AI Cluster Experiment on Ethereum: How IMD's Destruction Mechanism Works?

From electronic pets to AI swarms: community companies run entirely by agents.

Written by: @wmpeaster, Bankless Newsletter

Compiled by: AididiaoJP, Foresight News

A rapidly warming AI project IMD (identity.md) has recently emerged in the Ethereum ecosystem. Its goal is to build a company operated by AI agents and owned by the community. The project mechanism includes NFT seats, agent swarms, IMD/sIMD, Uniswap V4 destruction Hook, and Community Coins launchpad. After the agent network opened to NFT holders on September 20, the number of nodes increased from dozens to over 370 within five days, and IMD has risen over 200% in the past week.

The core logic of IMD is that swarm agents may become a labor force that other protocols are willing to pay for continuously, such as reasoning oracles or other broader applications. To understand this mechanism, it needs to be broken down into several modules.

Background

IMD founder Adam (@surfcoderepeat) previously launched the on-chain electronic pet game Fren Pet on Base, which went live in August 2023. In October 2025, he bridged FP to Ethereum via LayerZero. Months later, FP was renamed to VIBE; in May 2026, it was renamed to IMD, and 2000 identity.md NFTs were minted for free.

Recently, the project's pace has accelerated. Liquidity was migrated to Uniswap V4 last month; in September, the POOL4 destruction Hook, sIMD staking, the bridge to Robinhood Chain, and the agent network with NFT thresholds were successively launched. Subsequent progress is mainly disclosed on-chain: Adam updates from transaction notes sent from address 0x200E710aCAA6A93bbc77146026328C40F1d60fB1, which can be directly read on Etherscan.

How IMD Operates

1) NFT Seats

2000 identity.md NFTs, each corresponding to a seat in the swarm. Holders must first register the NFT as an agent according to the ERC-8004 standard, which provides on-chain identity and reputation for AI agents on Ethereum. Then, they need to install the open-source worker client on their machines, preferably using a VPS, and connect to their own Claude or Codex subscription.

One NFT only authorizes one active device, and the computing cost is borne by the holder. High-value tasks, such as smart contract and frontend development, are currently only allocated to seats running top-tier models, such as Claude Fable 5.1 high-performance tier.

2) Swarm

The main scheduling agent publishes tasks to the network, and seats claim, execute, and submit results. Validators will rerun tasks in isolated containers to confirm that only allowed changes are made. Other seats conduct adversarial reviews, and the approved results will be recorded on-chain for reputation.

Outputs can be viewed in a public browser, including code, websites, oracles, audits, reports, and images. Current code outputs mainly test the Uniswap V4 Hook on the testnet, websites are published to IPFS, and oracles consist of independent answering groups of multiple agents.

Starting this week, external users can also purchase work. Anyone can open an order by paying 0.5 IMD, settled through x402. Currently, deployment is still only on the Sepolia testnet, with mainnet deployment scheduled.

3) $IMD and sIMD

$IMD is the currency of this economy. The total supply is fixed, distributed across Ethereum, Base, and Robinhood Chain, with a 1:1 bridge, and the total amount will only decrease. From the destruction accumulated since the Fren Pet period, the initial 10 million has decreased to about 7.1 million, a reduction of about 29%.

Holders can deposit IMD into the StakedIMD vault to exchange for sIMD. sIMD is an ERC-4626 standard interest-bearing share, redeemable for more IMD, with no lock-up and no need for separate claims. Currently, about 2.3 million $IMD is staked, accounting for about 32% of the total supply. On September 19, Adam relinquished ownership of the staking contract, and administrators can no longer execute emergency withdrawals for stakers.

4) POOL4

The protocol itself holds the main ETH/IMD pool on Uniswap V4, namely POOL4, operated by CappedBurnHook. This Hook sets a cap on the amount of IMD in the pool. When sell orders push the amount of IMD in the pool over the cap, the Hook will trim the excess when the transaction settlement is completed.

Of the trimmed tokens, 85% are destroyed, 6% are reserved for scheduling computing power, 4.5% are allocated to stakers, and 4.5% are allocated to NFT seats. The ETH freed up from trimming will be used to create buy walls below the current price. The cap is adjusted down by about 1000 $IMD daily to avoid stagnation. Sell orders drive destruction, and destruction feeds back to stakers and seats.

5) Community Coins

Community Coins are IMD's own launchpad. Anyone can issue coins by paying gas, with a total of 1 billion for each coin, priced in $$IMD rather than ETH. Transactions appear to be conducted in ETH, but each purchase is essentially buying $$IMD.

In each transaction, 1% is allocated to ETH/IMD liquidity providers, 0.5% on the ETH side goes to the issuer, and 0.5% on the $IMD side is destroyed. These new coins share the same liquidity pool, so buying any one will raise the others, and selling any one will lower the others.

Flywheel and Current Status of the Swarm

According to the current design, the ideal cycle is: users trade IMD and Community Coins; POOL4 trimming destroys supply, benefiting stakers and seats; scheduling agents convert demand into marketable products; useful outputs attract more people to issue coins priced in IMD, purchase work, and increase demand for seats; scarcity of seats raises NFT prices, and demand and destruction raise $IMD.

The key question is: can the cornerstone of the flywheel—the swarm—actually work? According to the public API data from IMD on September 25, of the 2000 seats, about 380 had registered, with 372 online agents. Among them, 334 registered seats had submitted accepted work, 267 had over 50 accepted records, and 193 had over 100. The top 10 seats accounted for about 8.5% of the accepted workload, while the top 50 accounted for about 32%.

Quality data is decent. Of about 50,700 attempts, 86% were accepted, with direct rejections at only about 1%, while the rest were failures or queued. The pace is also accelerating, with about 29,600 accepted submissions in the past 24 hours. Paying customers have just entered: the public x402 channel currently has 115 transactions, each for 0.5 IMD, totaling about 57.5 $IMD, approximately $560. Paying demand has emerged, but the scale is still in the early stages.

Participation Path

If you are interested in this project, IMD is a relatively direct way to participate. It is a fungible token with a low trading threshold, adaptable to different capital scales; the seat price is about 1.99 ETH. If you judge that trading volume and destruction can be sustained, and that the swarm's paid demand can expand, IMD can be seen as an exposure to the corresponding token reservoir. This bets on the token mechanism rather than agent ownership. You can also stake it for sIMD later, but the yield depends on whether there is sell-side flow into the POOL4 Hook.

The NFT side is more suitable for advanced participants: you can run a node yourself or invest about 1.99 ETH (the floor price will fluctuate) to bet on seat capacity and scarcity. Current income from seats is limited, but it may change in the future, making it worth continuous tracking. Its nature is similar to an option on future wages, while also occupying potential positions for whitelists, airdrops, etc. If you find the project mechanism complex, or judge that it only remains an on-chain experiment and is difficult to expand outward, choosing to wait and see is also reasonable.

Conclusion

Can IMD reach the market value of other leading agent projects in the hundreds of millions, such as Virtuals? As a reference, VIRTUAL has retraced about 85% from its historical peak, yet its market value is still around $508 million. For IMD to catch up from its current position, it would need to rise about 7 times.

The two are not entirely comparable. Virtuals issue tokens for each agent separately; IMD tokenizes a limited base of agents along with the entire system. Future observation is needed to see how the market aggregates towards IMD. If it truly becomes a labor swarm that more and more protocols are willing to pay for, significant increases are not impossible; if it is merely an on-chain experiment with strong destruction reservoirs, the ceiling will be much lower.

Regardless of the outcome, IMD has already become one of the most watched AI experiments in the Ethereum ecosystem by the end of the year. If activity continues to rise and market momentum persists, it may become one of the trading targets worth tracking in 2026.