What is an AI Chief of Staff?

Every company has bought the AI seats, and very few have changed how they work. In our AI Readiness Diagnostic, 86% of Chiefs of Staff and senior operators use AI every day, yet only 7.3% scored as AI-native, meaning AI runs in their workflow on a schedule, without anyone opening a chat window.

The best tools for knowledge work are only as good as the context they're given, and a Chief of Staff holds more of that context than anyone. CEOs weighing software against a hire will find that decision in the first three sections. If you're new to the role, start with our Chief of Staff handbook.

What is an AI chief of staff?

An AI chief of staff is usually software: an assistant connected to a leader's email, calendar and meeting notes that writes briefings, tracks follow-ups and drafts replies. The same phrase is also used for a job, a human Chief of Staff whose mandate is leading AI adoption, and loosely for any Chief of Staff who works through AI. Mixing those meanings up leads to bad decisions on both sides.

Good assistant software never forgets a follow-up, and it turns a leader's morning scramble through email into a short read. What it can't do is decide which of twelve open threads should reach the CEO today, or tell the head of sales that her number isn't credible.

Compare them on what each returns before you compare price. Software gives back minutes on follow-ups and drafts, for a subscription. A Chief of Staff gives back executive hours, gets decisions made sooner and ships cross-functional projects, and averages $167,954 in base salary, according to the Chief of Staff Network's 2025 Salary Report (nearly 500 respondents).

If your CEO asks whether they "need an AI chief of staff", ask what problem they want gone. Follow-ups and meeting prep need an assistant; decisions, alignment or projects nobody owns need a person, and our guide to hiring a Chief of Staff is the place to start. If software could replace everything your Chief of Staff did, the role was designed as admin, and our guide to what a Chief of Staff actually does is the better fix.

What's the difference between an AI chief of staff, a Chief AI Officer and a Chief of AI?

An AI chief of staff is software. A Chief AI Officer is a C-level executive who owns a company's AI strategy, investment and risk. A Chief of AI is usually a Chief of Staff, or another senior operator, who takes AI adoption on as a portfolio alongside the rest of the job.

All three show up in the same conversation: a CEO reads about AI assistants, hears that peers are hiring a Chief AI Officer, and asks their Chief of Staff to "own AI". Those are three different purchases with three different costs.

Pay figures for postings come from the Chief of Staff Network's 2026 Pathways Report (more than 1,800 job postings); the base-pay average comes from our 2025 Salary Report.

My take: a company under a few hundred people needs someone with cross-functional context to pick the first real use cases, write the rules and measure what changed, well before it needs a Chief AI Officer. That's a Chief of AI, and it's often the Chief of Staff. A Chief AI Officer earns the title when AI becomes a product line, a large budget or a regulated risk. If your CEO is weighing that hire, write down the three AI decisions they expect the person to make in six months, and if you could make them with a clear mandate and budget, make the case for doing it yourself first.

Can an AI chief of staff replace a human Chief of Staff?

No. AI will replace the parts of the Chief of Staff's week that never needed a Chief of Staff, such as the meeting notes, the status deck and the first draft of the board update. It won't replace the role, because what remains is judgment, trust and the context behind both.

What separates a Chief of Staff from a capable analyst is knowing that the CFO will push back on the hiring plan because of a conversation last Tuesday, and that the CEO says "growth" but means one enterprise deal. That knowledge sits in one person's head, next to the leader, where no AI system can read it.

Sawyer Middeleer was the first hire and Chief of Staff at Aomni, an AI software company, and now helps operators bring AI into their companies. His verdict on AI chief of staff products was blunt.

Why the replacement math looks better than it is

Leaders look at a task from the outside, see what goes in and what comes out, and assume a bot can do the middle cheaply. Sawyer's example was a support ticket: a request arrives, a reply goes out, the ticket closes. The person answering it carries context and a sense of what a good reply looks like, and nobody ever wrote either down, so automating before you've mapped that middle gets you faster, worse work.

A Chief of Staff's week is that problem at its most extreme. The board update starts as metrics and ends as a deck, and in between sit the calls: which bad news leads, which director's number gets questioned first, what the CEO promised last quarter that this deck has to answer for. Sawyer also noted that the line between what AI should and shouldn't do keeps moving, so I'd redraw it every quarter, task by task.

The work that moves, and the work that gets harder

Alex Craig, Chief of Staff at the cybersecurity company PortSwigger, built himself an AI system that prepares his mornings, his one-to-ones and much of his board prep, and wrote up how it works for us. His morning context rebuild went from about 30 minutes to a 5-minute read, and his prep for a one-to-one from about 15 minutes to 2.

Then he told us in his episode on the incognito test that AI made his job harder. The routine work it took away was the restful part of the day, and what's left is decisions, conversations and judgment calls, back to back. The bar rises too, because your principal will notice when a polished draft hides thin thinking.

AI cut Alex Craig's one-to-one prep from about 15 minutes to 2, and left his days full of decisions and judgment calls.

Sorting your own week

Take a typical week for a Chief of Staff to a CEO, as a labeled illustration built from outputs most Chiefs of Staff produce every week.

1. Leadership meeting notes, about 2 hours. The transcript and action list go to AI. You keep the ten minutes deciding which unresolved disagreement to raise with your principal.
2. Monday metrics summary, about 3 hours. AI pulls the numbers and drafts the summary. You keep the paragraph on why a number moved, which needs a conversation with its owner.
3. One-to-one prep with your principal, about an hour. AI assembles the open threads. You choose the three that get airtime.
4. Board update first draft, two days a quarter. AI drafts from the context file. You set the order of bad news and check every commitment.
5. A deadline dispute between two VPs. Only the prep notes go to AI.

Close to a day a week of drafting moves to AI, and none of the five decisions do. The test for the sort: if your principal would act differently depending on a choice you made inside the artifact, that choice is yours. If they read an output and change nothing, it was admin, and it should go, however visible it makes you.

The opposite mistake is quieter. For many Chiefs of Staff, writing the meeting notes is how they stay across everything, so hand that to a notetaker and you lose the moment you absorbed what happened. I'd replace it on purpose with fifteen minutes a day on the summaries and the decision log.

What happens after the admin goes

Once the morning brief arrives on its own, it stops counting as your work, and by month three your principal will ask what you did with the hours. Have an answer that names a project shipped or a decision unblocked. Other leaders get the same tools, so keep capturing what never reaches a transcript: the hallway conversation, the direct message, the reason behind a decision.

At a startup under 50 people, the Chief of Staff often carries operations admin too, so more of the week moves at once and the role can look thin within a quarter; plan the replacement work before you automate. Reporting roll-ups for a functional leader move first, and the board and trust work of a Chief of Staff to the CEO doesn't move at all.

This week, list every recurring output you produce, with the hours each takes, and mark the ones a new analyst could do with a good brief. Those are AI's. Everything else is your job, and it's the part to grow.

Why does AI adoption land on the Chief of Staff?

AI adoption lands on the Chief of Staff because it's new, it crosses every team, and nobody else owns it. The CEO wants it done, the CTO owns the technical fit, and the Chief of Staff ends up running the work in between: choosing use cases, buying tools, writing the rules and getting people to use them.

In the Chief of Staff Network's Chief of AI Report, the CEO was the primary decision maker on adopting AI 31% of the time, with the Chief of Staff and the CTO tied next at 10% each. On choosing the tools, the Chief of Staff was involved almost as often as the CTO: 19% of tool selections against 20%, with the CEO at 17%. We call that group the AI Triad: the CEO sponsors, the Chief of Staff scopes and runs the rollout, and the CTO checks the technical fit. Download the full Chief of AI Report Part 2 for the breakdown by company size.

Abhishek Vora, who was Chief of Staff to the CEO at LaunchDarkly when we spoke (he now holds the same role at Asana), explained in his episode on Chiefs of Staff leading AI adoption that cross-company work with no owner is what Chiefs of Staff pick up. Kanika Choudhary, Chief of Staff at People IX, describes the same default in her episode on mapping work before buying AI tools: the work rolls downhill to whoever has cross-functional context and can't say no.

Frame the mandate, then pick a first project worth the risk

A mandate that arrives as "do more AI" has no budget, no authority and no definition of done. Kanika's advice is to answer it with questions: what's the point, what's the budget, and what can you decide alone? In our Chief of AI Report, 27% of companies funded AI from no specific budget, another 27% from the general software budget, and only 20% had a dedicated AI budget.

My take: say yes to owning AI, but only after you've written down what owning it means. A pilot nobody measures never technically fails, which makes this one of the easiest projects to fail at without anyone noticing.

The first project decides whether you get a second. Kanika sees Chiefs of Staff pick the process they can't possibly mess up; leadership sees no difference and decides AI doesn't work. I pushed back that small wins prove the tools work, and her answer was that it depends on what a win costs. Where a change takes three months and four signatures, spending that approval on a small fix burns credibility, so build a bare-bones prototype of the bigger idea, with no company data in it, and show leadership that.

Count the approvals to tell which situation you're in. If you can switch on an approved tool for one team this week, take the small, visible win and make it touch at least two teams. If every change needs security review, procurement and legal, spend that queue on a process leadership already complains about.

Company size changes the mandate too. Companies under 500 people leaned more on "no specific budget", and up to 19% left AI tool choices to individual employees, so your first job there is getting AI onto a budget line. In larger companies the triad repeats inside departments, such as the general counsel, head of legal operations and a senior engineer in legal, and a Chief of Staff to a functional leader sits in that department's middle seat.

This week, write a one-paragraph AI mandate and send it to your principal: the return you expect, the budget line, the decisions you can make alone, the ones that need them, and the date you'll report back with one result, such as "board prep down from five days to three by the end of the quarter". Our guide to the Chief of Staff mandate has the one-page template, and if the role is being hired for AI, the same fields belong in the job description.

How do you get useful output from AI as a Chief of Staff?

You get useful output by giving AI the context only you hold, then checking what comes back. We call that context management: capturing what the company knows, curating it into something AI can use, directing AI with that context, and checking the result.

Alex Craig uses a test at PortSwigger. When someone shows him AI-assisted work, he asks whether he could have produced the same thing in a fresh, incognito chat with no history. If he could, the person added nothing: their knowledge of the company and their preferences are missing, and the output is the generic answer anyone would get.

Curate is the step most people skip, and it decides the quality of everything after it. Alex's system runs on one folder holding about six months of decisions, corrections and role context, which he treats as its most valuable part. [SCOTT: one or two lines on the workspace you run your own week from, and what it changed?] The full template, with the decision log and commitments list that keep it current, is in our guide to context management.

Direct means a request that reads like a brief to a sharp new analyst. "Summarize this" gets a summary; "Draft a one-page memo for the CEO choosing between these two vendors, recommendation first" gets something you can use.

Check, because AI's mistakes look as polished as its correct work. Alex's system assigned a task directly to his boss without telling him, credited one colleague's update to another, and kept preparing notes on someone who had already left.

Check any AI output about a person, a number or a commitment against the source before it leaves your hands, and add each correction to the context file.

How does a Chief of Staff use AI across the four core responsibilities?

In each of the four core responsibilities, priorities and decisions, the operating rhythm, communication and cross-functional projects, AI takes the drafting and sorting, and the Chief of Staff keeps the call. Every request starts from the context file.

Priorities: decide faster with better memos

AI is good at preparing a decision: pulling facts together, laying out options, stress-testing a recommendation. It's bad at knowing which decision matters most this week. The highest-value use is the decision memo, decision first and options in a table; with the context file, a first draft takes minutes, and your time goes into whether the recommendation is right.

You are helping me, the Chief of Staff to [principal, title], prepare a decision memo.

Context file: [paste or attach]
Decision needed: [one sentence]
Deadline for the decision: [date]
Options we're considering: [list]
Inputs: [paste notes, data, emails from the people involved]

Write a one-page memo:
1. The decision needed and the recommendation, in the first two sentences.
2. A table comparing the options on cost, time, risk and fit with this quarter's priorities from the context file.
3. What we're giving up with the recommended option.
4. Who disagrees and why, using only what's in the inputs.
5. The next step and owner if approved, and any earlier commitment this decision affects.

Rules: use only facts from the inputs and context file. Mark anything you're unsure of as [CHECK]. Don't soften disagreements. Keep it under 400 words.

Then ask the model to argue against its own recommendation, and learn before the meeting whether the counterargument is stronger than you expected.

Rhythm: run the operating rhythm with less admin

The operating rhythm is where AI saves the most raw time: agendas built from last week's open actions, condensed pre-reads, notes turned into owners and dates within the hour. Eric Nehrlich, former Chief of Staff for Google Search Ads, told us annual cycles no longer make sense when plans change every few weeks, and AI makes a faster weekly and quarterly rhythm affordable because review prep no longer takes a day.

Our guide to building an operating rhythm covers the meetings themselves, and our OKR vs KPI guide has a weekly KPI scorecard you can copy. In the board update, put the most care into what you promised last time and haven't done, because boards remember.

Communication: hear what's going on

A Chief of Staff tracking forty stakeholders can't hold every open thread in their head. With meetings recorded on approved tools, AI can read a week of notes, tell you where things stand with each person, and catch what the room missed, such as three leaders raising the same concern in separate meetings. The output feeds your weekly update to your principal.

You are helping me, the Chief of Staff to [principal], prepare my weekly stakeholder review.

Context file: [paste or attach]
Meeting notes or transcripts from this week: [paste or attach]
People I track: [list names and roles]

For each person, in a table:
- Where things stand on the work we share
- Open commitments: what I owe them and what they owe us, with dates
- Concerns or objections they raised, in their own words where possible, with the meeting it came from
- Anything that contradicts what they said in a group setting

Then list:
1. Concerns raised by two or more people, even in different words
2. Commitments due in the next 7 days
3. Anything my principal should hear this week, ranked

Rules: only use what is in the notes. Never guess at motives or feelings. Mark anything uncertain as [CHECK]. This summary is for me only.

A summary of what people said is useful. A guess at why they said it is gossip with a confident tone, so keep anything about a person's motives, performance or conflicts out of shared tools.

Projects: build the first version yourself

Cross-functional projects are the important work no team owns. AI changes this responsibility the most, because a Chief of Staff can now build a working first version and show it, where they used to write a proposal, which is Kanika's prototype-first habit applied to every project.

Which AI tools does a Chief of Staff need?

A Chief of Staff needs three layers of AI tools: a general-purpose assistant for thinking and drafting, a productivity layer for meetings, documents and knowledge, and an integration layer that connects tools and runs work on a schedule. Almost every Chief of Staff has the first layer, and very few have the third.

The top half of that chart is the first layer, and drafting, summarizing and briefing are now normal. The bottom half is where the time savings compound, and where almost nobody is yet.

In our Chief of AI Report, a single general-purpose assistant was the most common company AI tool, named by 34% of respondents; pick the one your company allows and get very good at it before adding anything. Our guide to AI tools for Chiefs of Staff compares the categories, and our AI setup guide walks through building your own system, which took Alex Craig a few hours a week for two months.

How should a Chief of Staff handle AI governance and data privacy?

A Chief of Staff should write the company's AI rules before an incident forces them: which tools are approved, what data can go into them, what always needs a human check, and who decides when the rules change. Security, compliance and ethical concerns are the most common obstacle to AI adoption, at about 41% of responses in the Chief of Staff Network's Chief of AI Report, ahead of adoption and change management at about 38%.

The risks are concrete: a leadership meeting recorded by a notetaker nobody approved, board materials pasted into a personal account, a summary guessing at a colleague's motives saved in a shared folder. Each one only takes a busy person and no rules.

Delong Lu, Chief of Staff at Proton, the Swiss privacy company, had to adopt AI quickly at a company whose whole promise is privacy. In his episode on steering AI at a privacy-first company, he described getting leadership to agree on values first and letting those decide the tool questions, then writing explicit rules for how engineering uses AI. A policy that contradicts the company's values gets called out by its own people immediately.

Write the policy on one page and make the approved path the easiest one, because people route around rules that make their job harder. Adapt this version with your legal and security leads.

# How we use AI: [Company] leadership team
Owner: [Chief of Staff] · Approved by: [CEO], [legal/security lead] · Review date: [quarterly]

## Why we use AI
To spend less time on preparation and more on decisions, customers and people. AI drafts; people decide.

## Approved tools
| Tool category | Approved tool | Account type | Approved for |
| --- | --- | --- | --- |
| General assistant | [Tool] | Company account only | Drafting, analysis, research |
| Meeting notes | [Tool] | Company account only | Internal meetings with consent |
| Documents and slides | [Tool] | Company account only | Internal documents |

Anything not on this list needs approval from [owner] before use with company data.

## Data rules
- Never put into any AI tool: customer personal data, employee health or performance records, passwords or keys, unreleased financials outside approved tools, anything under NDA unless the contract allows it.
- Board and investor materials: approved tools only, company accounts only.
- Notes about individuals (performance, conflicts, motives) stay out of AI tools and shared folders.
- Personal AI accounts are never used for company work.

## Meetings
- Say when a meeting is recorded or transcribed. Anyone can ask to turn it off.
- Meetings that are never recorded: [the list your leadership team agrees].

## Human check required before anything goes out
- Every number, name, date and commitment
- Anything external: customers, investors, board, press
- Anything about a person
- Nothing is sent in a leader's name without that leader or their Chief of Staff approving it

## Disclosure
- Say when a document is substantially AI-drafted if the reader would reasonably want to know.
- The person who sends it owns it. "The AI wrote it" is never an explanation.

## Questions and exceptions
Ask [Chief of Staff]. Exceptions are approved by [owner] and logged.

## How this changes
We review this page every quarter and after any incident. Proposals go to [Chief of Staff].

Treat this as an operating policy for your legal and security leads to sign off, and expect a regulated company to need more; our AI & the Law course covers AI risk and contracts. A policy also works only if the approved tools are good enough that people prefer them to the one they use at home.

This week, find out which AI tools your leadership team actually uses, and compare them with the approved list. The difference between the two lists is your policy's first job.

How do you lead AI adoption across a company?

You lead AI adoption by getting a few people to a real result in their own work, turning them into champions who teach their teams, and then adding structure: shared tools, shared context and a way to measure what changed. Most companies are stuck between buying licenses and changing how they work.

Half the companies in our Chief of AI Report were piloting, which is where programs stall: a pilot can run forever without anyone deciding whether it worked.

Stage one: create a few real wins

Zen Foo leads AI adoption across Aspire, the Singapore finance company. On our podcast he described three stages: hype, a chaotic middle where usage spreads faster than anyone can organize it, and governance, where shared tools, knowledge and rules catch up. Early on, people adopt AI the first time it does something real in their own job, which a training deck rarely achieves.

So his workshops keep the same basics for every team and change the first exercise. For the product team, he arrived with ready-made skills that write a product spec in one pass, so the moment happens on the team's own work in the first hour. He's one person, so Aspire's champions program finds at least one excited person per team, builds a first workflow or agent with them, and then steps back while they teach their teams.

I'd spend most of the first month with the champions, chosen by two tests: they're keen, and their team's work runs through a process the company cares about. A keen champion on a side process produces a nice demo and nothing anyone else copies.

Stage two: manage the people as well as the plan

The second most common obstacle in our Chief of AI Report, at about 38% of responses, was adoption and change management: people who had the tools and didn't use them. Dominique Lear, Chief of Staff at the consulting firm Incandescent, argued in her episode on fear and AI transformations that transformations fall short because leaders plan the skills and structure and skip the fear, including their own.

Leaders have to sit with their own fears first and admit how much nobody knows yet, then make room for candid conversations where people say what feels hardest. She described the moment a colleague hears someone else name the worry and says, in effect, "wait, you feel that too?" I'd trust that moment over another training session.

The fear has a specific source. In our Chief of AI Report, 54% of companies named efficiency as their main reason for adopting AI. One Chief of Staff at a midsized fintech put the goal plainly: "My goal is to scale teams without adding headcount OR save $ on headcount but achieve the same results." That's exactly the sentence an employee imagines when they hear "efficiency", so say in the launch note what the time saved will go to, whether headcount plans change, and who to ask.

Stage three: measure what changed

After the excitement comes what Zen called beautiful chaos: non-engineers building apps and agents, and nobody able to say what's being built. His next step was tracking every AI project and setting goals that point teams in a direction, without the trap of a rule like "everyone builds one AI app a quarter", which produces apps nobody needed.

Our Chief of AI Report found that 32% of companies hadn't measured AI's impact at all. Of those that had, the most common measure was time saved (40%), well ahead of financial impact (16%) and quality gains (13%). Measure from the start what changed in the work, because "Board prep went from five days to two" is a line your CEO can repeat to the board.

A 30-minute weekly review keeps this honest: update the context file with the week's decisions and corrections, check every external number against its source, note one use case that worked and one place people routed around the approved tools, and log the hours saved on three outputs you're tracking.

By month three, the champions' teams pull ahead and other teams notice, which is the moment for a second round of champions. Tools multiply, so start a register of AI projects and tools with an owner and purpose for each before two teams buy different answers to the same problem. At a company of 1 to 50 people, where 13% had rolled AI out companywide against about two thirds at 5,000+, you can get the leadership team onto one assistant in a month; at a large one, your job is the register, the rules and the measures.

Is Chief of AI a career path, and how do you build the skills?

Yes, for Chiefs of Staff who can show a result. The market pays for what you changed with AI, and the skills come from using it on real work with real context.

We describe the Chief of AI in three parts: a mindset that treats AI as a change in how work gets done, capabilities that go past prompting into systems and agents, and a portfolio, owning the AI direction for your principal with a point of view. I wrote about the distance between being a power user and an AI leader in What it means to become Chief of AI.

One in four Chief of Staff postings ask for AI skills, according to the Chief of Staff Network's 2026 Pathways Report, yet those postings don't advertise higher pay on their own; companies that talk about AI pay about 8% more. AI-first Chiefs of Staff in our 2025 Salary Report out-earn AI novices by about $40,000, and those at companies still AI-curious average $243K. The full breakdown is in our salary guide.

A month is enough to get past the power-user stage. Week one, map one process end to end before touching a tool, because, as Kanika Choudhary put it, "you cannot automate what you can't see." Week two, build the context file. Week three, automate one recurring output, such as the morning brief, and in week four measure it and show your principal the before and after. For the full set of skills, including the eight competencies in our Leveling Framework, see our guide to Chief of Staff skills.