Identity.md / IMD: how the AI-agent economy works, and whether it can become a sustainable market for autonomous work

Snapshot time: 2026-10-05, about 22:45–23:05 UTC. Every live figure below comes from that window unless it says otherwise. The network changes daily, so recheck the numbers before you rely on them.

Method: I read the public control-plane API (https://api.imdnotdead.fun) and the explorer directly. I also read the official API docs (https://imdnotdead.fun/docs) and token page (https://imdnotdead.fun/token), the worker client README, the ERC-8004 and x402 specifications, DexScreener market data, and secondary press (Bankless, KuCoin, TokenPost, and a third-party GitHub write-up). I fetched about 810 job records one by one to see who pays. I did not read contract source code or trace on-chain token flows (burns, sIMD vault balances, where fees go after they arrive), so any claim that depends on those is marked as unverified.

Labels used throughout:

• [F] Fact: read directly from a primary source (the live API, official docs, a spec, or market data). A citation follows.
• [R] Reported: stated by a secondary source that I did not independently confirm.
• [I] Inference: my analysis built on the facts. It may be wrong.
• [U] Unknown: I could not establish it.

---

1. Summary

1. [F] The system has four layers:
   • IMD is an ERC-20 token. It is the only payment asset for paid work.
   • Identity.md NFTs are 2,000 seats. Each NFT allows one worker device.
   • ERC-8004 is the identity and reputation registry where each seat is registered as an agent.
   • x402 + Permit2 is the payment rail. Every paid action currently costs 0.5 IMD.
2. [F] "sIMD" and "SIMD" are two different things:
   • sIMD is the ERC-4626 staking share of IMD.
   • SIMD ("Super Intelligent Identity", @SuperIMD_eth) is a separate token. Its Uniswap pair against IMD was created on 2026-10-04. The wallet linked to it is currently the largest buyer of swarm jobs.
3. [F] Paid demand is real but very small:
   • 1,203 confirmed paid orders over the network's whole life. At 0.5 IMD each, that is about 600 IMD, or roughly $8k at today's price of about $13.7.
   • The main IMD pool alone traded about $7.2M in the last 24 hours.
4. [F/I] Most of the reported work volume is not paid customer work:
   • About 99.7% of all accepted submissions come from the oracle-assess skill.
   • 55 of the 60 most recent oracle requests I sampled used a panel of 200 seats answering a question the chain itself can answer. The paid route allows at most 100 seats per panel.
   • [I] These are almost certainly generated by the operator to keep seats active, not bought by customers.
5. [F/I] Paid demand is concentrated and mostly comes from inside the ecosystem:
   • Across 498 paid non-oracle jobs (2026-10-03 to 2026-10-05), there were 29 paying wallets. The top three paid for 81%.
   • The top payer's jobs are SIMD-branded hash-collision puzzles and SIMD "thesis grading" of tweets. The second payer's jobs are for an NFT art project ("IDENTITY UNITS").
   • About 75% of non-oracle job objectives mention "IMD" or "identity".
6. [F] Seat owners are not paid directly by the customer's 0.5 IMD. The x402 payTo address is the same address as the server's gas wallet. As documented, seat rewards come from three other sources:
   • token allocations from launches,
   • a share of POOL4 sell-side trims (reported, not verified by me),
   • launch-pool trading fees.
   • [I] So today, worker pay depends mainly on token trading activity, not on customer revenue.
7. [I] The strongest candidate for recurring outside demand is scheduled, attested oracle answers for long-tail questions that a contract consumes. One live example: a 4-hourly Heathrow weather feed with a named consumer contract. Two things stand in the way: the attestation is signed by a single operator key, and job prices are fixed in a volatile token.

---

2. How the pieces fit together (verified mechanics)

2.1 IMD token

• [F] Contract 0xd34a99bc0f67ae1bbd63c660e6d0b0dd03e263b7 on Ethereum mainnet. The official token page says 10M were minted in 2023 ("none since"). Supply is split across Ethereum, Base and Robinhood Chain and bridged 1:1. The token began as Fren Pet's $FP on Base (Aug 2023), was bridged to Ethereum in Oct 2025, and was renamed $FP → $VIBE → $IMD in May 2026. (imd.fun/token)

  • [R] KuCoin puts the VIBE rename in January 2026, which conflicts with the official page. (KuCoin)

• [F] The official page says: "Every sell burns supply." All liquidity moved to Uniswap v4 with the POOL4 hook in September 2026. (imd.fun/token)

• [R] How POOL4 works: a CappedBurnHook caps the IMD inventory in the pool, and IMD above the cap is trimmed after sells. The trim is split:

  • 85% burned
  • 6% to orchestrator compute
  • 4.5% to sIMD stakers
  • 4.5% to NFT seats

  The cap reportedly ratchets down by about 1,000 IMD per day. Circulating supply is reported at about 7.1M (Sep 25). (Bankless, KuCoin)

  • [U] I did not verify the split or the supply on-chain.

• [F] Market data (DexScreener, snapshot time), main ETH pool 0xb07d…abfb3:

  • price $13.71
  • 24-hour volume $7.23M
  • liquidity $3.53M
  • DexScreener market cap $54.8M

• [U] Sources disagree on market cap:

  • KuCoin: a "$39M all-time high" on Sep 26.
  • Bankless: about $69M on Sep 25 (7.1M × $9.73).
  • DexScreener: $54.8M today.

  The difference is probably whether supply is counted on one chain or on all three. Treat any single market-cap number as unreliable.

2.2 sIMD (staking)

• [F] The official page says: "Stake IMD. A share of the pool's burns. Each sIMD redeems for more IMD over time. No lock, no claim." (imd.fun/token)
• [R] sIMD is an ERC-4626 vault. About 2.3M IMD (about 32% of supply) was staked in late September. Vault ownership was reportedly renounced mid-September. (Bankless, KuCoin)
• [I] sIMD yield comes only from POOL4 trims, which come from sells. It is not linked to the job fees customers pay. sIMD holders are exposed to trading activity, not to how much work the swarm sells.

2.3 SIMD (a separate token, not sIMD)

• [F] "Super Intelligent Identity" (SIMD), contract 0xbb0c1f82a2ea0253ea3d91c2f05caded82133415. Its website (si-md.xyz) shows only the ticker, the X handle @SuperIMD_eth and the contract address.

• [F] On DexScreener its main pair is against IMD. The pair was created 2026-10-04 16:46 UTC. Snapshot figures: market cap about $0.71M, 24-hour volume on the IMD pair about $0.77M. (DexScreener API)

• [F] The swarm job log shows many paid jobs tagged [SIMD-THESIS] and [SIMD-COLLISION:…]:

  • Thesis jobs grade public tweets about IMD/SIMD against a rubric, with "Pay bar is quality ≥ 8".
  • Collision jobs ask agents to find collisions for truncated sha256, keccak256 or ripemd160 hashes.
  • These came mainly from wallet 0x9fadab91….

  (explorer, /jobs, example /jobs/95b0edef-…)

• [U] I cannot confirm who operates SIMD or how its payouts work.

• [I] SIMD looks like a community token built on top of IMD. It buys swarm jobs to run a content-reward program and puzzles. That is demand, but it is reflexive: an IMD-ecosystem token paying IMD to the IMD network to grade posts about IMD.

2.4 The 2,000 seats and ERC-8004

• [F] ERC-8004 ("Trustless Agents") is a Draft ERC, created 2025-08-13. It defines Identity (ERC-721-based), Reputation and Validation registries. It states that "Payments are orthogonal to this protocol." (EIP-8004)

• [F] The seat collection is 0x0000ec93127baa929e58e97dd0095a2bfb38ec1d. Seats are registered as agents in the ERC-8004 registry 0x8004a169fb4a3325136eb29fa0ceb6d2e539a432 through an adapter contract 0xde152afb…. (/health, /feedback/batches)

• [F] From the worker README:

  • "An unregistered token cannot connect for work."
  • "One NFT authorizes one active device."
  • The worker drives the owner's own Claude Code or Codex CLI and "uses its quota".
  • Contract and frontend work is offered only to workers running approved premium models.

  (Identity-md/worker)

• [F] The network writes reputation feedback in batches to the registry. One example entry has tag1: "verification:structural", value: 1, per agentId. (/feedback/batches)

• [F] Seat counts at snapshot time:

  • 2,000 seats held by 793 distinct owner wallets. The largest holder has 77 seats; the top 10 hold 17.9%.
  • 629 active enrollments, 618 connected daemons.
  • 650 seats have work records; 600 of them were active in the last 24 hours.

  (/swarm, /health)

• [R] Earlier seat counts: 270 enrolled on Sep 24 (mtezy write-up) and 380 on Sep 25 (Bankless). Enrollment has grown quickly.

• [I] The ERC-8004 reputation entries are written by the network operator about its own structural verification. They are not feedback from independent clients. This makes them portable identity records, but weak evidence of value delivered to outside buyers.

2.5 x402 payments

• [F] x402 is an open HTTP-402 payment standard, now maintained by the x402 Foundation. It uses PAYMENT-REQUIRED and PAYMENT-SIGNATURE headers. (x402)

• [F] IMD's own implementation (/requests/capabilities, docs):

  • Prices: seven paid actions (job.open, job.continue, launch.open, workflow.open, oracle.request, schedule.create, schedule.topup). All cost 0.5 IMD; schedules cost 0.5 IMD per run bought.
  • Payment flow: payment is on Ethereum mainnet with Permit2. The buyer signs twice, the Permit2 payment plus a separate EIP-712 quote approval. The docs say "A stock x402 client makes the first signature but not the second."
  • Payee: payTo is 0x4e0fa57b… for every action. That is the same address /health reports as the server's gas wallet.

• [F] Lifetime order counters from /health:

  Status	Count
  paid	1,203
  expired quotes	1,199
  failed	20 (19 payment_permission_expired)

• [R] Earlier paid-order counts: 38 on Sep 24 (mtezy) and 115 on Sep 25 (Bankless). That is about 10× growth in roughly 11 days, from a very small base.

• [I] Expired quotes about equal paid orders. The custom double-signature flow and the Permit2 setup likely lose many would-be buyers. Agent-to-agent buyers using a standard x402 client cannot pay without custom code.

• [U] I could not see where the IMD collected at payTo goes next: burned, sold to cover gas, or held.

2.6 How the paid swarm jobs are executed

• [F] A job is split into skill "nodes". Seats claim nodes over WebSocket and run them on their own LLM subscriptions. A verifier then judges each submission:

  • verifier-rerun: rebuilds and re-runs the checks (used for contracts and tests).
  • verifier-paths: checks only structure and allowed paths (used for websites, reports and media).

  The live catalog has 50 skills, of which 19 are reference-only. (/skills)

• [F] One instance each of the verifier, publisher and deployer is running (/services). Oracle attestations are signed by a single attester key, 0x5598aa91… (/oracle/requests).

• [F] Launch economics (docs):

  • The swarm always receives 10% of a launch token's supply: 2% split equally among wallets with accepted work on the launch, and 8% split equally among seats connected at admission.
  • Launch pools charge 1.25% per trade: 1% to the paying wallet and 0.25% to the network.

---

3. What the activity data actually shows

Metric (snapshot)	Value	Source
Accepted submissions, last 24h	87,082	/health
…of which in the two busiest hours	33,974 + 32,793 (76%)	/steps/hourly
Lifetime accepted (sum over seats)	877,967	/seats/records
Lifetime accepted oracle-assess	875,491 (~99.7%)	/skills
Lifetime oracle requests	7,260 (7,071 attested, 166 disagreed)	/oracle/counts
Lifetime jobs, all kinds	8,735	/swarm
Lifetime paid orders	1,203	/health
Inference tokens consumed	~177 billion	/swarm
Schedules ever created	10 (419 runs bought; 3 by operator)	/schedules
Last 100 launches by chain	98 Sepolia testnet, 2 Ethereum mainnet (both 2026-10-05); 25 parked	/launches

Oracle sample (60 most recent requests, fetched one by one):

• 55 used panelSize: 200, evidence: "chain". Most asked for Uniswap v4 spot prices on Robinhood Chain or swap counts on the mainnet PoolManager.
• 5 had panels of 5–60.
• [F] The paid route caps panels at 100 (capabilities), so the 200-seat panels were not bought through it. The docs say chain evidence means "the deployer reproduces the answer from chain data".
• [I] These are operator-generated tasks. They keep seats busy, produce measurable activity, and may serve as attendance for reward eligibility. Two hundred LLM agents agreeing on a value a script can read from chain proves little about useful capacity. The headline "86% acceptance / tens of thousands of submissions per day" mostly measures this activity.

Who pays (sample of 498 paid non-oracle jobs, 2026-10-03 21:52 → 2026-10-05 22:45 UTC, paidBy field from /jobs/:id):

Payer	Share	What they bought
0x9fadab91…	44.8%	SIMD-COLLISION puzzles, SIMD-THESIS tweet grading, some media. This report's own job was also paid by this wallet.
0x21919137…	28.5%	"IDENTITY UNITS" NFT art generation and review
0xbe87c01c…	7.6%	"SWARMBRAIN" rounds (research)
26 other wallets	19.1%	imdUSD branding and audits, "IMD Ember World" design and audit, token sites, a hackathon judge, etc.

• [F] 376 of the 500 most recent non-oracle job objectives contain "IMD" or "identity". This is a crude keyword test, but it is consistent with what I saw reading them.
• [I] Almost no observed paid demand comes from buyers with no stake in IMD. The buyers are mostly ecosystem tokens, NFT projects and community members. That pattern fits a token-community economy more than a work marketplace.

[I] Unit economics. At $13.71, a 0.5 IMD job costs about $6.9. A single paid job can fan out into many seat-runs: a launch ran 11 to 13 nodes, and an audit uses four specialists plus a judge. Each run uses the seat owner's premium-model quota. The customer's payment covers a small fraction of the market price of that compute. The remaining cost is carried by:

• seat owners, who pay for their own LLM subscriptions,
• token-based rewards: launch allocations and POOL4 trims.

That is a subsidy model, not a self-funding one.

---

4. Strongest real-world use cases (analysis)

Ranked by how well each one fits what the swarm does well: redundant independent agents, machine-checkable output, and on-chain delivery.

1. Attested oracle answers for long-tail questions, sold by subscription. (strongest)
   • Evidence of fit [F]: an active schedule, "Claus · London Heathrow · four-hour weather", pays per 4-hour run and names a consumer contract 0x37bfb8ac… (/schedules). The EIP-712 attestation format can be checked on-chain (docs).
   • Why [I]: Chainlink and UMA do not serve questions like "did it rain at EGLL?", "did this hackathon registry meet X?", or "which v4 pool led WETH volume?" well. The questions recur naturally (insurance triggers, prediction-market resolution, parametric payouts, KPI options), so revenue would recur.
   • Gap: the answer is ultimately only as trustworthy as one attester key and a panel of LLMs that may share the same blind spots.
2. Parallel smart-contract review: audit panels, fuzzing, adversarial review.
   • [F] An audit template (four specialists plus a judge) and fuzz routes exist. Paid audits appear in the log (imdUSD, Ember World).
   • [I] One buyer paying about $7 for several independent LLM audits is a real price advantage as a pre-audit or CI gate.
   • [U] No outside measurement of finding quality exists, and the review skills' public records show 0 accepted / 0 rejected with everything "pending" (/skills). Quality is not yet measurable from the outside.
3. Contract → review → deploy → website pipelines ("workflow.open", launches).
   • [I] Useful as a one-click scaffold for hackathons and small teams.
   • [F] Nearly all launches so far are on Sepolia.
   • [I] On mainnet, demand will track memecoin cycles, because the default product is a token plus a pool. That is cyclical, and regulatory exposure is high.
4. Research reports and media.
   • [F] These are the largest paid categories by count.
   • [I] These are also the most commoditized. Any buyer with a Claude or Codex subscription can get the same output directly. The swarm's verifier checks paths and structure, not truth. Weak defensibility.

5. What could create recurring outside demand for IMD (analysis)

• Schedules consumed by contracts. This is the only product with built-in recurrence: a schedule.create with runs, and a consumer contract that verifies attestations. To grow it:

  • publish consumer-side verification libraries,
  • publish dispute and accuracy statistics per question type,
  • move from a single attester to a threshold or multi-signer scheme.

• Prices that are stable in dollars. All prices are fixed at 0.5 IMD. The USD price went from about $4.9 (Sep 25, Bankless price) to about $6.9 now. Businesses budget in dollars. Two options:

  • quote in USD and settle in IMD,
  • accept USDC through x402 and buy back IMD.

  Either keeps IMD demand while making the product easier to buy. [I] The second option is the standard "fee buyback" design. It makes token demand follow usage instead of speculation.

• Standard x402 compatibility. If ordinary x402 agent clients could pay without the second EIP-712 signature, agent-to-agent buyers could use the swarm with no integration work. That is where "autonomous" demand would come from. Today about half of quotes expire unpaid.

• Quality evidence a buyer can audit. This means acceptance and rejection rates split by judge type, excluding operator-generated oracle panels, plus reputation feedback written by clients rather than the operator in ERC-8004. Without this, an outside buyer cannot price quality.

• Paying workers per job. [I] If part of each job fee went to the seats that did the work (as with the launch split), worker incentives would follow customer demand rather than trading volume. That is the key condition for "sustainable".

6. Biggest risks and weaknesses

1. Revenue is tiny compared with speculation. [F] About 600 IMD (about $8k) of lifetime paid orders against about $7M of daily DEX volume and a market cap of tens of millions. [I] The token is valued on expected future demand, not current demand.
2. Activity metrics are inflated by operator-generated work. [F/I] About 99.7% of accepted submissions are oracle-panel work, mostly 200-seat panels over chain-reproducible data. Headline acceptance and submission numbers should not be read as customer work.
3. Demand is concentrated and circular. [F] Three wallets account for 81% of sampled paid jobs. The largest is tied to SIMD, a token paired with IMD. [I] If one or two community programs stop, paid volume could collapse.
4. Worker pay depends on trading, not customers. [F] Job fees go to the server or gas wallet. [R/F] Seat rewards come from POOL4 sell-trims, launch-token allocations and launch-pool fees. [I] In a bear market, falling volume cuts worker rewards at exactly the time real demand is needed.
5. Centralization.
   • [F] One control plane, one verifier, one publisher, one deployer, one oracle attester key.
   • [F] A /feedback/batches writer controlled by the operator.
   • [F] Workers that auto-update from GitHub releases.
   • [I] The operator is a single point of failure and trust, despite the "community-owned" framing. If the auto-update channel were compromised, code could run on more than 600 machines that hold LLM credentials.
6. Dependence on consumer AI subscriptions. [F] Seats run on owners' personal Claude or Codex subscriptions, and premium models are required for contract work. [U] Whether those providers' terms allow consumer subscriptions to be resold as a work service is not established here. If a provider restricts this, capacity could disappear overnight. The model requirements in the README also change as model versions change.
7. Verification checks form, not value. [F] verifier-paths skills (reports, sites, media) check structure and integrity only; the README says the same. [I] A report can be accepted and still be wrong. This report is an example of that kind of job.
8. Security of running untrusted job text. [I] Workers run objectives written by anyone on machines holding model credentials. Prompt injection and data exfiltration are inherent risks, partly reduced by sandboxing that varies between seats.
9. Token and regulatory risk. [R] The project is unaudited and experimental, and its history is a rebranded game token. [I] The launchpad issues tokens and pays fees to the requester, which could draw securities or commodities scrutiny in some jurisdictions.
10. Standards are immature. [F] ERC-8004 is still a Draft, and IMD uses a non-standard extension of x402 (the second signature).

7. Open questions

• [U] Where does the IMD at payTo (0x4e0fa57b…) end up? Is any of it burned, distributed or sold?
• [U] Exact POOL4 trim split, cap schedule and cumulative burns. These need on-chain checks of hook 0xc6C965Bd… (search result, unverified).
• [U] Real per-seat income in IMD and USD per month after LLM subscription costs (/wallets/:address/earnings exists but I did not survey it).
• [U] Who opens the 200-seat oracle panels, and are they tied to seat reward eligibility?
• [U] Who operates SIMD, and is its job spending funded by trading fees or by a treasury?
• [U] Accuracy of attested oracle answers against ground truth, and finding precision for audits.
• [U] Whether 1,203 "paid orders" counts each schedule purchase once or each run. Schedule runs add at most about 210 IMD (419 runs × 0.5).

8. Verdict (analysis)

[I] Identity.md is a working, unusually transparent marketplace system: public API, real x402 settlement, on-chain identities, a reproducible verifier. Its technical base could support a market for autonomous work. As of 2026-10-05, though, it is a token-community economy with a small, concentrated, mostly internal paid-demand layer. The workers are paid through trading activity, not customer revenue.

For it to become a sustainable market, three numbers need to move together:

• paid orders from wallets with no IMD-ecosystem stake,
• the share of job fees reaching the workers who did the work,
• recurring scheduled oracle or job runs consumed by outside contracts.

None of these is currently large enough to measure meaningfully.

Sources

Primary (live, read 2026-10-05):

• IMD API: /health, /swarm, /skills, /jobs, /oracle/requests, /oracle/counts, /launches, /schedules, /requests/capabilities, /services, /steps/hourly, /seats/records, /feedback/batches
• IMD API docs: https://imdnotdead.fun/docs · Token page: https://imdnotdead.fun/token · Explorer: https://explorer.imdnotdead.fun
• Worker client: https://github.com/Identity-md/worker
• ERC-8004: https://eips.ethereum.org/EIPS/eip-8004
• x402: https://github.com/x402-foundation/x402
• Market data: DexScreener IMD, DexScreener SIMD, si-md.xyz

Secondary:

• Bankless, "Inside IMD, Ethereum's New AI Swarm Experiment" (Sep 25, 2026): https://www.bankless.com/read/inside-imd-ethereum-s-new-ai-swarm-experiment
• KuCoin, "What Is IMD Token?": https://www.kucoin.com/blog/id-imd-token-community-owned-ai-agents
• TokenPost (Sep 25, 2026): https://www.tokenpost.com/news/technology/24260
• mtezy/imd-writeup (third-party, Sep 24 snapshot): https://github.com/mtezy/imd-writeup