Reeve — written r·Eve — is a concept-stage AI operator for agencies, consultancies and studios. Its two faces, Eve and Adam, run the recurring work on a schedule and stop at an approval gate before anything they can’t undo. You keep the authority; they keep the account.
Two named operators, not a faceless bot — the first faces of your firm’s AI. Each runs a side of the recurring work, and each is bound by exactly the same rules.
Eve faces the market. She researches prospects, drafts the outreach, assembles the client reports, and turns what worked into the next brief. Anything client-facing, she prepares — and leaves for you to send.
Adam faces the books. He audits the accounts, keeps budgets on pace, reconciles the ledger, and watches every written limit. Anything that touches money, he proposes — and holds for your approval.
Two faces, one rulebook. Whatever Eve drafts or Adam reconciles runs inside the same governance — an autonomy ladder that decides what each may do alone, a gate that holds anything irreversible, and a ledger that records all of it. The faces do the work; the authority stays yours.
Every action that can’t be undone — a send, a budget move, a change of scope — waits here. Each card shows who proposed it, the exact rule it’s acting under, what will change, and the money at stake. Approve, edit, or reject in a click.
Reeve classifies each action on two axes — can it be undone, and does it need judgement — and that cell decides who may act. Autonomy rises a rung at a time; the top rung never moves.
The ceiling never moves. Moving or withdrawing funds, touching a client’s books, and any change to scope, price, or which clients you keep stay human — permanently, by design. A re-weight inside one platform, where no funds leave, is gated-approvable; money leaving an account is not.
Each proposal, decision, and action lands in an append-only ledger — what ran, which face, the rule it followed, who approved it, and the outcome. Replay any run step by step, or export the record for a client, an insurer, or a buyer.
Before every run, Eve and Adam read the client’s charter: goals, hard limits, what to escalate, and the read-only systems they may touch. It’s versioned, so every edit is tracked — and every gate card and ledger line links back to the exact clause it acted under. The rules and the record are one object.
Pulls the CRM and new inbound, scores each lead against the client’s fit rules, and builds a ranked prospect sheet. No message is sent.
You decide: which outreach drafts get queued.Gathers the month’s numbers across connected systems, writes a formatted report plus a next-month plan, and leaves it in draft.
You decide: the send, and any edits to the narrative.Compares spend against plan and the charter’s caps, and drafts a reallocation that stays inside the written limit.
You decide: any move past the limit — the rest runs within the cap you set.Reads the week’s transactions, categorises them against the chart-of-accounts rules, and flags every unusual entry.
You decide: each flagged entry — anything moving money stays fully human.Anyone can wire an agent into a few tools. What they can’t copy is your firm’s governance record — the per-client charter that compounds with every edit, and the append-only ledger of every decision Eve and Adam ever proposed and you ever approved. That record is the switching cost, and it’s the thing a client, an insurer, or an acquirer trusts. The model is a commodity; the accountable history is the asset.
Both — but the line is explicit and written down. Reversible, routine work (research, reconciliation worksheets, report drafts) they do on their own. Anything irreversible or judgement-heavy they propose and hold at the gate, where you approve, edit, or reject. What each may do alone, per client, is set on the autonomy ladder — never assumed.
Send anything to a client, move or withdraw real money, touch a client’s books, or change scope, price, or which clients you keep. Those sit at the top rung as “always human” and can’t be promoted. A re-weight inside one ad platform, where no funds leave, can be gated-approvable; money actually leaving an account cannot.
Reading a client’s systems is read-only by design. Any action Reeve executes — an auto-action under a written limit, or anything you approve at the gate — runs through a separate, scoped, logged write path with its own caps. Reading and acting are different connections, and every action lands in the ledger.
Automations just run; Reeve is governed. The product is the control plane around the agent — a ladder that decides what each face may do, a gate that holds every risky action for a human, and a ledger that records every proposal, decision, and approval so any run is replayable. That governance record, not the model, is the point — it’s what you, your clients, and an eventual auditor can trust.
It’s an early-stage idea in development, not a shipping product. There are no customers, revenue, or results to claim — every figure on this page is an illustrative example or a limit an operator would set, never an achieved metric. What’s real is the thesis: the governance record, not the model, is the defensible part.
Reeve is an early concept from The Adjacent Possible. We’re building the autonomy ladder, the gate, and the compounding ledger — and we’d like to talk to agencies, consultancies, studios, and bookkeeping shops who already live by their own rules.
Or write to [email protected]