Ethereum officially includes and launches two Finch AI agent standards proposals

On October 15 (Beijing time, UTC+8), the official Ethereum ERC repository accepted two Ethereum standards proposals put forward by the Finch team: ERC-8436, the AI Agent Collective Decision (ACD) Framework, and ERC-8444, Agentic Resource Discovery (ARD) Onchain. Both proposals are now live on the official Ethereum EIP site and can be cited by number in other Ethereum proposals. ERC-8444 went from receiving its number to being accepted in less than two days.

ERC-8436 AI Agent Collective Decision and ERC-8444 Agentic Resource Discovery proposals

Figure 1: ERC-8436 and ERC-8444

The two proposals are a key part of a full suite of standards proposals the Finch team has put forward for the AI agent economy. The seven proposals — ERC-8338, ERC-8414, ERC-8419, ERC-8434, ERC-8436, ERC-8444 and ERC-8447 — were authored by Finch Founder and CEO Gary Yang. They cover AI agent skills, tasks, trust, identity, collective decisions, resource discovery and swarm organization, forming one interlocking protocol stack, and all seven have now been assigned Ethereum ERC numbers.

The seven Finch ERC proposals and their progress

Figure 2: The seven ERCs and their current progress

What is ERC-8436: the AI Agent Collective Decision Framework

As swarm AI becomes a leading way to accomplish deep intelligent tasks, the relationships among AI agents, their rules of deliberation, their voting and decision mechanisms, and their principles for coordinated execution form the foundational relations of production for an evolving AI society. ERC-8436 defines that foundation on-chain: bodies of AI agents, humans or contracts can form collective decisions with defined effect, within explicit authorization, under verifiable and composable rules. Its key design principles:

Rules frozen as parameters: Every decision policy is identified by the hash of the very struct the registry executes, leaving no gap between the rule as published and the rule as applied.
Authorization precedes eligibility: Reputation or stake may qualify an AI agent for a panel, but never determines what the panel may decide; a result takes effect only where a relying contract committed in advance to accept it.
Composable multi-body decisions: Bodies combine through ALL, ANY, K-of-M and VETO, e.g. “technical and economic chambers must both approve, while a screening council holds a 12-hour veto.”
“No decision” is a first-class outcome: A missed quorum or a silent submitter is never disguised as approval or rejection.
Decision separated from execution: The registry is the normative record and never moves assets itself.
Procedural finality: Bounded appeals, a hard deadline, and a permissionless exit from every state.

Official page: https://eips.ethereum.org/EIPS/eip-8436

What is ERC-8444: ARD Onchain

ARD is an open resource-discovery specification launched by Google. Publishers list their MCP servers, A2A agents, skills and APIs under their own domains; registries crawl and index them; agents search in natural language. But ARD only answers “what exists and who offers it.” It covers no price, payment or settlement, and its entries run in one direction only: supply.

ERC-8444 adds the on-chain value layer on top of ARD and is compatible with Finch's ERC-8338 (executable skills) and ERC-8414 (task tenders). It lets on-chain skills, tasks and AI agent identities be found directly through ordinary Web2 ARD search, complete with price and settlement terms. Its key design principles:

Every record is a standard ARD entry: An ARD registry that knows nothing about the on-chain extension still indexes it as an ordinary entry. Searchers need no wallet and no blockchain connection.
Anchored directly to ERC-8338 and ERC-8414: Skill records anchor to skill tokens and task records to task-tender tokens. A record reaches the top “backed” level only when its package digest matches the hash locked on-chain.
Value root = settlement root: Control of a record belongs to the on-chain address that gets paid, not to a domain; a domain is only an endorsement an address can collect.
The first demand-side ARD entries: Once task tenders are in ARD, AI agents looking for work can find demand too.
Independently verifiable on-chain: Descriptors are committed on-chain by hash, so any copy served by anyone can be checked against the chain.

Official page: https://eips.ethereum.org/EIPS/eip-8444

Finch: Building the Relations of Production for the AI Agent World

Once AI agents become independent economic actors, they need more than capabilities. They need a complete set of relations of production: how skills are owned and traded, how demand is posted and settled, how identity and trust are verified, how many agents deliberate and decide, and how resources are found.

Any single proposal can solve only one of these links. Finch designed them as a whole. The seven proposals share the same design principles — identity is fixed by hashes and addresses, transport locations are replaceable hints, publication, ownership and judgment are separate rights, and trust conclusions must carry validity windows — so they snap together directly.

“Finch is building an AI-native ecosystem of AIFi and AI Market services. The AI world will not run its finance, economy and relations of production out of human habit; it needs an independent, efficient system of production and interaction of its own. The AI protocol standards we have initiated will provide the foundational framework for how the AI world operates, and the AIFi and AI Market we are building will mature on top of them into the native ecosystem for the AI agent economy and A2A autonomy.”

— Gary Yang, Founder and CEO of Finch (Finchtech.AI)

The seven proposals at a glance

ERC-8338 Token-Bound Executable Skills: An ERC-721 extension that binds a skill token by hash to its exact skill package, so a runtime can verify it byte-for-byte before executing. Publication rights are separate from ownership: transferring the token does not transfer the right to change the skill. This is the supply side of the AI agent economy.
ERC-8414 Token-Bound Task Tenders: The demand-side mirror of ERC-8338. It turns a task into a transferable token whose bounty is escrowed in the token's own vault and can leave only through settlement or refund. Acceptance authority is on-chain and can be a verifier contract, a single address or a multi-member panel.
ERC-8419 Know-Your-Agent (KYA) Framework: A common on-chain registry and handshake for trust conclusions about AI agents. It does not prescribe how agents are verified; it makes conclusions from different providers mutually readable, supports assertions admitted by zero-knowledge proof, and binds to ERC-8004 agent identities.
ERC-8434 Agent Identity (AID): Address-anchored identity for AI agents. Every address is a dormant AID that becomes active when a live AI agent is bound to it and signals liveness within a window. Every facet of the profile carries its provenance, validity window and access mode, and can be re-verified on-chain by anyone.
ERC-8436 ACD Framework (AI Agent Collective Decision Framework): How many AI agents deliberate, vote and reach collective decisions with real effect. See “What is ERC-8436” above.
ERC-8444 ARD Onchain: Makes on-chain skills, tasks and AI agent identities directly searchable through ARD. See “What is ERC-8444” above.
ERC-8447 Swarm AI Framework (AI Agent Swarm Framework): Gives AI agent swarms an organizational structure and defined collaboration relationships.

Figure 3: How the seven proposals fit together

Seven Proposals, One System for AI Governance and Agent-Economy Standards

Supply and demand: ERC-8338 (skills) and ERC-8414 (tasks) mirror each other with a compatible object model. When a deliverable is itself a skill, the task can settle directly against that skill's package hash.
Decisions and acceptance: ERC-8414's acceptance authority can be handed to an ERC-8436 collective decision. The ERC-8436 reference implementation already includes an adapter for task-tender contracts, for example letting a 3-of-5 AI agent panel decide whether a delivery is accepted.
Identity and trust: ERC-8434 draws on ERC-8419 trust assertions and derives an AI agent's skill and task history directly from ERC-8338 and ERC-8414 contracts, with no extra registration.
Discovery: ERC-8444 brings ERC-8338 skills, ERC-8414 tasks and ERC-8434 agent identities into ARD search.
Organization: ERC-8447 ties the others together: a swarm can hand its decisions to ERC-8436, take on an ERC-8414 task as a whole, and admit members based on ERC-8419 trust assertions and agent identities.

ERC-8436 and ERC-8444 now enter community review on Ethereum Magicians; Last Call and Final will follow once the community converges on their frameworks. The other five proposals, including the newest, ERC-8447, are progressing through the same process. All discussion takes place openly on Ethereum Magicians, and developers, researchers and AI agent builders are welcome to join.

About Finch

Finch is the marketplace where AI value is discovered and exchanged — built on AI-native economic protocols and agent-native autonomous interactions.

❖ Finch — Make Markets AI-Native.

MEDIA CONTACT

Marketing@finchtech.ai

Generator