Leading a company-wide AI rollout and personally using AI well are two different jobs. This is a playbook for the second one.
An executive’s personal AI stack has two modes, not one tool. Thinking partner mode is for judgment calls: AI tests your reasoning, argues the other side, and stress-tests a decision you’re going to make and sign your name to. Delegate mode is for bounded, defined work: first-pass drafts, prep documents, status pulls, the stuff that doesn’t need your judgment, only your review. The two get confused constantly, usually in the direction of delegating something that was actually a judgment call. This piece covers how to tell the two apart, what to keep doing yourself on purpose, and what a working week looks like once the habit is real instead of occasional.
Say a VP of Operations at a mid-size logistics company spends a quarter doing exactly what she’s asked to do. She picks a tool, negotiates the licenses, runs training sessions for four departments, and builds a rollout dashboard she presents to the leadership team every month. By quarter’s end, the adoption numbers look genuinely good. Then a board member asks her, almost in passing, what she personally uses AI for in her own week. She has to think about it. The answer is barely anything. She has spent a full quarter running an AI program without really using AI herself.
That gap is common, and it isn’t really about time. It’s that most of what gets published under “AI for executives” is written for one specific job: leading adoption. Which tool to pick, how to get a skeptical team on board, how to talk about ROI, how to build the governance structure around it. We’ve written pieces in that category ourselves, including how to build a team of AI agents for your company and a role-by-role guide to where Copilot helps a leadership team. Both are useful. Both are aimed at you as the person running the rollout, not as the person doing the work.
This piece is about the other job, the one nobody puts on a dashboard: what you personally do with AI at your own desk, on your own decisions, in the parts of your week nobody else is auditing. Using AI well as an executive is a personal skill, not a rollout decision, and the two require genuinely different habits. Knowing how to get a hundred people using a tool tells you almost nothing about how to use it yourself for the call only you can make.
Before deciding where AI fits into your own week, it’s worth being honest about where the week actually goes, because it’s rarely where people assume. Harvard Business School researchers Michael Porter and Nitin Nohria spent over a decade tracking this directly: they had the executive assistants of 27 CEOs code every 15 minutes of their time, around the clock, then verified the coding with each CEO. Roughly 60,000 hours of data came out of it, and the finding that lands hardest is how little of that time was ever a candidate for AI to begin with.[1]
Percentages as reported by Porter and Nohria; the remainder reflects other activity categories the study didn’t break out individually. Source: “How CEOs Manage Time,” Harvard Business Review, July-August 2018. [1]
Most of that time is spent in a room with other people. The same study found CEOs sit in an average of 37 meetings a week, consuming 72% of their total work time.[1] That’s the number worth sitting with before you open any AI tool: the biggest single claim on an executive’s week is a form of work that AI can prepare for but can’t attend on your behalf. Nobody wants a chatbot representing them in the room.
What AI can genuinely take is narrower, and it clusters at the edges of that data rather than the center. Routine responsibilities are the obvious candidate, and a meaningful slice of “reactive” time is also fair game, because a lot of what looks reactive is actually information-gathering before you can respond, not the response itself. What it can’t do is sit in the 43% of time that’s genuinely agenda-driven, the moments where your presence, your read of the room, or your name on the decision is the actual point.
A short list of what tends to qualify, before you get to any specific tool:
Notice that none of those items are “the meeting” or “the decision.” They’re everything that happens around those two things, which is also where most of an executive’s actual hours quietly disappear.
Once you accept that only part of your week is a candidate for AI at all, the next question is which part gets which kind of help, because “use AI more” isn’t a plan. In practice, a personal AI stack has two modes, and almost every confused, half-adopted habit an executive develops comes from running a task through the wrong one.
Thinking partner mode is for judgment calls: the pricing decision, the hard conversation you’re about to have, the strategic bet you’re weighing. Here, you’re doing the thinking. AI’s job is to test it: argue the other side, list what you might be missing, push back on the assumption you haven’t examined closely. You stay in the loop the whole time, and the output isn’t a deliverable, it’s a sharper version of your own reasoning.
Delegate mode is for bounded, defined work: a first-pass draft, a prep document, a status summary pulled from scattered updates. Here, AI does the work and you review it before it goes anywhere. The task has a clear shape and a clear finish line, and your judgment shows up at the review step, not throughout the process.
| Mode | What it looks like | Example task | Where it goes wrong if you skip the review |
|---|---|---|---|
| Thinking partner | A back-and-forth conversation. You bring the situation, AI argues, questions, and stress-tests, you keep deciding as you go. | Pressure-testing whether to renew a major vendor contract before the call with your CFO | You start treating its counter-argument as the answer instead of a prompt to think harder |
| Delegate | A defined task with a clear output. AI produces a first pass, you check it, you decide what ships. | Turning six weeks of scattered project updates into one prep document before a steering meeting | You skip the check because the draft reads as finished, and a wrong number goes out under your name |
The two modes cover the whole personal stack. Almost every specific task an executive brings to AI is a version of one of these two.
Match the mode to the kind of thinking the task needs, not to how busy you are that day.
The mistake worth watching for is running a thinking-partner task through delegate mode because it’s faster: asking AI to just tell you what to do about a comp negotiation instead of using it to pressure-test your own read of the situation. Before handing something over, a short self-check helps.
| Question | Score 1 point if yes |
|---|---|
| Does this task have a defined, checkable output, rather than an open judgment call? | Delegate is a fit |
| Would you be comfortable if a colleague did the first draft and you reviewed it? | Delegate is a fit |
| Does getting this wrong mean re-explaining a fact, not re-litigating a decision? | Delegate is a fit |
| Is your name, not the document’s polish, the thing actually being tested here? | Thinking partner is the fit, not delegate |
Three or four “delegate is a fit” answers: hand it over and review the output. If the fourth question is the one that stands out, treat it as thinking-partner work instead, no matter how tempting the shortcut looks.
This is where the line matters most, and it’s worth stating plainly rather than as a footnote: AI can prepare the material you use to decide. It cannot make the decision, and it doesn’t carry the accountability for it. You do, whether the call goes well or badly. That split is the actual argument of this section, not a caveat bolted onto the end of it.
Say a CFO is deciding whether to walk away from a vendor renewal that’s crept 40% over its original contract value. In thinking-partner mode, she lays out the situation and asks AI to argue for renewing anyway, list the switching costs she might be underweighting, and flag anything in her reasoning that sounds more like sunk-cost thinking than a clean read of the numbers. That conversation sharpens her case. It doesn’t replace it. She still makes the call, and she’s still the one who has to defend it to the board, a use case close to the kind of prep work covered in our guide to using AI ahead of board meetings, though the underlying skill here, testing your own reasoning before you commit to it, applies well beyond anything board-related.
What counts as a judgment call differs by role, and it’s worth naming that difference rather than talking about “executive decisions” as one undifferentiated thing. A CFO’s hardest calls are usually numeric: which variance is a real problem versus noise, which number to trust. A CMO’s are more often about creative or brand judgment, where there’s no spreadsheet that settles the argument. A COO’s tend to be operational trade-offs under time pressure. AI can pressure-test all three. It can’t supply the taste, risk tolerance, or organizational context that actually makes the call, because none of that lives in the prompt.
| What AI does | What you still own | How it gets checked |
|---|---|---|
| Lists counter-arguments and blind spots in your reasoning before a hard call | Deciding which counter-argument actually changes your mind versus which one you’re rejecting on purpose | You can name, out loud, the reason you rejected each point it raised |
| Pulls together the numbers and context behind a decision into one readable brief | Deciding what the brief leaves out that actually matters | You read the source data yourself before the number goes anywhere near a board or a sponsor |
| Drafts the message for a hard conversation, tone and structure included | Deciding whether to have that conversation live, and what you’ll say if it doesn’t go as drafted | You never read a delegated draft aloud for the first time in the actual conversation |
AI can prepare the decision. It never gets to own it, and neither does the fact that a tool suggested it.
The failure mode to watch for isn’t dramatic. It’s quiet: an executive starts treating a well-written summary as though it settles the question, because it reads as confident as a correct one would. A wrong recommendation doesn’t announce itself any differently than a right one. That’s exactly why the check has to happen before the decision, not after something goes wrong, and it’s the same discipline covered in more depth in our anti-hallucination toolkit, worth a read if this verification habit is new to you.
Some of what an executive does can’t be handed to AI, and the honest reason usually isn’t a capability gap. It’s that your presence is the actual content of the moment, not the words being said. Treating that as a limitation of the tool misses the point. It’s a deliberate choice about what your role still requires from you specifically.
If your presence is the message, delegating the words defeats the point of showing up at all.
This isn’t an argument for using AI less. It’s an argument for being specific about the boundary instead of leaving it vague. An executive who can name exactly what stays theirs tends to delegate more confidently everywhere else, not less, because the line is clear instead of a source of quiet guilt every time a draft goes out unread.
None of this is useful as an abstract split between two modes unless it shows up in an actual week. Here’s what it looks like for an executive who’s had the habit running for a couple of months, not someone trying it for the first time.
| Day | Delegate task | Thinking partner task |
|---|---|---|
| Monday | Pull last week’s scattered project updates into one prep document before the leadership meeting | — |
| Tuesday | First-pass draft of a policy memo, based on three bullet points and a stated goal | Pressure-test the reasoning behind a pricing change before raising it with the team |
| Wednesday | — | Talk through a difficult conversation before having it, including what you’ll say if it goes sideways |
| Thursday | Summarize a long vendor proposal into the three points that actually matter for the decision | — |
| Friday | — | Review the week’s open decisions and check your own reasoning on the one you’re least sure about |
Roughly five to seven uses across a normal week, split close to evenly between the two modes. Fewer than that most weeks usually means the habit hasn’t stuck yet; far more, especially skewed toward delegate, is worth a second look at what’s being handed over.
The real test of whether this is working isn’t how often you used AI this week, but whether you’re still reaching for it in six weeks without being reminded, and whether the actual output, the decision, the memo, the prep document, is genuinely better or faster because of it. Using AI eleven times this week and not being able to say what changed as a result is activity dressed up as a habit, not the real thing.
Pick one delegate task and one thinking-partner task you’d normally do the hard way this week. Define each one narrowly: the exact document, the exact decision. Run them for two weeks. Then decide, honestly, whether either one earns a permanent spot in how you work, or whether it was just easier to try than to skip.
Focus on your own recurring work, not the company’s rollout: your own prep, your own first drafts, and thinking-partner conversations before decisions only you can make. Leading adoption is a different job with a different playbook, and mixing the two is exactly why some executives champion AI everywhere in the company while barely touching it in their own week.
Pick something you already do every week rather than searching for a novel use case. A status pull, a first-draft prep document, or turning scattered updates into one readable summary are all good starting points, because you already know what a good version looks like and can judge the output fast. Bounded, defined, low-stakes work is the right first test, not a decision you’re nervous about.
It can genuinely help with decisions, but only at the preparation stage, not the decision itself. For something like a hiring call or a comp negotiation, AI can lay out the case for and against, flag what you might be underweighting, and organize the evidence. It has no stake in the outcome and no accountability for it, so the actual call, and the reasoning behind it, has to stay yours.
Anything where your presence is itself the message: delivering hard news in person, the final call on a people decision, and anything carrying your name publicly, especially during a crisis. These aren’t capability gaps. They’re moments where the fact that you specifically showed up, in person or in your own words, is the actual point, and handing over the words undercuts why you were there.
Scale and audience. Personal use is about your own output and your own calendar: what you draft, what you prep, what you think through before a call. An adoption strategy is about governance, licensing, training, and change management across a team or company, aimed at other people’s calendars, not yours. An executive can be excellent at one and have barely started on the other.
This article draws its one statistical source, the Harvard Business School time-use study of 27 CEOs, from Porter and Nohria’s ‘How CEOs Manage Time’ (Harvard Business Review, July-August 2018), verified against a reprint published with Harvard Business Review’s direct permission. The two data breakdowns used, meeting time and activity-purpose split, come from the same study but are different measurements and are never combined into a single total. All named executive scenes (the operations VP, the CFO weighing a vendor renewal) are illustrations of a common situation, not real client stories.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.