How Crew One actually works.
Four stations. One workflow to start. An operating layer over time.
You brief it, connect the tools one workflow needs, and delegate a real task. The Chief of Staff routes the work across eight specialist teams, holds the approval where it matters, and turns every call you make into precedent. Here is each of those four stations in full — the short version lives on the homepage overview.
You tell it how the company thinks. Once, not every time.
The setup is a conversation, not a form. You describe who you are, what the company is doing, and — the part that matters most — how you like decisions made. That last piece is what turns a generic assistant into your Chief of Staff: it needs your judgement encoded, not just your calendar.
Who you are, and what the company is trying to do this quarter.
How you like decisions made — and which ones you always want to see first.
The context you would otherwise repeat to every new hire on day one.
The brief is the seed of your institutional memory.
Everything the crew does from here reads against it. You are not configuring settings — you are giving the company a starting point it can build on, so the context you hold in your head starts living somewhere the whole crew can reach.
You connect what the first job needs. Nothing more.
Most tools ask for everything up front and hope you trust them later. Crew One works the other way. You pick the first real workflow — triaging the inbox, chasing invoices, keeping the CRM honest — and connect only the tools that workflow actually touches. Access expands when there is a reason, never speculatively.
The first workflow decides the first connection — nothing else.
Read access before write access; write access before anything irreversible.
Every scope is legible, and every scope can be revoked.
Minimal by default. Legible by design.
You can always see what the crew can reach and what it cannot, and you can pull a scope back at any point. The connections grow with your trust, not ahead of it — which is the same principle the approval loop runs on.
You hand over a task. The crew figures out who owns it.
This is the station most people picture when they imagine AI agents — and where Crew One diverges from a single do-everything bot. You give one plain-language request. Behind it, the Chief of Staff assesses how involved the work is, decides which specialist team should own it, and — for anything spanning more than one — drafts a plan and coordinates the hand-offs.
You brief the request in plain language.
The Chief of Staff assesses it and picks the team — or teams.
The crews do the work and draft the output.
Anything with real stakes comes back to you as an approval.
Simple requests go straight to one team. Complex ones get planned across several.
A runway question is Finance. A launch is GTM, Marketing and Product together. You do not route the work yourself — you describe the outcome, and the orchestrator decides the shape of the job. Routine steps run on their own; the moment real stakes appear, the work stops and waits for you.
Every decision becomes precedent. The next one starts from it.
A task that ends when it ends is just automation. Crew One is built so the work leaves something behind. Every approval, edit, rejection and exception is written down as a decision trace — not the answer, but how the answer was reached and why.
The decision, and who it was for.
The rationale — why this, not the alternative.
The outcome, and your edit if you made one.
Compounding.
The next similar task reads that trace and starts closer to right. The company stops re-deciding the same things and re-explaining the same context. Judgement accumulates instead of evaporating — and unlike a hire, this memory does not walk out when someone leaves. It is the difference between a tool you operate and an operating layer that gets sharper the longer you run it.
The crew proposes. You hold the gate.
Drafting the reply, the doc, the plan.
Reconciling records and tidying the CRM.
Gathering the numbers for a decision.
Anything reversible, low-stakes, in-house.
Payments and anything that moves money.
External messages sent in your name.
Irreversible or sensitive changes to systems of record.
Anything you flagged in your brief.
Human-in-the-loop is not a setting you switch on. It is how the system is built.
When the crew reaches something that matters, it stops and shows you the proposed action, the context behind it, and a way to approve, edit or decline. Approve and it proceeds. Edit and it learns your preference. Decline and that becomes signal for next time. The gate holds by default — nothing high-stakes slips through because a step ran unattended.
Start narrow. Widen as it earns it.
You do not roll out an operating system on day one. You give the crew one workflow, watch how it handles the judgement calls, and correct it where it is off. Every correction is a precedent it keeps. By the end of the week the first workflow runs with fewer interruptions — and you add the second.
That is the whole arc in miniature: brief it once, connect what the job needs, delegate the work, and let the decisions compound. The operating layer is not something you install — it is something that accretes, one approved decision at a time.
- How does Crew One work?
- Crew One works in four stations. You brief it once — who you are, what the company is doing, and how you like decisions made. You connect only the tools the first workflow needs. You delegate a real task, and the Chief of Staff assesses it, picks the specialist team (or teams), and coordinates the work. Then every decision compounds: approvals, edits and exceptions become precedent the next similar task starts from. Anything with real stakes waits for your approval throughout.
- What are the four stations — brief, connect, delegate, compound?
- Brief is the setup conversation where you encode who you are and how you make decisions, seeding your institutional memory. Connect is wiring up only the tools the first workflow touches, with access that expands when there is a reason. Delegate is handing over a plain-language task the Chief of Staff routes across eight specialist teams. Compound is the payoff: every approval, edit and rejection is written down as a decision trace, so judgement accumulates instead of evaporating.
- How does the Chief of Staff decide which team does the work?
- It assesses how involved the request is, then routes accordingly. Simple requests go straight to a single team — a runway question is Finance, inbox triage is the Executive Assistant. Complex requests get planned across several teams working together — a launch pulls in GTM, Marketing and Product. You describe the outcome in plain language; the orchestrator decides the shape of the job and coordinates the hand-offs. You never route the work by hand.
- Do I stay in control of what the crew does?
- Yes — Crew One is human-in-the-loop by design, not as an optional setting. Reversible, low-stakes, in-house work runs on its own. Anything that moves money, sends an external message in your name, or makes an irreversible change stops and waits for your sign-off, along with anything you flagged in your brief. The crew proposes and executes; you hold the gate, and nothing high-stakes happens without your approval.
- What does the first week with Crew One look like?
- You start narrow: give the crew one workflow, watch how it handles the judgement calls, and correct it where it is off. Every correction becomes precedent it keeps, so by the end of the week that first workflow runs with fewer interruptions — and you add the second. The operating layer is not installed on day one; it accretes, one approved decision at a time.
Brief it once.
Let the company compound from there.
Crew One is the operating system for AI-native, founder-led companies.