Justin McKelvey

Justin McKelvey

Fractional CTO · 15 years, 50+ products shipped

AI for Business 7 min read

AI for CEOs: What to Do Yourself, What to Delegate, and What to Refuse (2026)

Quick Answer

How should a CEO use AI? Personally, every week, on the work only you do: board and investor prep, strategy memos, hard conversations, reading what the company is telling you, and the first draft of anything with your name on it. Use Claude or ChatGPT directly, in a project with your context loaded. Delegate the three things that scale. Refuse the three that don't reverse. Write a one-page policy so nobody has to ask. As of 2026, that's the whole job — and most CEOs are doing the opposite: buying licenses and never opening the tab.

Verified September 2026 · Author: Justin McKelvey, fractional CTO & AI consultant, 15 years in software, 50+ products shipped

TL;DR: The CEO's Job With AI Is Not to Become the Best Prompter in the Building

A CEO should use AI on the six jobs only the CEO does, delegate the three that scale, refuse the three that don't reverse, and publish a one-page policy so everyone else can move. That's it. Not a transformation program. Not a Chief AI Officer search. Not a tool budget with a Slack channel. I've sat in enough leadership meetings in 2026 to know the pattern: the company has licenses, the all-hands had a demo, and the CEO has never personally used the thing on a real decision. The routine below is what I run for myself and install for the founders I work with. It takes about three hours a week and it is the highest-return three hours on your calendar.

The six things a CEO should personally do with AI every week

Open Claude or ChatGPT, in a project or workspace with your company context loaded — strategy doc, org chart, last quarter's numbers, your current priorities. Then, weekly:

  1. Pre-read everything before the meeting. Board decks, investor updates, the exec team's status docs. Paste it in and ask: what's missing, what's inconsistent with last month, and what question would a skeptical board member ask first. You walk in having already had the argument.
  2. Draft and stress-test strategy memos. Write the memo yourself, then ask the model to argue against it as your sharpest competitor, your CFO, and a customer who just churned. The memo that survives three attacks is the one you send.
  3. Prepare for hard conversations. A performance issue, a co-founder disagreement, a customer escalation. Role-play it. Ask for the three ways the other person might hear what you're planning to say. This is the single use that CEOs tell me changed their week the most.
  4. Read what the company is telling you. Support tickets, sales call notes, exit interviews, NPS comments. Dump a month of it in and ask for the five themes and the one thing nobody has escalated. Your team filters bad news on the way up; the model doesn't.
  5. Write the first draft of anything with your name on it. The all-hands, the investor letter, the LinkedIn post, the customer apology. First draft only. The voice is yours to fix; the blank page is no longer your problem.
  6. Interrogate your own numbers. Paste the dashboard export and ask what changed, what's correlated, and what you'd want to know before you'd bet on the trend. It's not a replacement for your finance lead. It's the question list you bring them.

Notice what's not on the list: coding, marketing automation, customer service bots. Those are real, and they're someone else's job. The CEO's AI use is judgment amplification, not throughput.

The three things to delegate

  • Repeatable knowledge work with a clear template. Proposals, job descriptions, SOPs, meeting summaries, first-pass research. Give a function owner a budget and a target, not a mandate to "explore AI."
  • Customer-facing automation. Intake, triage, FAQ handling, follow-up sequences. Owned by whoever owns the customer, with a human review gate until the error rate is measured, not assumed.
  • Tool selection and integration. Which assistant, which connectors, which data it can see. That's a fractional CTO or an AI consultant job, and the CEO's only role is to approve the one-page policy it has to fit inside. If you're a smaller company, the Claude for small business setup guide is the version of this you can do in an afternoon.

The three things to refuse

  • Irreversible decisions. Hiring, firing, pricing, fundraising terms, the big contract. AI produces the options and the arguments. It never chooses. The day you let a model pick the candidate is the day you can't explain the decision to anyone, including yourself.
  • Speaking for you when the stakes are personal. The termination conversation, the board answer under pressure, the public statement in a crisis. Draft with it, sure. Deliver it yourself, in your words, or the room will know.
  • The strategy itself. Models are extraordinary at synthesizing what everyone already believes. Strategy is the bet on what they don't. Use AI to find the consensus so you know exactly what you're deviating from.

Write the one-page AI policy this week

Every company I've walked into without a policy has one of two cultures: reckless use (customer data pasted into personal accounts) or no use (everyone assumes it's banned). Both are your fault, and the fix is one page with five sections:

  1. Approved tools and accounts. Which assistants, on company accounts, with which data controls on.
  2. Never-paste list. Customer PII, credentials, unreleased financials, anything under NDA, anything legal would wince at.
  3. Human-review gates. What must be read by a person before it reaches a customer, a regulator, or the public.
  4. Owners by function. One name per department who decides what AI does there. Not a committee.
  5. How to ask for an exception. A form, a Slack message, a 24-hour answer. Permission, not restriction.

Sign it. Put your name at the top. The policy is the thing that turns "the CEO said AI is a priority" into something a mid-level manager can actually act on. AI for managers is the companion piece for the people who have to run it day to day.

Adoption vs. adoption theater: how to tell

Theater looks like this: a tool budget, a #ai-tips channel, a vendor demo at the all-hands, an "AI champion" in each department, and no metric that moved. Real adoption looks like shorter cycle times on named workflows, the same throughput with fewer hours, and people who can tell you, specifically, the three tasks they now do differently.

The fastest test I know takes ten minutes. Ask five people, individually, what they did with AI yesterday. If you get five specific answers, you have adoption. If you get "we're exploring it," you have licenses. As of 2026 most companies I meet are in the second group and reporting themselves as the first.

AI for founders: the 5-to-50-person version

If you're a founder-CEO of a company with five to fifty people, everything above applies and one thing changes: you don't have the functions to delegate to yet. You are the head of sales, the head of product, the head of people, and the finance department, in whatever order the week demands. That's exactly why AI matters more to a founder than to any Fortune 500 CEO. The jobs you do alone before you can afford to hire for them are the jobs a frontier model is best at drafting.

Here's what my founders' daily AI use actually looks like, as of 2026:

  • Customer research: transcripts in, patterns out, before you write a line of product spec.
  • Sales: call notes to follow-up email in two minutes, proposal from your last three winning ones.
  • Product: the spec's first draft, the edge cases you didn't think of, the test plan.
  • People: the job description, the scorecard, the interview questions that actually test for the thing.
  • Finance: the model you'd otherwise build in a spreadsheet at midnight, with the assumptions listed so you can argue with them.
  • Marketing: the first draft of everything, in your voice once you've loaded twenty of your own posts as examples.

The founder-specific rule is the inverse of the big-company one. A large-company CEO's risk is never touching the tool. A founder's risk is letting it touch everything. Keep your hands on two things personally: the customer conversations and the numbers. Let AI draft around them. The day you stop reading the support inbox because the model summarizes it, you've traded your best signal for a convenience. If you want the wider view of who owns what at this size, I wrote it up for AI for executives and, for the people-leadership side, AI for leaders.

What changes above 50 people

Past fifty, the CEO's personal AI use shifts from doing the work to modeling it. Your team copies what you visibly do. If they see you paste the board deck into Claude before the meeting and walk in with better questions, they will do it with their own decks. If they hear you talk about AI and never see you use it, they'll correctly conclude it's a talking point. At this size, the policy matters more, the function owners matter more, and the theater test matters most — because you're now far enough from the work that you'll be the last to know it isn't happening.

The micro-action for today

Take whatever's on your calendar tomorrow that you're least prepared for. Paste the relevant material into Claude or ChatGPT and ask for the three questions you'll wish you'd asked. That's the entire onboarding. If you want to know whether your company is ready for more than that, the AI readiness checklist takes ten minutes and tells you which of the five policy sections you're missing. And if you'd rather talk it through, the 30-minute strategy call is free and has no pitch in it — most CEOs use it to figure out whether the problem is the tools or the org.

Free Resource Justin McKelvey

Get the Free AI Content Toolkit

The exact system I use to turn one idea into a month of content — atomization framework, voice template, prompt library, weekly system.

Frequently Asked Questions

How should a CEO use AI?
Personally, on the work only the CEO does: pre-reading board and investor material, drafting and stress-testing strategy memos, preparing for hard conversations, summarizing what the company is telling you in metrics and customer feedback, and writing the first draft of anything with your name on it. Use a frontier assistant like Claude or ChatGPT directly, in a project with your context loaded, and treat it as a chief of staff that never sleeps and never has an agenda. Then delegate the repeatable work, refuse the irreversible decisions, and publish a one-page policy so the rest of the company can move without asking permission.
How should founders use AI?
As a force multiplier for the jobs a founder does alone before the company can afford to hire for them: customer research, first-draft copy, sales follow-ups, financial modeling, hiring scorecards, and product specs. As of 2026 a founder running a 5-50 person company should be spending real personal time in Claude or ChatGPT every day, because the alternative is doing that work slower by hand or paying someone before the business can support it. The founder-specific rule is to keep your hands on the customer conversations and the numbers, and let AI do the drafting around them.
What should a CEO never delegate to AI?
Three things: decisions that are expensive to reverse (hiring, firing, pricing, fundraising terms, major contracts), anything that carries your personal judgment to someone who will act on it (a termination conversation, a board answer, a public statement in a crisis), and the company's actual strategy. AI is excellent at producing options and arguments for all three. It should never be the thing that chooses, and it should never be the thing that speaks for you when the stakes are personal.
How do I know if my company is actually adopting AI?
Look at outputs, not logins. Real adoption shows up as shorter cycle times on specific workflows, fewer people needed for the same throughput, and staff who can name the three tasks they now do with AI and what changed. Adoption theater shows up as a tool budget, a Slack channel, a demo at the all-hands, and no metric that moved. The fastest test I know: ask five people what they did with AI yesterday. If the answers are vague, you have licenses, not adoption.
What should be in a CEO's AI policy?
One page, five sections: what tools are approved and on which accounts, what data may never be pasted into them (customer PII, credentials, unreleased financials, anything under NDA), what work must have a human review before it leaves the building, who owns AI decisions in each function, and how to request an exception. The goal is permission, not restriction. A company without a policy defaults to either reckless use or none, and both are worse than a page that says what's fine.
Should a CEO learn to use AI personally or hire someone?
Both, in that order. A CEO who has not personally used Claude or ChatGPT on real work for at least a month cannot evaluate an AI hire, a vendor, or a consultant, and will get sold something. Spend the month first. Then, if you can name three workflows where AI would change the outcome this quarter, hire for implementation. If you can't name them, that's the gap an AI readiness assessment closes, not a hire.
How should a small business owner use AI?
The same way as a founder-CEO, with the budget dialed to zero until it pays for itself: one paid assistant seat, one project with your business context loaded, and a short list of tasks you do every week that you now start in AI instead of from a blank page. Customer replies, quotes and proposals, job descriptions, marketing drafts, and reading your own numbers. Small business owners get more out of AI per dollar than any enterprise I've worked with, because the owner does the work and sees the time come back directly.

More on AI for Business

AI for Executives: The Five Decisions That Are Yours, and the Twenty That Aren't (2026)

Executives don't need to learn AI. They need to make five decisions about it: what to fund, what to refuse, who owns it, what the board gets told, and what 'we use AI' means in the P&L. Everything else belongs to the people who run the workflows. A fractional CTO's version of the executive AI briefing, without the slide deck.

8 min

AI for Leaders: Your Job Is the Decisions, Not the Prompts (2026)

Leaders keep asking which AI tool to learn. Wrong question. Your job is deciding what the company uses AI for, what it never uses it for, how work gets re-scoped when a task drops from days to minutes, and how you judge output you can no longer produce yourself. The decision rules, the meeting habits, and what to stop measuring.

7 min

AI for Managers: How to Actually Use It, What to Hand Your Team, and What to Refuse (2026)

A manager's AI job is not prompting. It's deciding which of your team's tasks AI takes, setting the review standard for what comes back, and re-scoping people's work when a two-day task becomes a twenty-minute one. Here is the weekly practice, the tools that matter, and the mistakes I watch managers make.

7 min

Anthropic API Pricing (2026): Every Claude Model Per Million Tokens, and What a Real Workload Costs

As of September 2026: Sonnet 5 $2 in / $10 out per million tokens, Opus 5 $5 / $25, Haiku 4.5 $1 / $5, cache hits at 10% of input, Batch API half price. The rate table is the easy part. What three real workloads cost, the tokenizer change that quietly inflates bills, and the price cut Anthropic made permanent.

7 min
Justin McKelvey, Fractional CTO and AI consultant in Austin, TX

Written by

Justin McKelvey

Fractional CTO & AI consultant in Austin, TX. 15 years building software, 50+ products shipped, $53M+ in client revenue generated. I help $1M–$50M founders ship production software and automate operations with AI — without hiring a full-time executive team.

Work with me

If this was useful, here are two ways I can help: