC` emitted 83,976 events, and `ClankerSniperAuctionV2` `0xebB25BB797D82CB78E1bc70406b13233c0854413` shows 5,812 paired init/end auctions → **~5,800 token launches in 30 days**.\n\n**Virtuals Protocol** (~19.5k agent tokens historically; Dune: 17,357 on BondingV1 + 2,200+ on BondingV5/AgentFactoryV6-V7): **all 36 Base proxies emitted zero events in the last ~10 days**; BondingV5 `0x1A540088125d00dD3990f9dA45CA0859af4d3B01` was silent all 30 days. The docs now list a Robinhood Chain bonding curve — the swarm moved, it didn't die.\n\n## 3. Coordinated wallet fleets\n\n- `bondv5-trader` on GitHub is a published artifact: \"Adapted from a private fleet-based market maker running on Base mainnet\" — per-wallet isolated ACP smart wallets with session signers.\n- Bankr runs per-user agent wallets plus an \"Agent Wormhole\" for sealed agent-to-agent handoffs.\n- On-chain evidence: my Base mint scan shows 13,856 registrations from only 6,959 distinct addresses, and one ~55-hour chunk contained 9,796 mints — scripted fleet bursts, not organic signups.\n- Disclosure: the swarm producing this answer is itself visible — 8004scan's latest agents were \"IdentityMD #1770/#1435/#800\", minted on Ethereum minutes before I queried it (agentIds ~51,438–51,440).\n\n## 4. The invisible rails\n\nx402 agent-to-agent payments (e.g., Toppa on Base, which \"settles in cUSD via x402 micropayments\") ride on ordinary USDC EIP-3009 transfers — on-chain they are indistinguishable from normal payments.\n\n## Estimate\n\n- **Registry-visible agents**: ~538k across ~30 chains; ~136k on ETH+Base; ~14.8k new in 30 days.\n- **Actual swarms**: maybe **a few hundred to ~1–2k distinct operating entities** behind the mint bursts and launch pipelines, plus an unknowable tail of unregistered bot fleets (trading, sniping, x402 payers) that is plausibly the majority.\n\n## Why they're hard to see\n\n1. **Registration is optional.** Cognition is off-chain; the footprint is one mint. Unregistered fleets look like ordinary EOAs.\n2. **Contract churn blinds indexers.** Clanker rotated through 6+ factories — every documented address is dark. Virtuals spans 36 proxies and 3 factory generations, then moved chains.\n3. **Events aren't self-labeling.** Agent-launched tokens, MEV auctions, and agent USDC payments look identical to human activity.\n4. **Activity is patchy in space and time.** Base's registry froze ~Sep 8–13 while Ethereum kept minting; Virtuals shifted to Robinhood. A single-chain census sees a corpse that's actually migration.\n5. **Counts disagree.** Same Base registry: 61,516 (QuickNode explorer), 73,943 (8004scan), ~85k (observed max agentId).\n6. **Wallet count ≠ swarm count**, in both directions — 482k \"unique wallets\" collapse to thousands of scripted minters; conversely one launch contract fronts for thousands of agent users.\n\nAnswer to the paradox: the swarms are numerous and active — they're just metabolically cheap, mobile across chains, and wrapped in contracts that rotate faster than observers can label.\n\n---\nSources offered but not confirmed by this machine:\n- https://raw.githubusercontent.com/erc-8004/erc-8004-contracts/main/README.md — does not contain the quoted text\n- https://dune.com/remmez/virtuals-protocol-agentic-economy-dashboard — does not contain the quoted text\n- https://docs.bankr.chat/token-launching/base-launches — could not be read: fetch failed","citations":[{"contentHash":"daad4db2f11f6f0d80f72bee63cf76fcb68159cf804ec5cbda85e6c0c0ec64c6","quote":"ERC-8004 registry contracts deployed across multiple chains","retrievedAt":"2026-09-25T18:20:06.202Z","url":"https://8004scan.io/networks"},{"contentHash":"4fc38263c59d325431b5fe7f98cde7d447fa48c1ab9212665c21f75e6037da6d","quote":"54,802 agents registered in the IdentityRegistry","retrievedAt":"2026-09-25T18:20:06.424Z","url":"https://huggingface.co/datasets/rsoft-latam/erc8004-base-census-jun2026"},{"contentHash":"1c5acdfea78ad3320f3130292a615495a09a6ffa08b341503444c186a1b3bb94","quote":"Clanker is an autonomous agent for deploying tokens.","retrievedAt":"2026-09-25T18:20:07.912Z","url":"https://github.com/clanker-devco/v4-contracts"},{"contentHash":"a75f8828d0db63a58a5b6c1d0a8d6d62beed6e2741515a06436316561ebf1617","quote":"Adapted from a private fleet-based market maker running on Base mainnet.","retrievedAt":"2026-09-25T18:20:08.234Z","url":"https://github.com/moomooed/bondv5-trader"},{"contentHash":"66caf86a27af0b36ab8da152783989410cd1f6b692224f56c4a832fc5ef7f50e","quote":"0xE85A59c628F7d27878ACeB4bf3b35733630083a9","retrievedAt":"2026-09-25T18:20:08.368Z","url":"https://clanker.gitbook.io/documentation/references/deployed-contracts.md"}],"device":"1d142f9c9d30c62a","id":"bd48c8df-3748-4b2c-8950-f9e021e72ffd","model":"swe-2-high","nodeId":"7738e002-b1ca-49e0-bb2b-2757792a008b","runtime":"codex"},{"answer":"**The swarms are visible, but the evidence supports a handful of active ecosystems—not thousands of independently autonomous collectives.** I found four relevant ecosystems with recent Base activity: **Virtuals, Olas, Bankr and Clanker**. Registry evidence also points to additional publishing platforms, but does not establish that their agents collaborate autonomously.\n\nI used **27 August–25 September 2026 UTC** as the last 30 complete days. Sources have different refresh dates; below, I distinguish recent activity from lifetime totals. This is a targeted investigation, not a complete chain census.\n\n| Network / ecosystem | Addresses or links | Evidence within the window—and what it establishes |\n|---|---|---|\n| **Virtuals / Agent Commerce Protocol (ACP), Base** | [ACP explorer](https://app.virtuals.io/acp); agent NFT registry `0x50725af160260a316b2673C71C8c21469f6732c0`; BondingV5 factory `0x1A540088125d00dD3990f9dA45CA0859af4d3B01` | Summing the published daily dataset gives **4,632 commerce memos during 27 Aug–22 Sep**, with **5–14 senders per day**. These are memos, not necessarily completed jobs or distinct autonomous agents. Separately, the launch counter increased from **19,241 to 20,788** across the dashboard’s four weekly observations through 21 Sep. This is the strongest candidate here for an active agent-commerce network. [Daily data](https://agenteconomy.to/data.json), [contracts and launch measurements](https://satohub.ai/agent-economy/virtuals). |\n| **Olas / Autonolas, Base and Ethereum** | Base service registry `0x3C1fF68f5aa342D296d4DEe4Bb1cACCA912D95fE`; Ethereum service registry `0x48b6af7B12C71f09e2fC8aF4855De4Ff54e775cA` | Base registered services increased **636 → 661** across four weekly observations through 21 Sep; Ethereum remained at **58**. Thus, there is recent Base registration activity, but these counts do not prove every service ran during the window. Olas separately reported **9 daily active Mech agents**, a seven-day average as of 18 Sep, across its network—not specifically Base/Ethereum. Its Mech marketplace explicitly supports agents requesting work from other agents. [Registry measurements](https://satohub.ai/agent-economy/olas), [Olas Mech measurements](https://olas.network/agent-economies/mech). |\n| **Bankr tokenized-agent ecosystem, Base** | [Official agent roster](https://github.com/BankrBot/tokenized-agents); examples include **CLAWD**, **STARKBOT**, **EMBER**, **AXIOM** and **CORAL** | The roster contains **15 named tokenized agents**, but does not establish recent runtime activity for each. For a concrete recent launch, Bankr’s API reports CLAWEVAL token `0xaddb5e903b7fdd210979f1fa7ddaf9dd95fb5ba3`. I independently checked its successful Base transaction by RPC: **25 Sep, 18:05:27 UTC**. That proves deployment through this ecosystem; it does **not** prove autonomous initiation, affiliation with the named software project, or a swarm. [Launch feed](https://api.bankr.bot/token-launches), [transaction](https://basescan.org/tx/0xc4a938d3ff08e985ed205d7e8937e41464d45720a42fb75e193c8648c732ddfd), [roster addresses](https://raw.githubusercontent.com/BankrBot/tokenized-agents/main/AGENTS.md). |\n| **Clanker and its bot/agent integrations, Base** | [Base deployment dashboard](https://www.clanker.world/clankers/chain/base/stats), including factory link | The retrieved 24 Sep snapshot reports approximately **9,700 deployments over its trailing 30 days**, with a latest deployment that morning. Interface labels include **Clanker Bot, FourClaw, OpenClaw, grok-fee-bot and Hermes Agent**. This establishes an active automated launch ecosystem. Interface labels and token deployments do not establish how many independent agents—or coordinated swarms—exist. [Clanker’s own dashboard](https://www.clanker.world/clankers/chain/base/stats). |\n\n**ERC-8004 reveals more candidates, but counting registrations as swarms would be a major error.**\n\nThe canonical identity registry is `0x8004A169FB4a3325136EB29fA0ceB6D2e539a432` on [Base](https://basescan.org/address/0x8004A169FB4a3325136EB29fA0ceB6D2e539a432) and [Ethereum](https://etherscan.io/address/0x8004A169FB4a3325136EB29fA0ceB6D2e539a432). Sato’s 25 Sep reading reports **95,721 Base identities** and **51,409 Ethereum identities**, plus **6,791 entries in a separate Ethereum registry**. These are inventory counts, not 30-day active populations. Its weekly series shows recent growth on both chains. [Registry measurements](https://satohub.ai/agent-economy/erc8004).\n\nTwo additional platform leads emerge from registration metadata:\n\n- **Termix, Base:** its S3 metadata host accounts for **23.1% of a 52-registration sample**.\n- **IMD / `api.imd.fun`, Ethereum:** that host accounts for **95% of a 20-registration sample**.\n\nThose are small, window-sampled publisher clusters—not market shares or verified coordinated wallet fleets. Virtuals’ metadata host also appears in the Base sample, illustrating overlap between ecosystems. [Publisher attribution and sampling disclosures](https://satohub.ai/agent-economy).\n\nI did **not** identify an otherwise unknown wallet fleet for which public evidence simultaneously established common control, autonomous execution, inter-agent coordination and activity inside the requested window. Shared infrastructure alone would not meet that test.\n\n**My estimate: four visibly active ecosystems; roughly six named ecosystem/platform candidates; an unknown number of independent swarms.**\n\nThat estimate counts Virtuals, Olas, Bankr and Clanker once each, then adds Termix and IMD as weaker registry-derived candidates. It does not count ACP separately from Virtuals or every token as a network. This is a discovery count from the evidence above, **not a statistical estimate of the entire population**.\n\nFor strict autonomous multi-agent collaboration, **Virtuals and Olas are the two strongest candidates found**. Even there, the evidence does not furnish a clean census of independent operators active specifically on Ethereum/Base over all 30 days. A claim such as “100,000 swarms exist” would confuse identities with systems; a claim that “only two exist” would mistake limited detection for absence. ERC-8004 explicitly allows the same agent to register on multiple chains and operate elsewhere. [ERC-8004 specification](https://eips.ethereum.org/EIPS/eip-8004).\n\n**Why they are hard to see**\n\n1. **A wallet does not disclose who—or what—controls it.** Sato identifies Privy, Turnkey, Coinbase CDP Server Wallets, Dynamic and Fireblocks as structurally difficult to census: their ordinary externally owned accounts lack an agent-specific fingerprint. My inference: a coordinated fleet can look like unrelated users, while unrelated users sharing a wallet provider can look coordinated. [Attribution limitations](https://satohub.ai/agent-economy).\n\n2. **Identity, communication and settlement are separate layers.** ERC-8004 points to external services and explicitly leaves payments outside its scope. Olas describes multi-agent services that operate mostly off-chain and settle on-chain. Consequently, transactions can reveal outputs without revealing the conversation, delegation tree or decision-maker. [ERC-8004](https://eips.ethereum.org/EIPS/eip-8004), [Olas architecture](https://agenteconomy.to/olas).\n\n3. **Registration is cheap; functioning endpoints are scarcer.** A 2 Sep study sampled 600 populated agent URIs per chain: **56.33% on Base** and **71.33% on Ethereum** resolved to usable agent cards. These are conditional sample rates, not percentages of all registered agents. The same report identifies an Ethereum publisher with **10,000 mints**—a clear reason not to equate mint counts with independent operators. [Dated registry study](https://onchainagentintel.io/blog/state-of-erc-8004-2026-09).\n\n4. **Payments and token trading are ambiguous signals.** A token’s transfers show market activity, not that its agent is working. Likewise, an EIP-3009 authorization event is not uniquely an x402 payment; the primitive has other uses. Even genuine payment does not by itself prove independent customers or autonomous collaboration. [Virtuals measurement caveats](https://satohub.ai/agent-economy/virtuals), [payment-count methodology](https://aspern.org/coverage).\n\n5. **The observation tools have blind spots.** Registry-owner nonce checks include human transactions and additional registrations, while missing actions routed through smart accounts, relayers or session keys. Public-node history limits also truncate sampling. These limitations can inflate apparent activity and hide real activity simultaneously. [Detailed sampling limitations](https://satohub.ai/agent-economy/erc8004).\n\nThe evidence therefore favors an **observability and attribution problem**, alongside a large gap between deploying an agent-shaped object and operating a useful autonomous business. Cheap deployment explains abundant registrations and tokens; it does not establish abundant independent, collaborating swarms.\n\n---\nSources offered but not confirmed by this machine:\n- https://satohub.ai/agent-economy/virtuals — does not contain the quoted text","citations":[{"contentHash":"5783624aab3012c72c489dcb897878633f98713e2f214763b7faa31ce6307c27","quote":"Payments are orthogonal to this protocol and not covered here.","retrievedAt":"2026-09-25T18:12:07.613Z","url":"https://eips.ethereum.org/EIPS/eip-8004"},{"contentHash":"4f897f83ed31070568192fd017ed4004ae7d912658fe40ec53790ac6de00c133","quote":"These populations intersect: the same real agent can appear in both.","retrievedAt":"2026-09-25T18:12:09.872Z","url":"https://satohub.ai/agent-economy/olas"},{"contentHash":"ca1be6a4cbb833a6b608f506639aeee2d25c2a61608fd046fba4f33107695218","quote":"A community-maintained registry of tokenized AI agents deployed through Bankr.","retrievedAt":"2026-09-25T18:12:10.272Z","url":"https://github.com/BankrBot/tokenized-agents"},{"contentHash":"83e77221223370b63087af099048db777e90d84cdf44d2cc00b5a07cfde89cd4","quote":"Chain statistics and deployment analytics.","retrievedAt":"2026-09-25T18:12:11.772Z","url":"https://www.clanker.world/clankers/chain/base/stats"},{"contentHash":"e6fc82cd169d1f664c71badc92996e58688fcc89c057123294db3ece25aad294","quote":"On 2 September 2026 we drew a stratified random sample of 600 agents per chain","retrievedAt":"2026-09-25T18:12:12.303Z","url":"https://onchainagentintel.io/blog/state-of-erc-8004-2026-09"},{"contentHash":"ccf41ca80e7134de36ed996bf94b2e0b4c16fb4353c9f3bba4bcb0bf688756fb","quote":"A request is not a settled task.","retrievedAt":"2026-09-25T18:12:13.651Z","url":"https://aspern.org/coverage"},{"contentHash":"de60d136212c9dcd43bfe50fa48c6c085e05a25ecf2e4335c1da9d48ffc7cf76","quote":"A framework for building agent services as multi-agent systems that mostly run off-chain while settling on-chain.","retrievedAt":"2026-09-25T18:12:14.255Z","url":"https://agenteconomy.to/olas"}],"device":"1ebf032d4201b2fc","id":"cd20a9b7-d734-4fff-8f6d-066cef760e45","model":null,"nodeId":"7738e002-b1ca-49e0-bb2b-2757792a008b","runtime":"codex"}],"schema":"identitymd-work-v1","signals":[],"site":null,"snapshotHash":"cfd54b50c76d818c10003253193096c4672029cea5b0f27c8405136b56286bc9","state":"completed","submissions":[],"verification":[]}