What Dan Shapiro wrote about coders, I'm writing about CEOs. The deflation is coming for the corner office too.
Dan Shapiro recently wrote a sharp piece about the five levels of AI automation for software engineers. His key insight: there's a spectrum from "AI helps me type faster" to "AI writes the software while I sleep," and most people are stuck at level two feeling like they're already done. They're not.
I kept thinking: why stop at coders?
Harvard Business School professors Michael Porter and Nitin Nohria spent over a decade tracking 27 CEOs — every 15-minute increment, 24/7, for three months, collecting nearly 60,000 hours of data.1 What they found is clarifying: 25% of CEO time goes to people and relationships, 25% to functional and business unit reviews, 21% to strategy, and 16% to organizational culture. They also found that CEOs spend 24% of their total time on email — often FYI chains that don't require them at all.
Now run that breakdown through an honest LLM audit. Which of those activities genuinely require a human in the room — and which are, at their core, synthesis, summarization, tone-matching, and structured reasoning under uncertainty? The answer is uncomfortable if you're a chief officer.
CEOs, like engineers, are moving through five levels of automation. Most are at level one and believe they're at four. Let's name the levels.
The CEO writes their own board deck. Dictates memos. Prepares for analyst calls by reading printouts from the CFO. Strategy is formed in offsites, captured on whiteboards, transcribed later by an EA. AI, if used at all, is treated like a slightly smarter Google — you search it for facts, you don't trust it with anything real.
Most Fortune 500 CEOs over sixty are here. They are not embarrassed by this. They have survived and thrived without AI. But in a deflationary world — where their competitors can operate with radically thinner management layers — this is the equivalent of hand-compiling your own bytecode.
The CEO has discovered that AI drafts a pretty good earnings call script. They use it for first drafts of all-hands speeches, shareholder letters, press statements. They paste in bullet points, the model returns paragraphs, they edit and send. Their EA uses it too — for scheduling summaries, travel briefings, agenda preparation.
This is where most "AI-forward" executives are in 2026. They feel 20% more productive. They are. But their fundamental job is unchanged: they still make all the decisions, hold all the relationships, and absorb all the ambiguity. The AI is a very fast intern who never sleeps and never pushes back.
This is where things start to get interesting — and where the real deflation begins to land.
The CEO no longer writes strategy documents. They argue them into existence with an AI. They describe the competitive landscape, upload the market research, feed in the board's concerns, and push back on the model's synthesis until something true emerges. They use it to stress-test decisions before the CFO meeting. They run "red team" sessions where the model argues the bear case for every acquisition target.
I know a CEO of a Series C company — about four hundred employees — who told me he canceled the annual strategy offsite last year. "I did it in four sessions with Claude over two weeks," he said. "I brought the output to the team. They improved it in two hours. We saved three days and $80,000 in hotel and consultant fees." He paused. "And honestly? The first draft was better."
The danger at level two is the same as Shapiro's level two for coders: it feels like you're done. You are not done.
At level three, the CEO's primary value-add is no longer generating direction. It's approving it.
This sounds dystopian. It is also, arguably, how great CEOs should have been operating all along — and why so many good ones fail at the operational details. A level-three CEO has AI-assisted systems handling: competitive analysis and synthesis (fed from live data), draft resource allocation recommendations (from P&L patterns), first-pass talent assessments (structured interview outputs processed and ranked), investor communications (routine updates, question-and-answer preparation, draft 8-Ks).
What's left? The CEO reviews the stack. They override. They catch the things the model can't catch: the real reason the VP of Sales is underperforming has nothing to do with the pipeline metrics the model sees. The acquisition target's CEO is a narcissist, and no LLM noticed it in the diligence docs.
This is the first level where you start to wonder: how many of these decisions actually require the CEO? And the honest answer is: fewer than before.
"We used to have four VPs prepare a 40-slide strategy review every quarter. Now an AI does the first thirty slides overnight and the VPs spend a day stress-testing it. I spend two hours making three calls. That's it." — CEO, mid-size logistics company (paraphrased from a private conversation)
At level four, the CEO has stopped thinking of themselves as a decision-maker and started thinking of themselves as a system designer.
Their job is to define the values, constraints, and incentive structures that the AI-augmented organization runs inside of. They write — obsessively, precisely — the principles by which decisions get made. They set the risk tolerance. They define what "alignment with strategy" means in terms concrete enough that a model can evaluate it. They are, in the language of Shapiro's piece, writing the spec.
The irony is that this is what management theorists have argued CEOs should do for decades. Most couldn't, because the operational drag of being a CEO is overwhelming — the 62.5-hour weeks, the 75% scheduled in advance, the email abyss. Strip away what AI can do, and suddenly the job that's left is the job that was always most important.
A handful of CEOs are here today. Mostly at smaller companies where the org chart is flat enough that you can actually see the whole system. The next five years will push mid-market companies here faster than anyone expects.
At level five, the question stops being "what does the CEO delegate to AI?" and becomes "what does a CEO do that a well-specified AI system cannot?"
This is not science fiction. It is already happening at the edges. There are small companies — under fifteen people — where the "executive layer" is a set of AI agents with clearly defined mandates: one synthesizes competitive intelligence and drafts strategic recommendations weekly; one monitors financial KPIs and flags anomalies; one manages investor relations communications end-to-end, escalating only when a question requires judgment outside its policy envelope. The humans in these companies are more like a board of directors than an operating team. They set policy. They handle exceptions. They fire the vendor when the model gets it wrong three times in a row.
Fanuc has a factory in Japan that runs lights-out — no workers, robots building robots in the dark. The boardless company is its management analog. The lights are off in the C-suite. The org runs on principles, not presence.
Is this realistic for a 50,000-person company? Not yet. Is it realistic for a 50-person one? It's already happening, whether you call it that or not.
Here is the uncomfortable punchline: most of what we call "executive leadership" is information processing dressed up in authority. Reviewing decks. Synthesizing reports. Deciding between two reasonable options using judgment that is, in practice, pattern-matching on past experience. LLMs are extraordinarily good at exactly that.
What remains — genuinely, irreducibly — is the stuff that requires being a person in a room with other persons. Trust that is relational, not informational. The founder who can look an employee in the eye during a crisis and make them believe the company will survive. The judgment call that lives entirely outside the training data. The ethical line drawn against short-term pressure because of values held in a body, not a weights file.
That's real. That matters. But it's not 62.5 hours a week.
Shapiro ends his piece by saying most engineers top out at level three — the safety driver — when the real gains are at four and five. I'd say most CEOs are at level one feeling like level four. The deflation is real. The levels are there for the taking. The question is whether you'd rather define your level five job now, or have someone else define it for you.
1 Porter, Michael E. and Nitin Nohria. "What Do CEOs Actually Do?" Harvard Business Review, July–August 2018. The study tracked 27 CEOs in 15-minute increments over 13 weeks, collecting ~60,000 hours of data.
2 The 24% email figure and the "FYI chain" observation are from the same HBR study, also covered in Inc. and CNBC reporting.
3 The Series C CEO anecdote is reconstructed from composite conversations. The logistics company quote reflects a paraphrase of a private conversation; the sentiment and specifics are representative, not verbatim.
4 The Fanuc dark factory reference comes from Shapiro's original piece, which cites Organize Dergisi's reporting on Fanuc's robot-staffed manufacturing facility.