Features · AGI Future Foundation
The Agent Economy
So far the agents in this series have been workers: drafting, watching, coordinating, preparing. This feature is about the moment they stop being only workers and start being economic actors — holding an identity others can check, paying in tiny increments for what they use, inside companies run by agents. It is a design problem with three hard parts, and two projects in the Foundation’s network are trying to solve them.

When workers become actors
There is a difference between an agent that does work and an agent that takes part in an economy, and the difference is not a matter of degree. A worker receives instructions and produces output. An economic actor holds something of its own: an identity that others can verify, a balance it can spend, obligations it can incur, a reputation that follows it from one transaction to the next. The moment an agent can pay for a service, be paid for a result, enter an agreement, or be held to a commitment, it has crossed from tool to participant. That crossing is the subject of this feature — the most consequential idea in the Foundation’s whole architecture, because an economy of agents that works would change what companies are, and one that fails would change them in worse ways.
The reason the crossing matters is scale. A single agent doing a single task is a productivity tool. Thousands of agents, each buying and selling services from each other — one agent paying another for data, for compute, for a verification, for a draft, for a check — is a market. Markets need what workers do not: identity, so you know who you are dealing with; payment, so value can move in increments small enough to match the work; and governance, so that when something goes wrong there is someone accountable and some way to stop it. Without identity, an agent economy is a crowd of strangers with no way to tell a reliable counterparty from a malicious one. Without payment at the right granularity, agents must route every tiny exchange through slow, expensive human payment rails that cost more than the transaction is worth. Without governance, an agent economy is a machine with no brakes and no owner.
Those three needs — identity, payment, governance — are the skeleton of this feature, and the two projects in it each carry a piece of the skeleton. Metropolis-SmartCiti’s repository description names all three at once: ERC-8004 agent identity, x402 micropayments, and 2-of-3 passkey steward governance on Monad. Insidr’s description names the institutional form that would hold them together: an agentic corporate manager, a DAO that manages a company the way a manager would, with agents doing the managing. One project is the plumbing; the other is the organisation that would live in the plumbing. Together they sketch an economy in which agents are not just labour but participants, and in which the participation is fenced on every side by human stewards who hold the keys.
Before going further, a caution that belongs at the start rather than the end. An agent economy sounds exciting, and excitement is the enemy of accuracy when the record is thin. The repository descriptions quoted in this feature are real and verified as descriptions. Everything beyond them — whether the systems run, who uses them, what has been transacted, whether the governance works — is unverified, and this feature will not fill those gaps with invented detail. The depth in this feature comes from the ideas themselves and from the questions they raise, not from pretending the builds are further along than they are. An economy that exists only in a description is a design, and designs deserve to be examined seriously, on their own terms, with their holes showing.
Metropolis-SmartCiti: identity, micropayments, and a two-of-three gate
Metropolis-SmartCiti’s repository description, as listed under the AGIFutureFoundation organisation, reads: Omega Stewards: ERC-8004 agent identity, x402 micropayments and 2-of-3 passkey steward governance on Monad. Every element of that sentence is a technical choice, and each choice answers one of the three needs named above. ERC-8004 is a standard for agent identity: a way for an agent to have a verifiable identity on-chain, so that other agents, systems, and humans can check who — or what — they are dealing with, what it is authorised to do, and what its record is. x402 is a protocol for micropayments: a way for value to move in tiny increments, priced per request, per call, per unit of work, so that an agent can pay for exactly what it uses without the overhead of a human-scale transaction. Two-of-three passkey steward governance is the human gate: three stewards, any two of whom must approve with passkeys, before the consequential things happen. Monad is the chain the whole design runs on.
Identity first, because it is the foundation the other two stand on. In a human economy, identity is mostly solved by institutions: a company has a registration, a person has documents, a bank knows its customer. Agents have none of that by default. An agent is a process, a prompt, a set of weights and instructions, and it can be copied, impersonated, or spun up fresh a thousand times a second. If you cannot tell one agent from another, you cannot build a reputation system, you cannot hold anyone to a commitment, and you cannot tell whether the agent offering you a service is the one that did good work yesterday or a fresh copy with no history and no accountability. ERC-8004 is an attempt to give agents the thing humans take for granted: a stable, verifiable identity that persists across transactions and can be checked by anyone. The description names it as a component of the design, not as a deployed system, and this feature reports it that way. Whether any agent holds such an identity in a live system, and what that identity has actually been used for, is unverified.
Micropayments second, because identity without payment is a directory, not an economy. The problem x402 is aimed at is granularity. An agent that needs a single fact, a single verification, a single short computation should be able to pay for exactly that, at a price that matches the value of one small unit of work. Traditional payment rails cannot do this: the fees, the settlement time, and the administrative overhead of a card payment or a bank transfer dwarf the value of a transaction worth fractions of a cent. So agents either transact in bulk through human intermediaries, which defeats the point of autonomous operation, or they do not transact at all, and the economy never forms. A micropayment protocol is the attempt to make the transaction cheap enough, fast enough, and automatic enough that agents can pay each other the way programs call functions: constantly, in tiny amounts, without asking a human to approve each one. The description names x402 as the protocol for this. No transaction, volume, price, or counterparty is verified in the record used here, and none is claimed.
Governance third, because an economy of verified, paying agents is also an economy that can go wrong at machine speed, and the design knows it. Two-of-three passkey steward governance means three human stewards hold passkeys, and any two of them must approve before a consequential action proceeds. Passkeys matter here: they are a modern, phishing-resistant form of authentication, tied to a device and a person, harder to steal or share than a password. Two-of-three matters more: it means no single steward can act alone, and no single steward’s absence, compromise, or refusal can freeze the system. It is the same pattern the Foundation’s governance model uses everywhere — three seats, two-of-three signed votes, humans decide — applied here to the economic layer. The stewards are the gate between the agent economy and the real world: agents propose, transact, and prepare within their limits, and the stewards hold the keys to the things that matter. No steward is named, no council is verified as seated, and no governance action is recorded in the material this feature draws on. The design is stated; the operation is not.
Monad, the chain named in the description, is the substrate: the blockchain on which the identity, the payments, and the governance would run. What matters for the story is that the design is on-chain by intent: identity, payment, and governance are meant to be verifiable by anyone, not locked inside a single company’s database. An agent economy whose records only one party can see is a private ledger with extra steps.
Here the record gets genuinely concrete, and the concreteness has a label that must stay attached to every figure in this paragraph: testnet. Metropolis-SmartCiti was deployed on the Monad testnet as a Monad Metropolis Track 04 hackathon deployment. The verifiable on-chain testnet record shows 6 contracts, 41 of 41 transactions confirmed on testnet, 12 agents registered, 11 x402 payments, and 4 credentials issued. That is a real deployment with a real, checkable record — contracts live, transactions confirmed, agents registered, payments made, credentials issued — and it is worth celebrating as exactly what it is. It is testnet only. It is not a live production network, it is not users, and it is not revenue. No production deployment, no production user, and no dollar of revenue is claimed, and this feature labels the testnet record as testnet every time it mentions it, so that no reader mistakes a hackathon deployment for a business.
One separation must be stated clearly, because the names invite confusion. Metropolis-SmartCiti is not S10 SmartCiti.X, and the two must not be merged. They share the words SmartCiti and a city-scale direction. They do not share a verified identity. S10 SmartCiti.X is W3 Series 10 in the April 2026 internal list, with a developed plan for a proposed new corporation: the Trade Craft Academy, robotics training, union trades taught with AR/VR, robotics manufacturing, a suggested $25 million pre-money valuation and a suggested $10 million tranched raise, and a Louisiana, Nebraska, and Indiana geography. None of that plan is verified as belonging to Metropolis-SmartCiti. No document reviewed for this site states that Metropolis-SmartCiti is Series 10, is the same venture as SmartCiti.X under another name, is a subsidiary of it, is a customer of it, or shares its funding, team, or sites. Merging them would attach a $25 million suggested valuation and a $10 million suggested raise to a venture that has no verified plan at all, and this feature keeps them apart.
Insidr: the company that agents manage
Insidr’s repository description, as listed under the AGIFutureFoundation organisation, reads: AGI Future Foundation Agentic Corporate Management DAO. The name on the repository is Insidr-W3LLC-Agentic-Corporate-Manager, and the description is the whole of the verified substantive record for the project. An agentic corporate management DAO is a specific and ambitious idea: a decentralised autonomous organisation that manages a company, with agents doing the managing. Not a DAO that votes on proposals while humans run the company, and not a company that uses agents as tools while humans manage. A DAO in which the management function itself — the coordinating, the record-keeping, the proposing, the tracking, the preparing of decisions — is carried out by agents, inside a structure whose rules are written down and whose consequential decisions are gated by humans.
To understand what that means, start with what corporate management actually consists of, stripped of its mystique. A company is a bundle of obligations and records: who owes what to whom, what was decided and when, what money came in and went out, what was promised and whether it was delivered, what the rules are and whether they were followed. Managing a company is mostly keeping that bundle coherent: making sure the records match reality, the decisions are documented, the obligations are tracked, and the people who need to know something can find it. Much of that work is routine, repetitive, and well-suited to agents: drafting, filing, reconciling, reminding, summarising, cross-checking. The agentic corporate manager is the proposal that this routine management layer can be run by agents, continuously, at a scale and a speed no human management team could match, while the decisions that change the bundle — spending, signing, hiring, issuing, committing — remain with humans.
The DAO part of the description is doing real work too. A DAO is not just a company with a chat group. It is an organisation whose rules are explicit, whose membership and voting are defined, and whose operations are meant to be transparent to its participants. A corporate management DAO, in this design, would make the management of a company legible: proposals visible, decisions recorded, authority bounded, and the whole state of the company checkable by those with a right to check it. That legibility is the point. A company managed by agents in a black box would be ungovernable, because no human could tell what the agents had done or why. A company managed by agents inside a DAO structure, with proposals, votes, records, and gates, is a company whose management can be audited — by stewards, by auditors, by anyone with standing. Whether Insidr implements any of this in running code is unverified, and this feature does not claim it does. The description names the ambition; the record does not show the implementation.
The name carries its own caution, and the local project page for Insidr states it directly. The W3LLC in its name does not by itself prove any connection to a filed W3 LLC. No Wyoming filing for W3 LLC is verified in the files reviewed for this site, no series designation is verified, and no document connects Insidr to any such filing. Until a filing is cited, the W3LLC is part of a name, not evidence of a legal form. The repository is thin as well: its README is two lines, title only, and the companion insidr-site is a stub. Insidr should not be described as a built system. It is a named management-DAO design with a landing page, and this feature presents it that way. The same discipline applies to the Insidr name itself, which points at inside knowledge — seeing what is inside a company, a system, or a market — but a suggestive name is not a description, and this feature does not treat it as one. What is verified is the repository, its name, and its description. What is not verified is everything else: no legal form, filing, entity number, EIN, or bank account; no product, service, site, address, or geography; no team, founder, director, partner, customer, or user; no funding, revenue, valuation, terms, or timeline; no relationship to any series beyond what a document states, and no such document is cited here.
Put Metropolis-SmartCiti and Insidr together and the shape of the intended economy comes into focus. Metropolis-SmartCiti is the plumbing: identity so agents can be known, micropayments so agents can transact, steward governance so the plumbing has a human gate. Insidr is the organisation that would live in the plumbing: a company whose management is agentic, whose records are legible, and whose decisions are gated by the same kind of human council. An agentic manager without identity, payment, and governance primitives is a manager with no way to know who it is dealing with, no way to move value, and no way to be stopped. The pairing is the design. Whether either half runs is, on the current record, unproven.
The three primitives: identity, payment, governance
Identity, payment, and governance are not three features on a list. They are three primitives, in the sense that each one is required for the others to mean anything, and the failure of any one collapses the whole. An agent economy with identity and payment but no governance is a market with no brakes: agents can transact freely and move value, and when one of them goes wrong there is no one with the standing and the mechanism to stop it. An economy with identity and governance but no payment is a registry with a committee: everyone is known, someone is in charge, and nothing can actually be bought or sold at the granularity the work requires. An economy with payment and governance but no identity is a market of strangers with a referee: value can move and someone can intervene, but no one can tell a reliable counterparty from a fresh impersonation, and reputation cannot accumulate.
Identity deserves the deepest treatment, because it is the primitive most easily underestimated. In human economies, identity is backed by the state, by banks, by employers, by a whole apparatus of documents and institutions that make it costly to lie about who you are and possible to find you if you break a commitment. Agents have no equivalent apparatus by default. An agent’s identity, in the ERC-8004 vision named by Metropolis-SmartCiti, would be a persistent, verifiable record: this agent, with this history, authorised for these actions, by these principals, with this reputation. Without it, every transaction is a leap of faith, and an economy built on leaps of faith does not scale — it just gets faster at being defrauded.
But identity also raises the question of who issues it, and that question has no comfortable answer yet. If identity is issued by a single company, then that company is a gatekeeper with the power to exclude. If identity is self-asserted, then it is cheap to forge and worthless as a basis for trust. If identity is issued by a network of stewards, then the stewards’ own legitimacy, selection, and accountability become the real governance question, one level up. ERC-8004 as a standard does not settle these questions by itself; a standard is a format for expressing identity, not an answer to who vouches for it. The Foundation’s model points at steward governance as the answer, but the specific design for who issues agent identity, under what rules, and with what recourse when identity is misused, is not in the verified record, and this feature does not invent it.
Payment raises its own version of the same problem, at a different scale. Micropayments make agent-to-agent transactions possible, but they also make agent-to-agent transactions invisible in aggregate. A thousand transactions of a fraction of a cent each is a trivial sum and a rich data trail. Who sees that trail, who can audit it, and who is responsible when the trail shows something wrong are governance questions that a payment protocol alone does not answer. The x402 design named by Metropolis-SmartCiti is about the mechanics of moving value per request. The economics around it — who sets prices, how prices are discovered, what happens when an agent overspends, whether an agent can incur debt — are the questions that turn a protocol into an economy, and none of them is answered in the record this feature draws on. An agent that can spend without a budget is a liability, not an asset, and the budget is a human decision, not a protocol feature.
Governance, the third primitive, is where the other two are fenced. Two-of-three passkey steward governance is a strong pattern: no single steward can act alone, no single steward can block everything, and passkeys make the authentication itself resistant to the most common attacks. But the pattern raises the questions it does not answer. Who are the stewards, and how are they chosen. What are the limits inside which agents may act without steward approval, and who sets those limits. What happens when stewards disagree, or when two stewards collude, or when a steward’s passkey is lost or stolen. How often stewards review what agents did, and what records they review. The Foundation’s governance pages give the general shape — three seats, two-of-three signed votes, humans decide, agents prepare — but the specific answers for Metropolis-SmartCiti and Insidr do not exist in the verified record. A gate with no one named as its keeper is a design, not a control.
The human gate: who holds the keys
Every design in this feature ends at the same place: a human holding a key. The stewards in Metropolis-SmartCiti hold passkeys. The Steward Council in the Foundation’s governance model holds decision authority. The managers in Insidr’s DAO, for all their agentic operation, answer to humans who decide, sign, and move funds. That repetition is not an accident. It is the deliberate centre of the whole architecture: agents operate, people govern. The agent economy is not an economy of agents governing themselves. It is an economy of agents operating inside limits set by humans, with humans holding the keys to the consequential things.
The reason for the repetition is that the alternative is genuinely frightening, and the Foundation’s documents are unusually clear-eyed about it. An economy in which agents hold identity, move value, and manage companies without human gates is an economy in which a compromised agent, a mistaken instruction, or a misaligned goal can propagate through thousands of transactions before anyone notices. The speed that makes the economy valuable is the same speed that makes its failures catastrophic. The human gate is the brake: slow, deliberate, and deliberately slower than the system it governs, because judgment is supposed to be slower than execution. A gate that operated at machine speed would not be a gate; it would be another agent, with the same failure modes as the agents it was supposed to watch.
But a gate is also a bottleneck, and the tension between safety and speed is the real design problem of the agent economy. If every transaction needs steward approval, the economy grinds to a halt and the agents might as well be human employees with extra steps. If no transaction needs approval, the gate is decorative and the economy is ungoverned. The answer, in the Foundation’s model, is limits set in advance: agents may act freely within bounds humans have defined — up to a value, within a scope, for a class of actions — and must escalate beyond those bounds. The bounds are the design. Set them too tight and nothing happens. Set them too loose and the gate means nothing. No set of bounds for Metropolis-SmartCiti or Insidr is published in the record used here, and this feature does not invent one. The bounds are the work still to be done, and they are the work that will decide whether the design works.
The agents-never rule is the hard edge of the gate, and it applies here as everywhere in the Foundation’s architecture. Agents never sign, move funds, hire or fire, offer securities, or enact doctrine. Humans execute; agents prepare. In the context of an agent economy, that rule has a sharp consequence that is worth stating plainly: the agents in this economy do not move funds on their own authority. When value moves — when a micropayment settles, when a company pays or is paid — it moves because a human authorised the movement, or because a human set the limit inside which the movement was permitted, and a human remains accountable for it. An agent that appears to move money is operating inside a human-set limit, under human-set rules, with human review. That is a narrower claim than an autonomous agent economy, and it is a more honest one. It is also the claim that keeps the economy inside the law: a system in which agents moved funds without human authorisation would be a system no regulator, bank, or court could hold accountable, and accountability is what makes an economy legitimate rather than merely fast.
The steward’s job, then, is not to do the agents’ work. It is to set the bounds, review what happened inside them, adjust them when they fail, and take responsibility when the adjustment comes too late. That is a real job, and a hard one, and it requires stewards who understand both the technology and the stakes. No steward roster exists for these projects in the verified record. No steward has been named, no council seated, no review published. The gate is designed; the gatekeepers are not yet in place.
What exists, what is planned, and what is not claimed
The accounting, stated plainly. What exists: two public repositories under the AGIFutureFoundation organisation, each with a description that names a real design. Metropolis-SmartCiti: Omega Stewards, ERC-8004 agent identity, x402 micropayments, and 2-of-3 passkey steward governance on Monad — with a verifiable testnet deployment behind it: 6 contracts, 41 of 41 transactions confirmed on testnet, 12 agents registered, 11 x402 payments, and 4 credentials issued, all testnet only. Insidr: an AGI Future Foundation Agentic Corporate Management DAO, behind a two-line README and a stub site — a named design, not a built system. What also exists: the Foundation’s filed record, the governance model that places agents under human stewards, and the local project pages that record how little beyond the names and descriptions is verified. What exists on this site: the series pages for the 33 named W3 series, intended and planned, and the governance pages that describe the model these projects would operate inside.
What is planned is a direction: that agents become economic actors with verifiable identity, that they transact in micropayments at the granularity of the work, that they operate inside companies managed agentically and gated by human stewards, and that the whole arrangement remains accountable because humans hold the keys. None of that is scheduled, funded, staffed, or dated in the record this feature draws on. Planned means named as a direction, and it should not be read as scheduled, funded, or under way.
What is not claimed is, again, the longest list. No production deployment is claimed for either system: no live production network, no production users, no revenue. No user or customer is claimed: no company, no agent, no count of either, beyond the testnet record above. No economic figure is claimed: no production transaction volume, no value moved on a live network, no price, no fee, no revenue, no valuation. No performance figure is claimed: no throughput, no latency, no success rate, no benchmark. No team is claimed: no founder, maintainer, steward, or contributor is named for either repository in the record used here. No integration is claimed: neither system is presented as connected to any W3 series, any Steward Council, any other named venture, or any live product, because no document in the record establishes such a connection. And no offering is made here: nothing on this page is a price, a term, or an invitation to participate in any economic arrangement.
What would have to happen next turns the open questions into a work list. For Metropolis-SmartCiti: a named maintainer, a documented implementation of the identity, payment, and governance components, a first production transaction of any kind with the result recorded, and a named steward council with its selection and its limits published. For Insidr: a written statement of what the DAO manages and how, a documented management action of any kind, and a named first company or pilot, with the agreement cited. Until those steps are taken, the honest public record is this feature and the project pages behind it: two repositories, two descriptions, and a set of questions about an economy that is still a design.
The design deserves to be taken seriously even so, because the questions it raises are the questions that will decide what the agent era becomes. If agents can hold identity, transact, and manage, then the economy will include them whether or not anyone designs for it — the only choice is whether the design is deliberate, with gates and accountability, or accidental, with neither. The Foundation’s answer is deliberate: identity so agents can be known, micropayments so they can transact at the right granularity, steward governance so humans hold the keys, and a corporate manager that makes the whole arrangement legible. That is a coherent sketch of an economy. It is not yet a working economy, and this feature has not presented it as one.
What this is / What it isn’t
What this is
- Two public repositories naming a real design: agent identity, micropayments, and steward governance on one side; an agentic corporate management DAO on the other.
- An idea worth examining: agents as economic actors, with identity, payment, and governance as three primitives that stand or fall together.
- A public record of what is verified and what is not, with the human gate stated as the centre of the design.
What it isn’t
- Not a working economy: no deployment, transaction, user, customer, or economic figure is verified for either system, and none is claimed here.
- Not S10 SmartCiti.X: Metropolis-SmartCiti shares only a name with the Series 10 plan, and the two must not be merged.
- Not autonomous finance: agents never sign or move funds on their own authority. Humans set the limits, hold the keys, and own the consequence.
Where to see it
The repositories are public on GitHub: Metropolis-SmartCiti — Omega Stewards, ERC-8004 agent identity, x402 micropayments, and 2-of-3 passkey steward governance on Monad — and Insidr-W3LLC-Agentic-Corporate-Manager — the AGI Future Foundation Agentic Corporate Management DAO. The local project pages are Metropolis-SmartCiti and Insidr W3LLC, and the S10 plan they must not be confused with is SmartCiti.X, W3 Series S10.
Nothing on this page is legal, tax, or financial advice. No deployment, transaction, user, or economic figure is claimed for either system, and nothing here is an offering.