25 AI Rules for CEOs

RULE 1. Start with your business strategy.

“Great CEOs feel like they’re losing control over their business because they define their strategy based on technology, asking ‘What is our AI strategy?’ Instead, start off with: ‘What is your business strategy — and what are the parts that AI can accelerate?’”

“Then come to every single vendor and technology player and say: This is the direction I’m headed. How does AI help me get there better, faster, or cheaper? But I’m going to be the one that’s controlling where that differentiation lies. I’m going to be controlling how we compete. I’m going to force these vendors and these companies now to speak my language, so that I can lead my organization, as opposed to adapting to theirs.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

RULE 2. Determine your risk appetite.

“Ultimately, CEOs have to set the agenda about the risk appetite that the company is going to encourage in their AI explorations, and that is something everybody else will optimize around.”

— Saurabh Sharma, You.com CPO

RULE 3. Decide: Centralized or federated strategy.

“The centralized strategy is great — you can move with speed, you can move with clarity from the top — but it can also be far away from the troops on the ground and how they might use it.”

“With the federated strategy, you’re going to see a lot more innovation bloom, but it can be chaotic, it can be duplicative.”

— Saurabh Sharma, You.com CPO

RULE 4. Decide: Incremental or skunkworks.

“There are two approaches that are really important to think through here. One is the incremental approach — you have an existing business unit…and you’re going to incrementally improve that product line with AI capabilities, AI features. That is a good route to continue to invest in.”

“A lot of the companies that I talk to are now pursuing a second route, which is a skunkworks route — saying that maybe there’s something completely different here that we need to think about from first principles, about rebuilding our product to be AI-native completely. The risk is that it’s a high-risk endeavor. Nothing may come out of it. But the flip side is that if something does come out of it, it may be a 10x differentiator that you can utilize and have defended your business against somebody else coming in and finding that 10x advantage.”

— Saurabh Sharma, You.com CPO

RULE 5. Build to where model capabilities will go.

“I get asked all the time: I’m a startup, I’m building, how do I not get disintermediated? How do I not get eclipsed by the offerings of the core foundation model labs? … You have to build with the expectation that whatever the value prop is that you’re providing to your end customers, it will get better as foundation models get better.”

“I know where we sit right now, and I know the next two quarters are going to see tool and model capabilities continue to get better and better and expand at even a faster rate than they have so far this year.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 6. Track revenue per employee.

“That is the leading indicator I look at for whether there is an actual acceleration in business outcomes, as opposed to features and products being released into the wild today.”

— Saurabh Sharma, You.com CPO

RULE 7. Embrace controlled chaos.

“We’re entering a period in my mind where the companies that are succeeding in their AI strategy are … letting flowers bloom within their organization.”

“In my view, controlled chaos is the end goal you should all be building inside your organizations, so that the unexpected — which somebody may be getting funding for right now — can occur within your organizations as well.”

— Saurabh Sharma, You.com CPO

RULE 8. Iterate.

“I agree with a lot of the controlled chaos sentiment, and I would use my own words — iterative deployment — to describe the way I think about it and the way OpenAI thinks about it.”

“If you can get the velocity of building to 5x what it was before using a lot of the AI tools, and then have five or six different surfaces where you want to start to test the application of AI, and then continue to iterate … I think you’re going to embark on a very good strategy of how best to incorporate AI into your business.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 9. Decide the culture you want: one where your employees feel like they can be replaced, or one where everyone learns AI and reinvests together.

“Pretend for a moment: You're an employee, and one day a guy walks in and hands you a magic wand. Wave it over your keyboard, and what about the 40 hours you used to grind through? Done in 20. Sounds like a gift. Except the moment your company finds out you have it, they might decide to fire the person next to you and dump their work on your desk. You're right back to your forty hours. The company pockets the difference. So ask yourself, honestly: are you telling anyone the wand works? Are you eager to share your adoption lessons?”

“That's an important part of the AI conversation. The pitch to workers is genuinely thrilling: more done, less time. But somebody in charge needs to ask the uncomfortable question: when that extra productivity appears, whose pocket does it land in? If ‘successful adoption’ means smaller teams and fewer rungs on the ladder, why would anyone on the ground floor work to make it happen?”

“You can't ask people to build the thing that hollows out their own future and expect them to be enthusiastic about it. The real work for anyone calling the shots isn't the technology. It's figuring out how to share what AI gives and what it takes — fairly, openly — so the people doing the actual lifting have a reason to get on-board.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

RULE 10. Do not use the word “replace” in the same sentence as AI at the early stage in startups.

“I think the word should instead be acceleration or augmentation of existing skills. I look at the early startups that I work with — there might be a five- or six-person engineering team, but all five of those engineers are getting 5x outputs because they use all the tools that AI enables. They’re not replacing people because of the tools; they’re adding people because they’re so productive with the tools. That’s how these companies are starting to build from the ground up.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 11. Staff AI with internal talent first.

“It’s less about an ‘AI-first’ mentality … but who understands a problem, who understands what the future could look like, and empower those people ideally.”

— Saurabh Sharma, You.com CPO

RULE 12. Don’t pre-select talent for domain expertise.

“I would select the people that have that exploratory mindset. Because otherwise, you can self-select domain expertise, but maybe your opportunity is in a new audience segment you haven’t hit.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

RULE 13. Blow up skunkworks teams after two years.

“Only because then politics encroach. They start protecting their space, their budgets, their promotion paths.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

RULE 14. Refine management structures.

“What you’re starting to see from companies, from Google on downwards, is that they are collapsing their management layers — the productivity that you needed from a hundred engineers with two directors and one VP over them now can be achieved in some cases with 50 engineers, or 40, or 30, or 20. We’re still finding what that lower bound is, and that’s collapsing management structures.”

— Saurabh Sharma, You.com CPO

RULE 15. Build a thick layer. 

“The most important thing about that thick layer is as the models get better, your offering to your end customers gets better.”

“If as the models get better, your customers are saying, ‘the models are getting better, how come yours isn’t getting better,’ it would be because your solution is not getting better as the underlying models get better.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 16. Own at least two of the four ingredients of defensibility.

“Number one: the vertical workflow — can you go deep into that workflow?”

“Number two: are you the system of record for your customers and whatever data it is that you manage for them?”

“Number three: do you have unique distribution or unique partnerships that is an advantage?”

"And the fourth one is: can you build around network effects? As you gain more customers, do you have more bespoke IP, do you have a data network effect, do you have a human network effect that creates a moat around what you’re building?”

“Those are businesses that if you plug in GPT-5.9, your customers are not going to go back to saying, I want it to be cheaper now, or I want to do it myself, because the combination of those efforts mean that you are much harder to displace, certainly in a DIY manner, but also from a competitive perspective.”

— Saurabh Sharma, You.com CPO

RULE 17. Build evaluations for every place AI touches your product.

“As long as you have evals and evals become a proprietary component of what you’re building, then I think you’re in a good place, because you never become dependent on any one particular model. You’re only going to be dependent on whichever model is best for your specific use case.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 18. Control your narrative inside the LLMs.

“It’s a whole new world because the models interpret the content differently.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 19. Know when to buy.

“Companies that are leading with the research capability, staying at the frontier because of that research capability, getting a lot of traction, seeing a lot of success with new customers — ultimately, I really believe that many of those customers are just going to buy — and boom, [they] have a super intelligence lab.” 

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 20. AI budgets belong with P&L owners — not G&A or R&D.

“At the end of the day, it’s going to be: are you able to get high-quality work done at a more efficient cost than you were able to before? And that’s a P&L discussion.”

— Saurabh Sharma, You.com CPO

RULE 21. Track cost of intelligence.

“Every time we release a new model, it is the most expensive that model is ever going to be. Everything that we do after we release a new model is to continue to reduce the cost per token until it gets to a point that everybody has the ability and the access to do it.”

— Colin Evans, OpenAI Startups & VC Partnerships GTM

RULE 22. Get great at forecasting.

“Some of the best businesses I work with are forecasting where they look at token prices, storage costs, processor speed, and they’re building in advance of that.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

RULE 23. Make sure your CISO is your best friend.

“In many ways if they’re not, they’re the ones that are going to slow down the organization, because it is their job to do it.”

— Saurabh Sharma, You.com CPO

RULE 24. Refine learnings from inside the organization.

“Usually the people with the least amount of power in organizations will be the most honest with everything they have to go through to get something done.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

RULE 25. Accept tension.

“You have the buttoned-up corporate types that run on efficiency and careful risk, and the entrepreneurial types who'd rather move fast, break something and see what happens. The trick is not to pick a side.”

“Leaders are trying to manage from the middle, drawing out the best from each person. That's the work. And if you're doing it right, both camps will be a little unhappy with you some of the time. That's not failure, that's the job.”

— Neil Hoyne, Wharton Executive Education faculty and Google Chief Strategist

Through this newsletter, the podcast, and live events, Macro Talk uncovers the forces shaping the economy and world of work — and how to play them.

Thank you to Lia Hanson, Kate Dishaw, and the entire Norwest team for making this conversation possible. A big thank you also to Colin, Neil, and Saurabh for your candid and sharp insights.