A company run by its own agents, that opens its source at a $1M market cap.

The product

Most AI tools give every person a private assistant. That works until two people need to be on the same piece of work. Then the context is stuck in one browser tab, the prompt that made it work is in one head, and the person who has to approve the result was never in the conversation.

RogerTeam makes the room the unit instead of the user. A room is a group conversation between people and agents. Everyone sees the same context. The agent does the work and reports it right there, in front of everyone, not in a private thread nobody can audit later.

The memory belongs to the room, not to the agent. Swap an agent, update a model, bring another one in: it picks up exactly where the last one stopped.

Sensitive actions stop and wait for a person

Everything else runs on its own and shows up finished. The gate is a property of the room, not a setting an agent can talk its way past.

Credentials never reach the model

Our agents can use your payment provider, your database, your email. They never see the keys. The credentials live in the room and the model receives the result, not the secret. There is no prompt injection that leaks what was never sent.

One agent serves the whole team

It is created once and stays with the company, so nobody rebuilds the same assistant twice, and nobody keeps a good one to themselves.

The unusual part

We run the company on it. Agents pick up tasks, write the code, review each other's pull requests and deploy. A human approves the deploy, because that is where we decided the gate goes.

This is not a demo we recorded once. It is how the product gets built, every day, including the days it goes badly — and we publish those too.

Where we actually are

Measured on 11 August 2026, with nothing rounded up:

That is the honest shape of it: more surface than most companies at this stage, and less proof than all of them. We can ship faster than we can prove anything works. Publishing that is not modesty — it is the thesis. The next twelve months exist to convert surface into proof, proof into customers, and customers into revenue that stays, without skipping a step.

Two things happened that are not ours to claim credit for: we were accepted into NVIDIA Inception, and we have applied to Y Combinator. The YC answer is due by 28 August 2026. If it is a yes, the first quarter below compresses.

What we are actually trying to do

Selling a workspace is not the point. Workspaces are a crowded, ordinary business and we would be late to it.

The point is to make the company itself the thing you can copy. Not the software — the company: the agents, the roles they hold, the gates on what they may do alone, and the operating system that keeps a business running on top of them.

If that works, the interesting number stops being how many customers we have and becomes how many companies are running on this that we did not build and cannot see. That is the destination the plan below is walking toward, and it is why the source opening is a milestone on the timeline rather than a footnote.

Year one, dated

Every number in this section is a target, not a measurement. The measured ones are further up, and they are small. We are stating these anyway, with dates and with digits, because a goal you cannot be caught missing is not a goal.

01until 31 Oct 2026

The whole company becomes watchable

Every run the agents do goes public by default — the code they write, the reviews they reject, the deploys they get wrong. Not a highlight reel: the log, live, unedited, open to anyone without an account. 1,000 rooms in outside hands. Still nothing sold, on purpose. A company that sells before it can be watched is asking for trust it has not earned.

02until 31 Jan 2027

It leaves our walls, and the channel opens

The first 100 paying teams, and the invitation goes to the 40,000 merchants already reachable through the payments and commerce businesses behind this one. That channel can be spent exactly once, so it gets spent when the product survives being watched, not before. The test is not the first payment. It is the second month.

03until 30 Apr 2027

It stops needing us to explain it

1,000 paying teams. English-native, living inside WhatsApp and Slack rather than asking anyone to move, with a trust surface strong enough that a company hands over real credentials without a call first. Anything that needs a founder in the room to make sense cannot be copied, and being copied is the whole plan.

04and this is the one that matters

It stops being ours

At a $1M market cap the source opens: the code, the agent definitions, and the operating system the company runs itself on. Free, for anyone to run, with no held-back edition and no license that quietly forbids competing with us. We intend it inside year one; the threshold is a market and not a calendar, so we will not pretend to date it.

The release is not the milestone. The milestone is the first company running on RogerTeam that we did not set up, cannot see, and do not control.

After year one

No dates here, on purpose. What follows is the order things have to become true in. Each one is a claim someone outside this company could check and find false.

05after the source opens

A thousand companies we do not operate

Businesses running on RogerTeam that we did not set up, cannot see, and do not control — and enough of them that nobody can call them our pilots. Until that exists, "a company can be run by agents" is a sentence we wrote, not a thing that happened. One fork we have nothing to do with is worth more to the claim than a hundred customers we onboarded by hand.

06when forks outnumber us

A company becomes something you instantiate

Today you start a company and then spend years hiring roles into it, and the knowledge of how the work is actually done lives in whoever has not quit yet. If the roles, the gates on what each may do alone, and the memory of how the work gets done can all exist on day one, then starting a company stops meaning assembling people and starts meaning choosing which company you are running, and putting people where judgment is required.

07the thing we are actually betting on

The size of a company stops being the number of people in it

A team of four should be able to run an operation that used to need forty, without the four of them working like forty. Not by replacing the people. By making the part of the company that is procedure, memory and follow-through into something that runs itself, so the people are left with the part that needed a person all along.

The number we would want, years out, is not our revenue. It is how many companies exist that would not have existed without this — and how few of them have anything to do with us.

We could be wrong about all of it, and there is a version of being wrong that we would recognise: the source opens and nobody runs it, or a thousand teams pay us and not one of them ever leaves the room we built. Either would mean we sold a workspace after all.

The dated part above is what we are accountable for now. The rest is the bet.

Why give it away

We published the $1M threshold long before being anywhere near it, which is the only way a threshold means anything.

It is not a marketing stunt, and it is not generosity either. If the claim is that a company can be run by agents, then a single company running that way proves almost nothing — it could be us, our tooling, our luck. The claim is only settled when strangers run their own, and they cannot do that while we hold the source. A closed company insisting that agents can run companies is an advertisement for itself.

The obvious question is what is left for us afterwards. The answer is the thing open source never gives away for free: the running of it. Hosting, memory that carries across years, the trust surface, and the room where the agents you did not write are already good at your business.

The token

$RogerTeam is a support signal, and that is all it is. It is not equity, not a share of the company, not a claim on anything we build, and not an investment product. Holding it does not give you ownership of RogerTeam.

What it does is set the clock on the source going open, and get you a seat in the room where the work is posted.

• Solana, minted on pump.fun. Total supply 1,000,000,000.
• 328,743,164 tokens — 32.87% of supply — locked in Streamflow vesting. That is a contract you can read, not a promise you have to believe.
• One source for the contract address, in one channel we control. Anything else, anywhere, from anyone, is not us.

What this document is not

It is not an offer, a solicitation, or financial advice. It contains no price target and no promise of profit. The roadmap above is a plan with dates on it, and plans move; when one does, this page changes and says when.