10 ways a CEO can put AI to work this week

Three of every four professionals already let an AI notetaker sit in their meetings. You probably pay for ChatGPT or Claude and use it the way you'd use a smarter search box. That gap, between what the tool can do and what you ask it for, is the whole opportunity. Here are ten specific jobs you can hand it in the next five business days.

Most "AI for CEOs" advice tells you to write a strategy. This is the opposite. No committee, no roadmap, no platform. Just a menu of ten high-leverage uses, ordered from the ones that cost you almost nothing to start (your meetings, your inbox, the decision on your desk right now) to the one with the highest ceiling (building a small piece of software by describing it).

Pick two or three. Run them this week. Notice which ones save you an hour and which ones waste ten minutes. That is your map. Everything below is real work a non-technical operator can start today, and each one links to the deeper walkthrough when you want it.

If you'd rather see the strategic case for where to begin and what to ignore, we wrote that pillar too. This piece assumes you've decided to act and want the concrete list.

1. Stop taking meeting notes

The job: let software capture every decision, owner, and deadline while you stay fully present in the room.

You sit in meetings all day and most of what's said evaporates by Thursday. You half-remember, you scribble a note you never reread, you forget who owes whom what. Three of every four professionals already use an AI notetaker for exactly this reason, and managers report losing four-plus hours a week just summarizing calls by hand. That's a full afternoon, gone, on a task a machine does better than you while you focus on the person across the table.

Granola is the one I run. It joins the call, transcribes it, and turns each meeting into clean notes. The first action is small: let a notetaker join your next call, then afterward ask it for one thing only, "give me the decisions, the owners, and the deadlines." That single prompt is the whole unlock. Once you trust the capture, you can ask your meeting history questions the way you'd ask a chief of staff who never forgets, and review your own meeting habits with a meeting-effectiveness pass. For the full case on why this is the first install, the meetings deep-dive is the place to go. Nothing falls through the cracks.

2. Turn your inbox into drafts, not reading

The job: have AI pre-draft your replies in your voice and sort the pile by what actually needs you.

Email is the tax you pay for running a company. An AI email assistant reads the thread, drafts the reply the way you'd write it, and surfaces the ten messages that need a human before the forty that don't. These tools are built for executives clearing 100-plus messages a day, and as of spring 2026 agentic triage is shipping inside the tools you already open, Outlook among them, so this stops being a separate app you have to adopt.

The first action: connect one assistant and let it draft your next ten replies. Read the draft, fix the one line it got wrong, send. You're editing instead of composing, which is roughly a third of the keystrokes. You stay the judge of tone and commitment, which is the part you'd never want to hand off anyway.

3. Brainstorm the hard decision with a thinking partner

The job: paste the full context of the decision on your desk and make the model argue both sides, then write the memo.

One founder put it well on X: when he feels stuck on what to do next, he opens Claude or ChatGPT and brainstorms, "not for the AI to give me an answer, for me to find the right question." That's the move. A large language model, the engine inside these chat tools, won't make the call for you and shouldn't. But it will hold every variable at once, steelman the side you're resisting, and hand you a clean decision memo to react to.

The first action: take the live decision you've been circling, the hire, the price change, the partnership, and dump everything you know into the chat. Then ask it to argue both sides as hard as it can and write the one-page memo. You'll find your own answer faster by arguing with a draft than by staring at a blank page. If the decision is a recurring leadership pattern, you can even pull your own coaching themes out of your meetings and codify the doctrine you keep repeating so you stop relitigating the same calls.

4. Make every sales follow-up same-day

The job: turn your call notes into a personalized follow-up before you leave your desk.

The deals you lose aren't lost on the call. They're lost in the gap after it, when the rep means to follow up and the week runs them over. Speed is the whole game here: replying to a positive reply inside five minutes can lift your close rate by something like 8x versus letting it sit, and reps who automate the drafting get back two to five hours a week. That's not a productivity stat, that's revenue you're currently leaving on the table because the follow-up went out Thursday instead of Tuesday.

The first action: after your next sales call, have AI turn the call into a follow-up draft before you stand up. While you're at it, let it enrich the CRM record from the call so your pipeline data is real, and read the transcripts to surface what's slipping before your next pipeline review. You supply the judgment; the AI supplies the discipline your team never quite maintains.

5. Batch a week of marketing in one sitting

The job: feed the model your last few posts plus a topic and generate a week of content in your voice.

Small businesses that use AI overwhelmingly point it at content first, because content is high-frequency, low-stakes, and easy to judge. You already know your voice when you see it. The model just has to get close enough that you're editing, not writing from scratch, and many operators report cutting their content time sharply once they stop drafting cold.

The first action: paste your last three posts or emails into the chat, give it next week's topic, and ask for a week of captions or one newsletter in the same voice. Edit for the lines that don't sound like you. If you run meetings all day, the richer source is sitting right there: you can pull content ideas straight out of your meetings, and if you're writing a book or a long-form piece, mine your meetings for the ideas you already said out loud. Push this far enough and the whole content function changes shape: here's how I replaced a $10K/month agency with an AI stack.

6. Write the job description and screen the pile

The job: draft the JD from your must-haves, then have AI rank the inbound resumes against them.

Hiring is two slogs bolted together: writing the post and reading the pile. Roughly two-thirds of recruiters now draft job descriptions with AI, and the screening time drops sharply when you let the model do the first pass against an explicit rubric. The key word is explicit. You define the three must-haves; the model ranks against those, not against its own guess at what "good" means.

The first action: for your next open role, ask AI to draft the JD from a few bullets, then write a screening rubric that names your three non-negotiables. When resumes come in, have it rank them against the rubric and flag the top handful. You read ten instead of a hundred, and you're still the one who decides who gets the call.

7. Build your runway and board model in a spreadsheet

The job: build a 12-month runway model inside the spreadsheet you already use and scenario-test your next hire.

This one got dramatically easier as of May 2026, when ChatGPT inside spreadsheets went generally available across plans. You can now talk to your model the way you'd brief an analyst: "build me a 12-month runway from these actuals," "show me what happens if I hire two engineers in Q3," "write the variance commentary for the board." Board-ready commentary that used to eat hours now takes minutes, and it lives right next to the numbers instead of in a separate deck.

The first action: open last quarter's actuals and ask AI to build the runway model with you, in the spreadsheet. Then scenario-test one real decision you're weighing. You verify the formulas, because you're the one who knows the business; the model does the assembly you'd otherwise pay someone to do.

8. Digest the pre-read before any meeting

The job: paste the deck, contract, or report into the model and ask for the five things that matter and the three risks.

The reason you walk into meetings underprepared isn't laziness, it's volume. Nobody reads the forty-page pre-read. In 2026 the context windows on frontier models hold a whole contract or a month of transcripts at once, which means you can hand the model the entire document and get a real digest, not a skim.

The first action: before your next board meeting, paste the deck in and ask for "the five things that actually matter and the three risks I should push on." For recurring meetings, formalize it as a pre-meeting brief that reads everything you've ever discussed with that person or that account before you walk in. You show up as the most prepared person in the room, on documents you'd otherwise have skimmed in the elevator.

9. Turn tribal knowledge into an SOP

The job: record yourself doing a task once, then have AI write the step-by-step SOP from the transcript.

Every company runs on processes that live in exactly one person's head. When that person is on vacation or quits, the process goes with them. You've been meaning to document it for two years. The reason you haven't is that writing SOPs is miserable and you're busy. So do the part you don't mind, the doing, and let AI do the part you hate, the writing.

The first action: pick one undocumented process this week, a vendor onboarding, a monthly close step, a refund decision tree. Record yourself doing it once with Loom or any screen recorder, or just narrate it on a call your notetaker is in. Hand the transcript to AI and ask for a clean step-by-step SOP. You'll have in twenty minutes what's been on your list for two years. The same instinct extends to your team's follow-through: AI can pull the to-dos out of every meeting and keep a running ledger of who committed to what, and turn the week's meetings into team and investor updates that write themselves.

10. Build one tiny internal tool

The job: describe a small admin dashboard, calculator, or intake form to an AI builder and ship a working v1 with no engineers.

This is the highest-ceiling item, which is why it's last. For twenty years, a small internal tool meant a budget line, a hire, or a six-week agency quote, so the little tools never got built. Now non-technical founders ship working apps in hours by describing what they want, and some of those apps replace SaaS subscriptions outright. The practice has a name, vibe coding: you describe the outcome and judge the result, and you never touch the code.

The first action: name one manual spreadsheet or repetitive task you'd love to kill, a deal-margin calculator, a clean intake form, a dashboard for the one number you check every morning, and describe its replacement to an AI builder. The full walkthrough, from never having opened a terminal to a tool you can open in your browser, is in build your first software tool. The one safety rule: vibe-code the internal, disposable, verifiable stuff freely, and the moment a tool touches customer data, payments, or the open internet, bring in someone who can read the code.

Inside that boundary, you've removed the budget meeting from the path between "I wish we had a tool that does X" and having it. As one founder put it on X: the trick isn't the time you save today, it's that next month future-you runs it fifty times for free. Leverage compounds.

What to do this week

You're not going to do all ten. You shouldn't try. Trying to start everywhere is how CEOs start nowhere.

Pick two or three from the top of the list, the ones with the lowest activation energy, and run them before Friday:

• Let a notetaker join your next call and ask it for decisions, owners, and deadlines.
• Connect an email assistant and let it draft your next ten replies.
• Take the decision on your desk and make the model argue both sides.

Those three cost you nothing but the five minutes it takes to start, and they touch every day of your week. The win you feel from one of them will fund your appetite for the next. When you're ready to graduate from chatting to delegating, the move is into Claude Code and the terminal, where AI stops suggesting and starts doing the work: that's where you'd build a skill it can reuse, a CLAUDE.md file so it knows your business, and routines that run on a schedule without you in the loop.

Want to maximize your AI leverage? Upgrade to Pro.

Tell me in thirty days which one you started with and what it gave back. Those are my favorite messages to get.

Andrew

Related reading

• As a CEO, where should we start with AI? · the strategy view, what to start with and what to ignore
• Granola for CEOs: the highest-ROI AI install of 2026 · the meetings install, in depth
• Which AI releases actually matter · how to filter the noise so you only act on what counts
• What is an AI agent? · the difference between chatting and delegating