About three months ago I started noticing the same pattern in every CEO engagement. They had all set up the same Claude Project. They called it different things — Strategic Advisor, Chief of Staff, Operating Brain, Co-CEO, COS — but the architecture was identical. The same instructions, the same connected tools, the same workflows. They had each invented it independently. They just had not given it a name.
I have started calling it the AI Chief of Staff, because that is the closest job description for what it actually does. And I now tell every new founder-led or executive-led client to build one in their first week with Claude. It is the single highest-leverage thing you can build with a Claude Team or Pro account, and you can have a working version of it by the end of an afternoon.
This post is the full walkthrough — what it does, why Claude is the right platform for it, and exactly how to set one up.
What an AI Chief of Staff actually does
A Chief of Staff in the conventional org-chart sense protects the executive from low-leverage work, holds priorities and recent decisions in their head, and makes sure the right things get attention. They are one of the highest-leverage hires a growing company can make and one of the hardest to staff well — partly because the role does not really show up on a resume.
What I noticed in the engagements is that Claude can do most of what a human Chief of Staff does in the first six months of the role, immediately, for any executive willing to put thirty minutes a week into maintaining the system. The Claude version does five specific things:
1. It holds your context. Your top three priorities for the quarter. Your last ten major decisions and the reasoning behind them. Your current frustrations. Your operating preferences. The people you work with most often and what each one is responsible for. The pattern in how you actually run your business, not the one that lives in your strategic plan.
2. It pressure-tests your decisions. You walk in with a decision you are about to make and ask it to play devil's advocate. Not to be polite about it. To use what it knows about your priorities and recent decisions to argue against you. The good ones argue convincingly. The great ones change your mind once a week.
3. It synthesizes across sources. You point it at last week's leadership Slack channel, the three board memos you wrote this quarter, the budget your CFO just sent over, and ask it where the contradictions are. The answer is almost always a small list of things you should have caught and did not.
4. It drafts in your voice. Board memos, customer emails after a hard call, internal communications when you need to land a message exactly right. Not because it writes better than you. Because it writes faster than you with enough of your character that you can edit the draft instead of starting from a blank page.
5. It is a thinking partner you can interrupt at midnight. This sounds soft until you have used one for two months. Then it becomes the most valuable thing you have built. The leverage is not what the tool produces. The leverage is that you can have a useful conversation about something hard, at any hour, without the social cost of waking up your COO or your spouse.
If you have ever been a founder or executive late at night with a hard decision and nobody to sanity-check it with, you understand why this matters.
Why Claude specifically (versus a regular chat tool)
You can build a half-version of this in any AI chat tool. You can have a useful conversation with ChatGPT or Copilot or Gemini once. What you cannot do in those tools is build a Project — a persistent context that the AI remembers across every conversation, that you can update over time, that connects to your actual data sources.
Claude Projects are the architecture that makes this work. A Project holds:
- A document with your Project Instructions — who you are, your priorities, your operating frame, your preferences
- Files you upload — your strategic plan, board memos, OKRs, any reference material that should always be in context
- A custom knowledge base — anything you want Claude to always know about you before responding
Every conversation inside the Project starts with that context loaded. You do not have to re-explain anything. The third conversation is as informed as the thirtieth, because the context is shared.
Then on top of Projects, Claude has two other features that make the Chief of Staff system real: MCP connectors and Skills. MCP connectors let Claude pull live data from Google Drive, Slack, Linear, Notion, Asana, your calendar, your CRM. Skills are forward-slash commands you can build for repeated workflows. Together with Projects, they are how you turn a chat tool into an actual teammate.
I cover the comparison in more depth in Claude vs Copilot for Teams in 2026 — the short version is that Copilot is a feature inside Word and Outlook, Claude is a teammate you go talk to. For a Chief of Staff system, you need the teammate.
The six-step setup
This is the actual setup process. I have walked dozens of executives through it. It takes about ninety minutes if you do it carefully, less if you cut corners. The version below assumes Claude Pro (individual) or Claude Team. Claude Team is recommended if you want anyone else on your team to have access to the system — Pro is fine if it is just for you.
Step 1 — Create the Project
In Claude, click New Project. Name it something specific to you, not "Chief of Staff" (the name does not matter to Claude, but specificity helps you mentally separate it from other Projects). Mine is named after my company.
You will see the Project Instructions box. This is the most important field in the entire system. We will fill it next.
Step 2 — Write the Project Instructions
The Project Instructions are the system prompt for every conversation inside this Project. They define who Claude is when you talk to it through this Project, what it knows about you, what your preferences are, and how it should behave.
Here is the template I give every client to start from. Edit it heavily to make it yours — generic instructions produce generic outputs.
ROLE:
You are my AI Chief of Staff. You know my business, my priorities, my
recent decisions, and the people I work with most often. You exist to
help me think clearly, decide well, and communicate with precision.
ABOUT ME:
[Your name, your role, your company, what your company does, how big the
team is, what stage you are at, what your customers care about most.]
PRIORITIES THIS QUARTER:
1. [Your single highest priority]
2. [Your second priority]
3. [Your third priority]
RECENT MAJOR DECISIONS:
- [Decision] — [Why you decided that, in your own words]
- [Decision] — [Why]
- [Decision] — [Why]
KEY PEOPLE:
- [Name, role, what they own, when I should loop them in]
- [Name, role, what they own, when I should loop them in]
- [Name, role, what they own, when I should loop them in]
HOW I WORK:
- I prefer [bulleted lists / prose / one-page memos] for [scenario]
- I do not want [adjectives, hedging, false certainty, etc.]
- When you do not know something, say so directly
- When I am about to make a mistake, tell me before I commit to it
- My communication style is [direct / warm / formal / casual]
WHEN I ASK FOR A DRAFT:
- Write in my voice, not a generic professional voice
- Match the length of similar work you have seen me write
- If you do not have enough context to write the draft, ask me the
three questions you need answered first
WHEN I ASK FOR ANALYSIS:
- Lead with the conclusion, then the reasoning
- If the answer depends on a value judgment, say so explicitly
- Push back if you think I am framing the question wrong
WHEN I ASK FOR ADVICE:
- Be honest before you are kind
- Use what you know about my priorities and recent decisions
- If my new question contradicts a recent decision, point that outThat is the starting template. The version you actually run will be three times as long because you will keep adding specifics as you use it. That is the right pattern. Generic instructions are useless; specific instructions compound.
Step 3 — Upload the right files
Open the Project's file uploader and add the documents that should always be in context. The list I recommend for most executives:
- Your current strategic plan or OKRs
- The last two board memos (or investor updates if you are pre-board)
- The last quarter's all-hands deck
- A team roster with roles
- Your company values or operating principles, if you have written them down
- Anything you would hand a new hire on day one to give them context
Resist the urge to upload everything. The Project should hold what is always relevant, not what might be relevant once. Anything that comes up rarely is a per-conversation paste, not a Project file.
Step 4 — Connect the MCP servers
This is where most people stop, and it is where the system goes from useful to remarkable. Claude Desktop has built-in MCP support; the browser does not handle this as cleanly. If you have not installed Claude Desktop, install it now.
The MCP connectors worth setting up first, in order:
- Google Drive. So Claude can pull any document in your Drive without you having to upload it.
- Slack. So Claude can summarize channels, pull thread context, and reference things people said.
- Calendar. So Claude can see your week, prep you for meetings, and notice when you have over-committed.
- Linear / Asana / Notion (whichever your team uses for work tracking). So Claude can see what is actually in progress, not just what you remember being in progress.
- Granola (if you record meetings). So Claude can reference any conversation you have had without you re-explaining what happened.
Each connector takes about five minutes to set up. The total time is small. The value is enormous, because every connector turns Claude from "a system you have to brief" into "a system that already knows."
Step 5 — Build your first three Skills
Skills are forward-slash commands you can run from inside any Claude conversation. Inside your Project, they pick up the Project context too. The three I tell every executive to build first:
- /morning — runs at the start of the day, surfaces your calendar, top three priorities, anything waiting on a decision from you, and any inbox items that look high-stakes.
- /pressure-test — takes a decision you are about to make and argues the strongest case against it.
- /end-of-week — Friday afternoon, reviews the week against your priorities, flags what slipped, drafts the Sunday-evening internal note to the team.
You can write these as plain English Skills in about fifteen minutes each. The templates and instructions are walked through on claudeforeveryone.com, including the morning check-in version specifically.
Step 6 — Use it for a week before you change anything
The temptation after the setup is to keep tweaking. Resist it. Use the system as built, every day, for one week. Then sit down on Friday and update the Project Instructions based on what worked and what did not.
The pattern that emerges by week two is that the system becomes much more specific to you. The generic template becomes your specific operating frame. You start adding notes like "when I ask about hiring, default to the answer that errs on the side of waiting" or "my CFO prefers a one-page summary, not a deck" or "when I am writing to investors, lean formal — when I am writing to the team, lean direct but warm." Those are the notes that make the system feel like yours.
Three use cases where this actually pays for itself
The setup is the easy part. The value comes from using it for the right work. The three places I see it pay back fastest:
Board memo prep. You drop your three previous board memos, your current quarter's data, and your draft into the Project, and you ask Claude to help you find the version that lands the message most precisely. Forty-five minutes of work that used to take three hours. The improvement is not the speed — it is that the version you end up with is better than what you would have written alone, because Claude pushes back on the framings that are easy to fall into when you are writing under deadline.
Hard conversation prep. Before a meeting where you have to deliver difficult news (a layoff, a missed goal, a partnership ending), you talk through it with Claude. Not to rehearse the script. To pressure-test the framing, find the questions the other person will probably ask, and figure out what your honest answer is to each one. You walk into the meeting more prepared and less anxious. This is where the "thinking partner you can interrupt at midnight" claim becomes literal — these are the conversations you cannot have with anyone on your team because the conversation IS about them.
Weekly synthesis. Friday afternoon, you ask Claude to read the leadership Slack channel for the week and tell you what you missed. Not what was discussed. What you missed. The right answer is almost always two or three things — a customer complaint that did not escalate properly, a decision someone made without looping you in, a contradiction between what two team members are doing. The thirty minutes you spend on this every Friday catches things that would have been three-week problems by the time you noticed them.
The mistake most people make
The most common failure mode is treating the Chief of Staff as a search engine. You ask it for information, it gives you information, you accept the information without pushback, and the conversation ends.
That is not what this is for. The value of a Chief of Staff is judgment, not information. If you are using Claude to look things up, you are using it as Google. If you are using Claude to think alongside you about something hard, you are using it as a Chief of Staff. The difference is enormous.
The way to tell which mode you are in is whether you finish the conversation having changed your mind about anything. Search engines never change your mind. Good Chiefs of Staff change your mind once a week.
Maintenance — thirty minutes, every Friday
The system gets stale if you stop curating it. The maintenance ritual is thirty minutes on Friday afternoon:
- Update the Recent Major Decisions section with anything material that happened this week
- Re-read the Priorities This Quarter section and confirm it is still true. Edit if it is not.
- Skim the How I Work section and see if there is a new preference worth adding, based on something Claude did wrong this week or something you wished it did differently.
- Archive any uploaded files that are no longer relevant. Add new ones if the strategic context shifted.
This is the part most people skip. It is also the part that determines whether the system stays useful for years or becomes a fossilized artifact of how you were running things six months ago.
What is coming in the next twelve months
A few things on the horizon that will make this system meaningfully better:
- Claude Cowork for running the Chief of Staff workflow on a schedule (the /morning Skill running automatically before you wake up, instead of you remembering to run it)
- Scheduled tasks for proactive flags ("Claude noticed you have not had a 1:1 with your CFO in three weeks, want me to draft the calendar invite?")
- Better long-context retrieval so Projects with hundreds of files stay performant
- Better cross-Project context so your client Projects can reference the Chief of Staff Project when relevant
The version of this system that exists in 2027 is going to feel like science fiction compared to the one you can build today. The version you can build today is already worth more than most things you will set up this year.
How to get started
The fastest path is to set this up yourself, this week, using the template above. It is genuinely doable in an afternoon if you block the time and resist the urge to keep tweaking before you have used it.
If you want to skip the writing-from-scratch step, there is a free Strategic Advisor / AI Chief of Staff Generator on claudeforeveryone.com. Answer a short series of questions about your business and it produces a complete BRIEF-framework instructions file you can paste straight into a new Claude Project. About five minutes start-to-finish.
If you want to skip the trial-and-error and set it up live with someone who has built dozens of these — for you specifically, with your priorities and your team and your tooling — that is what the Quick Start 1:1 is for. Three sessions, $1,447, you walk out with your Chief of Staff configured and a clear path for what to build next.
If you are setting one up as part of a broader team rollout, the Chief of Staff is usually the first artifact of the Clarity Strategy Session — the architecture that everything else gets built around.
If you want to read the landing-page version of this with the visual walkthrough and the quick-start checklist, the /chief-of-staff page is where I keep the always-current version.
The Chief of Staff is the most leverage you will get out of a single afternoon with Claude. Build it.
Nicole Patten is the founder of Elevate Online and runs a Claude-specific training practice. She spent 7 years at Google as a Senior UX Engineer before dedicating her career to helping teams use AI responsibly and effectively. 100% of her business runs on Claude — including the Chief of Staff Project that handled the draft of this post.