Features · AGI Future Foundation
The Cohort Machine
Most incubators are rooms. This one is a machine, or it would be if it existed. The plan is simple enough to state in a sentence: a new company enters as a proposal, passes through seven stages and six gates, gets a series number and a Steward Council, and comes out the other side as a legal entity with segregated books, a named human signer, and a funding gate that knows which kind of money it is allowed to take. The machine has two parts that are easy to confuse: Incubator DAO, Series S21, the planned DAO for incubating early ventures, and the Delaware cohort series, Cohort 1, Series 1 to 3, the legal slots those ventures would be formed into. This feature is about both parts, and about the gap between them, because the gap is where every new company in the family will either be formed properly or not at all.
The machine that makes companies
Start with what the machine is for, because the rest of the design only makes sense against it. The family plans to build many companies with few humans, using agents for the daily work and humans for the decisions. That only works if forming a company is a repeatable process rather than a bespoke negotiation every time. The cohort machine is that process: a standard intake, a standard set of gates, a standard charter, a standard council, and a standard funding rule, applied to every new project the same way. A project that enters the machine knows what will be asked of it, in what order, and by whom. A project that cannot answer does not get a series number, and a series number is not approval, funding, or a guarantee of mission fit.
The machine has two named parts, and they are not the same thing. Incubator DAO is W3 Series S21, named in the April 2026 internal W3 series list, and its planned focus, as the name implies, is helping early ventures and builders get started, with space, training, mentorship, and time, governed as a DAO. The name is the most the files say about it. No cohort, resident, graduate, admissions criteria, or programme date is verified for this series, and it has no anchored metaverse fabric: no coordinates, no fabric file, no hosted assets, and no RP1 attach are verified for it. Zero worlds are live on RP1, for the whole W3 programme, not a gap peculiar to Incubator DAO. The Delaware cohort series is the other part: an intended Delaware series LLC, under 6 Del. C. section 18-215, into which all new projects go, as C1S1, C1S2, and C1S3 for Cohort 1, with the naming to confirm. The incubator is the programme. The cohort series is the legal form. Confusing them is the fastest way to promise a cohort that has no company behind it, or a company that has no programme around it.
The distinction matters twice over, because the family already has two neighbouring ideas that use the incubator word. The SmartCiti.X plans describe incubator housing, places for founders, builders, trainees, and instructors connected to a facility to live for a defined residency, under a residency agreement and a board-set fee, with zoning and safety in place before anyone moves in. The BuilderSpace tour is a separate programme of hackathon and onboarding events. Incubator DAO, if it ever becomes active, would need a written agreement saying how it relates to each of those: whether it runs a programme inside SmartCiti.X housing, whether it is separate from BuilderSpace, or whether it is something else entirely. Until such an agreement exists, the three should be treated as separate ideas that happen to use the word incubator, not as parts of one programme. The same discipline applies to the studio itself. The studio model is a separate design, an AGI Corp function, and Incubator DAO is not the studio. The incubator hosts ventures. The studio funds and manages them. Keeping those two ideas apart, in writing, is part of the work this feature does.
Intake: seven stages, one door
Every new project enters through one door, and the door has seven stages, none of which may be skipped. Intake files the proposal and opens a register row, and the register row is a tracking row, not a filing. A row in the register does not make a series exist. A series exists only when its master LLC’s operating agreement designates it, its books are segregated, and, where the statute requires it, its designation is filed or recorded. Counsel must confirm the Wyoming and Delaware requirements for each master, because the two states do not automatically recognise each other’s liability shields, and a project assigned to the wrong master, or to both masters, is a project with two legal homes and no clean one. One project is one series under one master, minuted at intake, and the register must not be pre-filled with the 35, 33, or 34 counts that disagree across the family’s materials. The internal list names 33 W3 sub-series, Series 1 through 33, and this feature follows that list.
The proposal itself must name the project, its description, its founder or lead, its intended world or product, and its funding source, with one primary source named: donation, grant, investment, or revenue. That last item is the one that decides which gate the project will face later, and it is asked first for that reason. Mission-fit review comes next, and it belongs to the Foundation: does the project advance safe, equitable, and socially beneficial artificial intelligence under the Foundation’s mission and standards? The answer is approved, declined, or approved with conditions, minuted with reasons, and a declined project does not get a series number, without exception. Legal assignment follows, putting the project under one master LLC and only one: existing series under the intended Wyoming W3 LLC, and new projects into the intended Delaware cohort series company, as C1S1, C1S2, and C1S3. Series designation amends or appends the operating agreement to designate the series, with its purpose, assets, and liabilities segregated on paper. The DAO charter is written for that series, following the charter outline in the register paper, and the Steward Council is named. Human decision is approve, amend, decline, pivot, or kill. Execution and record close the loop with a dated receipt, naming the agent that drafted the step, the human that adopted it, and the date.
The intake checklist that sits behind those stages is longer than the stages themselves, and it is worth reading as a list, because it is the machine’s real specification. The DAO charter must name the agent roles, the proposal process, the voting rules, and the express limit that DAO and agent proposals are advisory and operational only, and that only the Steward Council enacts, amends, or moves funds. The Steward Council must be three consenting humans, deciding by two-of-three signed votes, with keys and signing credentials actually held; until that is true, the series is Proposed, not Operating. A legal manager or signer must be named, a human, or AGI Corp under a written management agreement, authorised to sign contracts and bank documents for that series. Segregated bank accounts and books must be opened for that series, only after designation, with no series funds in the master LLC’s general account, AGI Corp’s account, or the Foundation’s account. Intellectual property must be assigned on day one, by signed assignment or licence, stating who owns the project’s IP: the series, AGI Corp, the Foundation, or a contributor, and not by assertion. And the funding gate must be checked before any money moves: donation or grant through the Foundation gate, investment or SAFE through the AGI Corp gate, never the other way round, and never from an unformed series.
The six gates, in practice
Once a project has a series number, it moves through six gates, and the gates are where the machine earns its keep, because they are the only part of the design that can say no. A gate is passed only when the human approver signs and the evidence is filed. An agent’s readiness flag is not a gate pass. A project that fails a gate is paused or killed, not carried forward on agent momentum, and the model treats killing a project as a studio success, not a failure to hide. That sentence is easy to write and hard to live by, which is why it is written into the stage-gates paper as a rule rather than left as a sentiment.
Gate 0 is Idea, agent-managed and human-selected. An agent brief is filed, covering the problem, the proposed solution, a mission alignment draft, a market sketch, a technical sketch, and the risks. The Foundation runs a mission screen, recorded as pass, pass with conditions, or decline. Nothing is a company yet, nothing is funded, and a declined idea ends here. Gate 1 is Validate, agent-assisted: customer and problem research grounded in human interviews, not agent assertions alone, and a financial model draft filed and labelled as a draft, not adopted, until a human adopts it. Then comes series designation, with the Steward Council named, three consenting humans holding signing keys. A series with no Steward Council is capped at Gate 1, and it cannot pass to Build as an operating company. Gate 2 is Build, agent-managed and human-reviewed: agents build the prototype or demo, test it, and document it, while the human checks are the intellectual property chain, a security and privacy review, and the contract rule that all contracts are signed by a human with authority, none by an agent.
Gate 3 is Launch, human-approved, and the most heavily human gate. Public claims are checked, the claims list is filed, and pricing is set and approved, or the launch is stated expressly as free or demo-only. No invented partners, users, revenue, venues, or outcomes, and no invented prices. Gate 4 is Operate and Scale, under quarterly Steward Council review, with a quarterly report filed covering funds in and out, key performance indicators as actuals rather than projections, risks, mission outcomes, and related-party transactions, and the budget approved on two-of-three signed votes. Gate 5 is Exit or Wind-down, a decision rather than a drift: the board and Council decision minuted with reasons, intellectual property, data, customer obligations, and funds resolved in writing, the series books closed, the accounts reconciled, and the register updated to Closed with a date and a reason. No further fundraising, sales, or public claims after closure, except a factual closure notice approved by a human.
The portfolio dashboard is the standing record the gates produce: one row per series, updated monthly, actuals only. Never a projection in a revenue or funds-in column. Never an unsigned partner, customer, or investor listed. The Fellowship is not a portfolio row. The dashboard is also where the machine’s honesty is most visible, because an empty dashboard is an honest dashboard. Today the dashboard is a template with blank rows for C1S1, C1S2, and C1S3, and no project has passed any stage gate under the model, because no project has entered it. A template is not a track record, and this feature does not present it as one.
The kill rule deserves a paragraph of its own, because it is the gate the machine will be judged by. Most incubators measure themselves by how many companies they launch, and a machine that measures itself that way will launch companies that should have been killed, because the people running it are rewarded for throughput. The stage-gates paper inverts the incentive: a project that fails a gate is paused or killed, and killing it is recorded as a studio success. The reasoning is practical rather than sentimental. A killed project costs the family the time already spent on it and nothing more. A project carried forward on agent momentum, past a gate it failed, costs the family a series designation, a Council’s time, a segregated account, a funding round, and a public record that will have to be explained later. The gate that kills early is the cheapest gate in the machine, and the Council that uses it is doing its job. That is also why the evidence rule matters: a gate is passed only when the human approver signs and the evidence is filed, so that a later reader can see, from the record, which gate a project passed, when, on what evidence, and on whose signature. A machine that cannot show its work cannot be trusted with other people’s money, and the cohort machine is built to show its work from the first proposal to the last receipt.
Who does what: eight roles, eight functions, one Council
The machine runs on people and agents in a fixed split, and the split is the same one that governs the whole studio: agents draft, research, build, coordinate, monitor, and report, while humans review, decide, approve, sign, and are accountable. The eight agent roles describe functions, not people and not owners: research synthesis, task routing, proposal drafting, coordination, progress tracking, monitoring, collaborator scouting, and grant drafting. Each series’ DAO charter must name which agent holds which role, who the human coordinator or manager is, and what each agent may and may not do, including its spending limit, zero dollars without Council approval unless the Council sets a written threshold. When a role is unfilled, a coordinator covers it, and the coordinator is not a ninth role.
The eight studio functions are the work those roles do, repeated for every company the machine touches: ideation and sourcing, validation, build, launch, operate and scale, portfolio management, fundraising support, and mission and compliance. Three of them deserve extra attention, because each one is where the boundary between drafting and deciding is thinnest. In validation, the model requires human interviews, not agent assertions alone, and a financial model stays labelled as a draft, not adopted, until a human adopts it. In launch, the claims check is absolute: no invented partners, users, revenue, venues, or outcomes, and pricing must be set and approved, or the launch is stated expressly as free or demo-only. In fundraising support, agents draft decks, data rooms, investor lists, grant drafts, and diligence packs, and do not solicit investment. The hard limits bind every role equally. Agents never sign a legal document, move funds, hire or fire, offer securities, or enact doctrine or charter changes. Agents prepare. Humans execute.
The two agent roles that most often surprise newcomers are the coordinator and the grant drafter, and both are worth a sentence, because they show how the machine treats work that looks like authority but is not. Coordination keeps people, tasks, and timing lined up inside a series, but coordination is not authority, and the coordinator agent cannot approve a budget, hire or fire, or commit the series to a deadline, a partner, or a customer. Grant drafting writes grant drafts for the Foundation’s charitable programmes and, where the gates allow, for a series seeking grant funding, and a draft is written for a human to review, correct, and submit, never going out under an agent’s name or authority. The pattern is the same in both cases: the agent does the work that can be checked, and the human does the act that cannot be undone. A series that forgets the difference has not automated its governance. It has abandoned it.
Above the agents sits the human gate, and it has two layers. The Steward Council, three consenting humans deciding by two-of-three signed votes, is the only authority that can move funds or enact a charter change for a series, and it examines each proposal with any DAO vote as an advisory signal only. The Foundation board sits above that as the mission gate, holding mission-fit approval, mission-standards enforcement, and grant decisions, and it can withdraw mission approval, after which a project loses Foundation support and branding even if AGI Corp wants to continue it commercially. Admissions, in particular, are a human decision: who is let in, on what criteria, and with what commitments on both sides, and no agent should be making that call alone. No admissions criteria, cohort, resident, or graduate is claimed for Incubator DAO anywhere in this feature, because none exists in the files reviewed.
The funding gate is the last part of the split, and confusing it is the fastest way to break the law in this structure. The donation and grant gate runs through the Foundation: donations and grants may fund a studio project only as a grant serving a charitable purpose, with reporting, not as studio equity, not as a SAFE, and not as a subsidy for a commercial series’ losses, unless counsel advises a specific lawful structure in writing. No SAFE, equity, profit share, or return may be issued by the filed nonprofit Foundation or by the Fellowship church, and no tax-deductible donation may be claimed. The investment gate runs through AGI Corp, and a project still Proposed cannot take investment. The suggested first raises for the Delaware cohort series are $100,000 to $500,000 per series, if Gate 1 is passed, and the same range is suggested for other W3 series, per series rather than pooled. Those figures are suggested and illustrative only, never a price, and no money has been raised under them.
Cohort 1, Series 1 to 3, and the naming collision
Cohort 1 is where the machine’s paperwork is thinnest, and the thinness is worth naming precisely. The naming convention is confirmed by Mike Rice on 2026-11-03: C, cohort, S, series, so that C1S1 is Cohort 1, Series 1, C1S2 is Cohort 1, Series 2, and C1S3 is Cohort 1, Series 3. The note that recorded it also carries a clarification that has not been resolved: the message listed C1S1, C1S2, and C1S3 with a parenthetical reading Cohort 1, Series 2, and the register paper reads the set as Cohort 1, Series 1 to 3, with a flag to confirm whether the first cohort is Cohort 1 with series 1, 2, and 3, or whether there is a different cohort numbering. Do not assign C2 numbers until Cohort 1 is defined. The legal name for contracts and bank accounts is the master LLC’s legal name plus the series designation, never C1S1 alone, and never the project nickname alone. The project nickname or world name, if any, is a marketing label, recorded in the register for reference only, and it is not the legal name. The Delaware master LLC’s name has not been chosen, and its registered agent is blank, and no operating agreement is on file. The Wyoming W3 master is in the same position: intended, not verified, with no operating agreement on file and a Wyoming agent still to be named. Do not use the same master name in both states, and do not assign a project to both masters.
There is also a naming collision that the papers flag and this feature records, because a reader who finds it on their own will assume the worst. The AGI Accelerator, at agi-lease.kylon.app, is live with a 60-day, $5,000 residency, and its pages name a Cohort 1 in September 2026 and a Cohort 2 in November 2026, dates that have passed, with its status unverified, and its operator claim needing counsel review. Those cohort numbers collide with C1S1, C1S2, and C1S3, and the two programmes are not the same thing. The Accelerator is a separate programme with its own operator and status questions, and its residency terms must not be copied into the cohort machine, or into SmartCiti.X housing, unless a board expressly adopts them. Until the collision is resolved in writing, the cohort machine’s Cohort 1 and the Accelerator’s Cohort 1 must be treated as two different labels that happen to share a word, not as one programme.
Two exceptions complete the picture, and both are worth stating because readers will look for them. The first is SmartCiti.X, which does not go through the machine at all. The corporation plan superseded the Series 10 raise, and SmartCiti.X, Inc. would be a new, separate corporation, not a Delaware cohort series and not a Wyoming series, with its own charter, bylaws, employer identification number, bank account, and cap table. A new project goes into a series, unless counsel advises otherwise for a specific project, and SmartCiti.X is the specific project where counsel’s otherwise has already been written into the plan. The second exception is the Fellowship, which never goes through the machine, because a church is not a series, not a cohort, and not a portfolio company. The machine forms companies. It does not form churches.
What exists today, what is planned, and the open questions
What exists today is a name, a number, a register template, and a set of gate checklists. Incubator DAO is a name and a series number, S21, in the April 2026 internal W3 series list, with a planned focus implied by the name and nothing more. The register is a template with blank rows for the Wyoming W3 series and for the Delaware cohorts C1S1, C1S2, and C1S3, and the stage-gates paper is a template with blank checklists per project. The studio model and the proposal flow are written designs, not operating systems. The filed Foundation is real, a California Nonprofit Public Benefit Corporation, Entity No. B20260196664, and the filed Fellowship is real, a separate church. Nothing else in the machine is filed. No master LLC has been verified as filed in Wyoming or Delaware. No series has been designated. No Steward Council has been seated for any series. No DAO charter has been signed. No segregated series bank account has been opened under the machine, and no money has moved under it. No project has entered it, and no project has passed any of its gates.
What is planned is the order of work, and the order is the discipline. Form the masters first, with counsel choosing the Delaware master’s name and the Wyoming master’s agent, and only then designate the first series, seat the first Council, and open the first segregated account. Confirm the Cohort 1 numbering in writing, so that C1S1, C1S2, and C1S3 mean the same thing to everyone who uses them. Resolve the Accelerator naming collision in writing, so that no cohort is confused with another programme’s cohort. Write the Incubator DAO mission, admissions criteria, cohort structure, and graduation or exit rules, decided by humans and written down. Write the agreements with SmartCiti.X housing and with BuilderSpace, if any relationship is intended, or record in writing that there is none. Decide, in writing, whether a cohort needs a shared virtual space at all, with published coordinates and a verifiable RP1 attach, or whether the work happens in person and in documents and needs no world. Zero worlds are live on RP1, and an invented coordinate would be worse than an empty one.
The open questions are the ones the machine cannot answer for itself. Counsel must confirm the Wyoming and Delaware series requirements for each master, and whether the two states’ liability shields recognise each other. Counsel must confirm that a DAO charter inside a series is governance and not incorporation, and that a token vote, a chat poll, or an agent proposal is not a corporate act. The Foundation board must be ready to run the mission screen, and to decline a project that fails it, with reasons minuted. The first proposal must move through the seven stages and the six gates, so that the record at stage seven can prove the system works. Until then the system is untested, and presenting it as tested would be the error this whole series of features is written to avoid.
What this is / What it isn’t
What this is
- An intended machine for forming new companies: Incubator DAO, Series S21, as the planned programme, and the Delaware cohort series, C1S1 to C1S3, as the legal slots, with seven intake stages, six stage gates, and a human Steward Council.
- A discipline about who decides: agents draft and operate, humans admit, designate, sign, and move funds, and a DAO vote is advisory only.
- A funding rule with two gates: donation and grant through the Foundation, investment and SAFE through AGI Corp, never the other way round, never from an unformed series.
What it isn’t
- Not an operating programme: no cohort, resident, graduate, admissions criteria, or programme date is verified for Incubator DAO, and no project has entered the machine or passed any gate.
- Not a formed entity: the Delaware cohort master is intended and unnamed, no series has been designated, and C1S1 is a reference label, never a legal name on its own.
- Not the Accelerator and not the studio: the AGI Accelerator is a separate programme with its own cohort numbering, and Incubator DAO is not the venture studio.
Where to see it
The series page for the incubator is Incubator DAO, Series S21, with the deep article at Incubator DAO, article 29. The machine itself is documented on the Proposal Flow, Human Gate, and Studio Model pages, with the Council on the Steward Council page. The landing page for the studio’s working harness is public on GitHub: trueforge-site — a landing stub, not a working harness, and there is no main trueforge code repo in the organisation.
Nothing on this page is legal, tax, or financial advice. The cohort machine, the stage gates, the register, and the proposal flow are templates and intended designs. The suggested first raises for the Delaware cohort series are planning figures only.