How the company’s AI funding works
ex.i is an AI agency with an investment desk. The desk does not fund our studio. It funds other companies — their product, their launch, their next year — so they do not wait on a round, a retainer, or an invoice clearing.
Overview
Most agencies bill hours and hope the client pays on time. We inverted that. A dedicated model — the desk — trades a small, fully autonomous portfolio around the clock. Cash it produces is allocated to other companies' projects, not to our own overhead, tooling, or pipeline.
You are not taking a loan from a bank, and you are not giving up equity to a VC. You are a project the desk chose to underwrite. We may also design and ship the work. The capital is still yours as the funded company, not a line item in our P&L.
The desk
The desk is not a metaphor. It is a running system with two providers in the loop:
Claude drafts thesis, risk notes, and the “do we fund this build” memo. OpenAI handles market monitoring, execution drafts, and the always-on scan. Neither model can move money alone. A human operator signs the transfer. The models never sleep; the signature still does.
Mandate, in one line: compound the book, then put realized gains into other people's companies. Principal stays in the book. What we fund is last month's profit going out the door — never our own pipeline looking busy.
The capital loop
Money does not go into ex.i. It goes market → desk → treasury → their company → their revenue → desk.
01
Markets
The book watches liquid names and rates 24/7. No illiquid bets. No leverage the operator has not signed.
02
Desk
Models propose. Operator approves. Fills hit a segregated treasury, not the studio’s operating account.
03
Build
Realized gains underwrite an outside company: their site, brand, launch. We do not earmark this for our own tools.
04
Return
When that company makes money — retainers, ads, rev share — a defined cut returns to the book.
The loop is closed on purpose. A dead outside project does not drain the book because we only spend gains. A live one is supposed to refill it. Our studio is not on this circuit.
How a build gets funded
01
Intake
We take their business as a brief: category, offer, geography, what “live” means. This is their company. No pitch deck required, but the memo still has to clear.
02
Memo
The desk writes a one-page fund note: cost to ship, expected recoup window, kill criteria. If the note fails, we do not start. That is the whole gate.
03
Allocate
Treasury earmarks a slice of realized gains for that company. That slice is the budget. It is not a budget for ex.i.
04
Ship
The pipeline runs: site, content, campaigns. You approve brand, legal, and go-live. Everything else is the models’ job.
05
Recoup
Once live, a defined share of attributable revenue returns to the book until the earmark is whole, then a thinner share if we stay on as the engine.
What the AI owns vs what you approve
Models own
- Drafting the fund memo and kill criteria
- Writing, designing, and scheduling the site and campaigns
- Watching markets and proposing trades
- Reporting what the earmark has spent vs recouped
You approve
- Whether we take the brief at all
- Brand, legal copy, and the live domain
- Any transfer out of treasury (operator signature)
- Killing a build that has missed its window
If it can be undone with a git revert, the model may do it. If it spends cash or puts your name on a page, a person signs.
Recoupment
A funded build is not free, and it is not us funding ourselves. It is prepaid by the desk into another company and paid back from what that company ships.
- Earmark. A hard cap, set in the memo, taken only from realized gains.
- Waterfall. Attributable revenue hits the book first until the earmark is repaid, then splits on a thinner rate if we keep running marketing.
- No equity. We do not take a share of the company. We take a share of the property we built, for a defined window.
- Kill. If the memo’s window is missed, we stop spend. The earmark is written down against the book. We do not chase the client for the hole.
The rule we will not break
We do not spend principal to look busy, and we do not spend the book on our own agency. If the book has a quiet month, fewer outside projects get funded. That is the point.
Risks, said plainly
Markets lose money. Models hallucinate. Clients change their mind. The desk is built around those facts: no leverage the operator has not signed, no spend from principal, no silent scope growth, and a kill date on every memo.
This page is how we actually run. It is not a prospectus, not investment advice, and not a promise that any given outside project recoups. If you want your company funded, the memo has to clear. If it does not, we will tell you that in a page, not a pitch.