The same assistant does a different job at each level. Here's what changes for individual contributors, managers and executives, and how to tell it's working at yours.
AI executive productivity isn’t manager productivity with a bigger title. An individual contributor checks AI against a source on their screen, a manager checks it against their own notes, and an executive often decides on other people’s summaries, so the executive’s first job for AI is to find the holes in a pack rather than to give the answer. Pick your level, choose one recurring load, write down the before number, and check at week eight whether it’s still running and whether an output your level owns has moved.
Say an analyst, a director and a VP all open the same AI assistant at half past eight on a Monday. The analyst pastes in a messy sales export and asks for a first-pass summary. She’ll check the totals against the export before anything leaves her desk. The director pastes in six weeks of one-to-one notes and asks what he’s promised his team and not yet done. The VP pastes in a 30-page pre-read for Wednesday’s investment committee and asks whether the committee should approve it.
All three get a fluent answer in about ten seconds. The first two asked for something they can check with their own eyes. The VP asked the tool to make a call at the point where she’s furthest from the source, and the answer will sound just as sure whether it’s right or not.
That’s the argument of this article. AI executive productivity isn’t manager productivity with a bigger title, because what you’re accountable for, where your information comes from and what a mistake costs all change as you move up. So the first task, the check and the measure change too. The sections below take individual contributors, managers, directors and VPs, and executives such as the CEO, COO, CFO and CHRO in turn, with the chief of staff as a role of its own.
If you want the personal stack an executive builds, a thinking partner plus a delegation tool, Hina’s personal playbook for using AI in your own role covers it. This piece works on either side of it: which level of the job you’re doing, and how to tell whether AI is working there.
It’s tempting to read who uses AI most as a scoreboard for who’s ahead. Gallup’s fourth-quarter 2025 workforce survey of US employed adults found 44% of leaders using AI a few times a week or more, against 23% of individual contributors.[1] Gallup also notes that the gap has grown since mid-2023.
Share of US employed adults using AI at work a few times a week or more, by their own level. Source: Gallup workforce survey, 30 October to 14 November 2025, published 25 January 2026.
Managers sit in between at 30%, and Gallup’s own explanation is that leadership and management roles are more often desk-based and built around writing, planning, analysis and communication, work that AI tools handle readily.[2] That explains why AI is convenient to use at these levels. It doesn’t say what makes it useful there.
Both sets of numbers count how often people use AI, and frequency says nothing about whether a leader’s decisions got faster or better. That gap matters most at the top: a leader can log frequent use while nothing about the week has changed. Frequency is only the first step of Use, Persistence and Impact.
The numbers cover US employees only, and the groups are people’s own labels. Nothing in them says which tasks the leaders were doing, which is the question the rest of this article works on.
Here’s the difference that matters, laid out so you can find your own row. It’s our own framing, a judgment about how these jobs are usually organised, not a published finding, and real jobs blur the lines.
| Level | Accountable for | Information arrives as | Best first AI job | Mistake this level tends to make |
|---|---|---|---|---|
| Individual contributor (analyst, coordinator, specialist) | A finished piece of work | Raw: the files, data and threads you work in | Draft or tidy something you’ve produced often, then check it against the source | Sending more, faster, without counting the rework |
| Manager or senior manager | A team’s output and follow-through | Part raw, part what your team tells you | Turn one-to-one notes into commitments you can check | Letting AI write the words about a person that your own judgment should have shaped |
| Director or VP | Several managers’ output and the trade-offs between them | Mostly your managers’ summaries, with the raw material a message away | Cross-check those summaries for gaps before you decide | Treating a summary of summaries as if it were the source |
| Executive (CEO, COO, CFO, CHRO) | Decisions, and the company’s answer for them | Briefed by others, each accountable for their own part of the story | Test the pack before the meeting, track what was promised after it | Asking the tool for the answer instead of for the holes |
| Chief of staff | The quality of what reaches the executive | Raw material from across the company | Build the brief so every claim carries its source | Producing a smooth brief that hides where each claim came from |
A Future Factors framing based on our own judgment, not a published benchmark.
A quick way to place yourself is to ask whether you can reach the source in one message. If you can, you’re on the manager or director row and can verify what the tool tells you in a couple of minutes. If you can’t, you’re on the executive row, whatever your title says. Typical signs:
A large set of CEO diaries shows why that row looks the way it does. Economists followed 1,114 CEOs of manufacturing firms across six countries, using a week of diary calls with the CEO or an assistant, and recorded that the average CEO spent about 70% of working time interacting with other people, mostly insiders.[3] It’s one type of firm and one type of executive, so treat it as an illustration. It fits the picture, though. When most of your day is other people talking to you, most of what you know arrives as somebody else’s account.
Choose your first AI use from what your level is accountable for, not from what the tool is good at.
For a Copilot-specific view of the executive row across the CEO, CFO, COO and CHRO, see our Copilot role guide for executives. Everything here works with any assistant.
An individual contributor’s early wins tend to come on work where the source is on the screen: a weekly summary you’ve written a dozen times, a thread to answer, a table to tidy. That’s a good place to start, because you can tell in seconds whether the output is right. A randomised field experiment shows what it buys you. Across 66 firms and 7,137 knowledge workers, those who used an AI tool built into email, meetings and writing spent about two fewer hours a week on email in the second half of the six-month experiment, and the authors detected no shift in the quantity or mix of tasks people did.[4]
These were knowledge workers in general rather than executives, and the result is a fair picture of the individual contributor’s position: the saved minutes are real, and they don’t turn into a different job on their own. In our judgment the level’s typical mistake is to bank them as more of the same output, so decide up front what the recovered time is for.
Managers get a different kind of first job, because their material is partly their own notes. Say you manage eight people and you’ve kept rough notes from one-to-ones for six weeks. You couldn’t say what you promised one team member about the training budget in week two. That’s a good job for AI, since the answer sits in text you own and you can check every line against it.
| Below are my one-to-one notes for the last six weeks with the eight people on my team, with names replaced by initials. 1. List every commitment I made to someone, with the date I made it and whether the later notes show it was done. 2. List anything a person raised twice or more without a recorded outcome. 3. For each item, quote the line from my notes it comes from. Do not judge anyone’s performance and do not suggest what I should say to them. If a note is unclear, say so instead of guessing. [paste notes] |
Use it only in a tool your organisation has approved for people data. Quote-the-line is what makes the output checkable.
The boundary for this level sits in the prompt’s last paragraph. What you say about a person stays yours: AI finds the commitment you forgot, and you decide what to do about it. Directors and VPs can use the same prompt on their managers’ summaries, with one addition: ask for the claims that have no supporting detail. For the rest of a manager’s week, our guide to using AI to be a better manager covers meeting prep, feedback drafts and difficult conversations.
Say you’re the COO and the monthly operating review pack lands on Sunday night: forty pages, five functions, each section written by the person accountable for that function’s numbers. You could ask an AI assistant what should worry you. You’ll get a tidy paragraph built from five accounts, each written by someone with a stake in how the month looks, so the answer tends to inherit their optimism.
The more useful question is about the pack itself: where it disagrees with itself, which claims have no number behind them, and what quietly changed since last month. Those are questions about gaps, and you can check every answer by turning to the page it names.
The further you sit from the source, the more the brief has to show where each claim came from.
| I’m the COO. Below are this month’s operating review pack from five functions and last month’s pack. I’m not asking what to do. I’m asking where the pack is weak. 1. List places where two functions describe the same event or number differently, with the page for each. 2. List claims of improvement or risk with no number or date behind them. 3. List what was in last month’s pack and is missing or reworded now. 4. Give me the five questions I should ask in the room, each tied to a page. Quote the line each item comes from. Do not recommend a decision or rank the functions. If you can’t tell, say “unclear” instead of guessing. [paste this month’s pack] [paste last month’s pack] |
Swap in your own functions and pack. Check the two or three items you’d act on against the pages they name before you carry them into the room.
Executives already use AI, though lightly. A survey of senior executives in the US, UK, Germany and Australia, fielded between November 2025 and January 2026, found respondents using AI about 1.5 hours a week on average, with 28% not using it at all in the working week.[5] Those are self-reported hours from a mix of CEOs, CFOs and other senior managers. It doesn’t follow that executives should use it more. A pack check before a review and a commitments check after it take a few minutes each, and they land at the two points where they change what happens in the room.
That’s where the human split matters. Here’s how we’d write it down for two executive tasks:
| What AI does | What the executive still owns | How it gets checked |
|---|---|---|
| Lists conflicts between functions and unsupported claims in the pack | Which gaps matter, and which function head to ask | You read the pages behind the top three items before the meeting, then note how many were real gaps |
| Turns the meeting transcript or your notes into decisions, owners and dates | Whether that is what was decided, and what goes out under your name | Read the list against your own notes for the first month; owners confirm their line within a day |
An example split for two executive tasks. Adjust the reviewer and the sampling rate to the risk in your own business.
The chief of staff’s version is the Source Rule applied first, because the brief is their product: every claim carries where it came from, so the executive can go one step back when a number looks off. AI can tell you a number is missing. It can’t tell you which function head is being hopeful, and that read stays with you. For the thinking-partner side of decisions, the personal playbook goes into detail. Our guide to board and investor prep shows the same pack-check habit on a board deck, and the anti-hallucination toolkit covers checking habits in general.
The trap at every level is to measure what’s quick to count. Frequency is quick, and so is “time saved” as people remember it. Economists who linked Danish worker questionnaires to administrative records show how far the remembered version can sit from the recorded one. Workers in eleven occupations, among them accountants, HR professionals and marketing professionals, reported time savings from chatbots, yet the records showed precise null effects on earnings and recorded hours, ruling out effects larger than 2%.[6]
That doesn’t prove the savings aren’t real. The authors point toward reorganised tasks rather than changes in recorded outcomes, which is why a worker’s own estimate can’t stand in for an outcome. Our guide to botsitting looks at the same gap inside teams. So choose an impact number your level owns and can count from its own records.
Measure an output your level owns, not the minutes you remember AI saving.
Here’s one recurring load per level, filled in with the Use, Persistence and Impact lens we apply to adoption work. The specifics are ours, for illustration, not benchmarks.
| Level and load | Use (weeks 1 to 4) | Persistence (week 8) | Impact (compare with the four weeks before) |
|---|---|---|---|
| Individual contributor: the weekly status update | AI first draft used in at least 3 of the 4 weeks | Still doing it, with no one having asked | Minutes from start to sendable, timed on three updates before and three after, plus reviewer corrections |
| Manager: one-to-one follow-ups | Follow-up finder run after each round of one-to-ones | Still run each round, and the list gets acted on | Share of commitments made in one-to-ones that were done by the date you promised, counted from your own notes |
| Executive: the monthly operating review pack | Pack check run before each of the four reviews | Still run in month three, without the chief of staff having to remind you | Questions in the room that came from the check, and decisions made in the meeting rather than deferred, across four reviews against the four before |
A Future Factors example, one load per level. Swap in your own, and write the before number down first.
Two things make the executive row harder. A dozen or so reviews a year means small counts, so treat a change as a hint, not a result. And the chief of staff or a peer has to be asked to reinforce the habit, because nobody else is placed to check an executive’s use. Our guide to tracking time savings from AI covers how to time a task properly for the individual contributor row.
You can run this yourself:
You can get a long way alone. Expert-led, role-specific training tends to shorten the learning curve and to build shared capability across a leadership team instead of leaving it with one enthusiast. It also helps to keep the four ideas apart. Training builds the capability. Workflow design gives it a named, recurring load to live in, which is what step two does. Reinforcement, the week-eight check, turns it into a habit. Adoption is what you see when all three have happened.
Our guide to the AI-powered professional covers how a person changes the way they work, and it’s the one to read next if the week-eight check comes back mixed. Our corporate training is built around a team’s real roles and workflows.
This week, do the first three steps and nothing else: find your row, choose one load, and write down the before number.
Run the one-message test again on each recurring load. Some of what you did as a manager, such as following up one-to-ones, stays, but a new load appears: summaries from your own managers. Retire habits from the old level, such as drafting everything from raw material, and add one check on those summaries.
The chief of staff produces the brief, so the discipline sits with them: a source line against every claim, and a mark on anything the model inferred instead of quoted. It also helps to log what the executive changes in each brief. After a month, the log shows which kinds of claim keep needing correction, and that’s what to check first.
Then keep that material out of it until someone with authority has answered two questions: where does the tool store what you paste, and can it be used to train models. Meanwhile you can practise the prompts on a redacted or invented pack, which builds the habit before the policy catches up.
Usually, yes. Generic tool training builds a basic capability, but the task, the check and the measure differ by level, so practice needs to happen on each group’s own recurring work. Expert-led, role-specific training tends to shorten that learning curve.
Pick one summary a month, choose the claim you’d act on most, and read the original for that claim only. Note whether the summary was accurate, left something out or leaned optimistic. After three or four rounds you’ll know how much checking that source needs.
The leader, manager and individual contributor comparison comes from two Gallup workforce articles read on gallup.com, one published 25 January 2026 (survey of 22,368 US adults, 30 October to 14 November 2025) and one published 12 April 2026 (survey of 23,717 adults, 4 to 19 February 2026). The Gallup text I could read didn’t define who counts as a leader, manager or individual contributor, so treat those groups as respondents’ own labels. The CEO time figure comes from the Journal of Political Economy paper by Bandiera, Hansen, Prat and Sadun, read from an author-hosted PDF, and it covers manufacturing firms only. The field experiment and the executive survey were read on NBER (working papers 33795 and 34836, the first with its arXiv abstract), and the Danish administrative-records study on NBER (working paper 33777). Porter and Nohria’s Harvard Business Review article on CEO time was left out because only its teaser is readable without a subscription, and our earlier executive playbook already cites it. The level table, the prompts and the register are our own frameworks and illustrations, and every named scene is an illustration, not a real client story.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.