![\"AICEOSkynet.jpg\"](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/01\/AICEOSkynet.jpg\" "\\"AICEOSkynet.jpg\\"\/")
<\/p>\nYour Next CEO (or Manager) Might Be an AI<\/p>\n
Or, if not an AI, your actual human boss will probably be augmented by one<\/p>\n
Here\u2019s something that might not surprise you: When the weather is nice, CEOs play golf instead of working.<\/p>\n
Ben-Rephael et al. (2024) actually quantified this in their paper on executive effort. Researchers obtained Bloomberg terminal usage data for C-suite executives at publicly traded firms and matched it with SEC filings. The findings? A one-standard-deviation increase in good weather corresponded to roughly 20 fewer hours of effort per quarter for CEOs. Effort also declined when annual targets became unreachable\u2014why work hard in the second half if it won\u2019t affect your bonus?<\/p>\n
Research also indicates that the acquisition of power reduces perspective-taking and empathy\u2014which may explain why the C-suite plays golf on Friday, while demanding longer work hours from everyone else.<\/p>\n
Human CEOs have issues. They may not care enough about society or their workers, but they\u2019re also not singularly focused on shareholder value. They\u2019ve got plenty of room left for status, leisure, reputation, and control. Hence the usual greatest hits of agency problems: Perks (yes, the corporate jet), the quiet life, headline-friendly philanthropy, and a little empire-building on the side.<\/p>\n
There\u2019s another problem with human CEOs: They have specific types and competencies that may not match what a firm needs. Dahlstrand et al. (2024) found that firms led by the \u201cwrong\u201d CEO type lose up to 20% of revenue per employee. Eliminating CEO mismatch across their sample would raise aggregate productivity by approximately 9%.<\/p>\n
By contrast, an AI \u201cCEO\u201d won\u2019t crave applause, corner offices, or a weekend place in Aspen. It doesn\u2019t want $1 billion per year in compensation and won\u2019t skip out to the golf course after a bad first half. The AI CEO will have available the sum of all human knowledge, real-time access to all data inside the company, and the ability to be present at multiple meetings simultaneously. And it can shift type (e.g., operations focused or visionary leader) on request. In the near future, it\u2019s likely to be able to decide – on its own – what type of CEO is needed for a given situation.<\/p>\n
Of course, there may be other downsides to consider.<\/p>\n
<\/p>\n
An AI CEO that\u2019s available 24\/7, calm, rational, and strategic sounds ideal\u2014unless, of course, it\u2019s biding its time while planning to eliminate humanity. (I\u2019m not being entirely flippant; for more details, see the Wikipedia article on existential risk from AI<\/a>.)<\/p>\n And the agency problem doesn\u2019t disappear\u2014it just moves upstream. Who programs the AI CEO\u2019s goals? What guardrails does it have? Who audits what it\u2019s doing? And the humans around it may treat it like a very expensive ventriloquist\u2019s dummy: \u201cDon\u2019t blame me\u2014the algorithm wanted the jet.\u201d<\/p>\n But aside from the above: How technologically feasible is it for an AI to do the work of a CEO?<\/p>\n Keep in mind that all the examples I\u2019m going to discuss are about older versions of Large Language Models (LLMs). As Mollick (2024) put it, \u00a0\u201cWhatever AI you are using right now is going to be the worst AI you will ever use.\u201d The capabilities will continue to improve.<\/p>\n AI Can (Mostly) Outperform Human CEOs<\/p>\n Mudassir et al. (2024) found that AI can mostly outperform human CEOs even now.<\/p>\n University of Cambridge researchers ran 344 participants (students and senior executives) in competition with GPT-4o\u2014a model that\u2019s already obsolete\u2014through a simulated CEO role in a U.S. automotive industry digital twin.<\/p>\n GPT-4o consistently beat the human players on growth, product design, market share, and profitability. When it failed, it got fired faster by virtual boards, largely because it tended to over-optimize for short-term gains. It also couldn\u2019t handle black swan events (such as pandemic-style demand collapses)\u2014human players adapted better to those.<\/p>\n But the bottom line: AI is probably already useful for augmenting a human CEO.<\/p>\n Where AI Hits Hardest: The White-Collar Surprise<\/p>\n Contrary to earlier waves of automation that displaced factory workers and routine clerical staff, generative AI is now penetrating the upper echelons of the labor market.<\/p>\n Felten et al. (2023) established that exposure to generative AI is highest for highly paid, highly educated, white-collar occupations\u2014lawyers, PR specialists, and instructors\u2014not assembly-line workers. Eloundou et al. (2023) found that roughly 80% of U.S. workers have at least 10% of their tasks exposed to LLMs, with higher-wage jobs showing greater exposure than lower-wage positions.<\/p>\n A field experiment with 758 BCG consultants by Dell\u2019Acqua et al. (2023) demonstrated that this isn\u2019t merely theoretical: On creative and strategic tasks (such as ideation, market segmentation, and persuasive writing)\u2014the bread and butter of elite professional services\u2014AI users finished 12% more tasks, 25% faster, and at 40% higher quality.<\/p>\n A more recent study with 776 P&G professionals by Dell\u2019Acqua et al. (2025) offers even more striking results. Individuals working with AI matched the performance of two-person teams without AI. Teams using AI were significantly more likely to produce exceptional (top 10%) solutions.<\/p>\n Perhaps most surprisingly, AI use increased positive emotions (excitement, energy) and decreased negative ones (anxiety, frustration) relative to a human working alone\u2014the AI provided similar emotional benefits as working with a human.<\/p>\n Working in a team with an AI looks increasingly plausible. And that brings us a step closer to the AI leading the team.<\/p>\n AI as Master Persuader<\/p>\n One of the CEO\u2019s essential roles is being \u201cpersuader-in-chief\u201d\u2014the person who must convince boards, employees, investors, and customers to follow a particular direction.<\/p>\n LLMs have proven to be remarkably effective persuaders<\/a>, often outperforming humans by significant margins.<\/p>\n Salvi et al. (2024) found that GPT-4 with access to personal information was 82% more persuasive than human debaters. A field experiment on Reddit\u2019s r\/ChangeMyView (2025) found AI persuasion rates three to six times higher than humans. In head-to-head comparisons, LLM-generated arguments have an 80\u201390% probability of being judged more persuasive than arguments written by incentivized human persuaders.<\/p>\n The Wade Test: Can AI Be the CEO\u2019s Voice?<\/p>\n Choudhury et al. (2024) looked at a core question: Can generative AI effectively automate CEO communications?<\/p>\n Their field experiment was conducted around a specific CEO, called Wade. An AI was trained on the CEO\u2019s prior communications, then 105 employees were randomly shown either Wade\u2019s own answers or AI-generated answers to ten questions.<\/p>\n The result: Employees couldn\u2019t tell the difference.<\/p>\n But here\u2019s where it gets interesting. When employees believed a response was AI-generated\u2014regardless of the actual source\u2014they rated it as less helpful. Labeling answers as AI-generated reduced perceived helpfulness even when the answers came from the actual human CEO.<\/p>\n The good news: We\u2019ve got an AI that can communicate like the CEO.<\/p>\n The bad news: Its success currently depends on people not realizing they\u2019re dealing with the AI.<\/p>\n But with wider use and social acceptance, that might change.<\/p>\n What This Means for You<\/p>\n Many of the tasks that defined \u201cknowledge work\u201d as protected from automation\u2014synthesis, persuasion, creative problem-solving\u2014are now within AI\u2019s expanding capability boundary. As Dell\u2019Acqua et al. (2023) noted, we still face a jagged frontier in AI capabilities; it cannot do everything a human knowledge worker can do, but that frontier is constantly shifting, and the tasks that are human-only for knowledge workers are shrinking.<\/p>\n And being a CEO is the ultimate knowledge work.<\/p>\n Even with current technology, AIs can do many of the things a CEO does. Jokes and snark aside, I think we\u2019re a considerable way off from a company truly having an AI CEO, for all sorts of legal, socio-political, and practical reasons.<\/p>\n What\u2019s more likely: CEOs will begin to use AI to augment themselves, delegating more as they and their boards become comfortable with AI\u2019s abilities. This is already happening. Eric Yuan of Zoom, Reid Hoffman, and other high-profile leaders are experimenting with AI clones<\/a>\u2014sophisticated digital twins that use their voice, likeness, or knowledge bases to scale their presence.<\/p>\n If you are an early career knowledge worker\u2014a starting management consultant, a new hire in a marketing department, or similar\u2014look for opportunities to manage AI \u201cteam members\u201d now. This is what you\u2019ll be doing a lot of over your career.<\/p>\n But there\u2019s another reason: Weidmann et al. (2025) found that leadership effectiveness with AI agents strongly predicts leadership effectiveness with human teams (\u03c1 = 0.81). Successful leaders of both AI agents and humans share the same behaviors: more questioning and conversational turn-taking. Learning how to lead AI well now – will quite possibly make you a better leader of humans later.<\/p>\n If you\u2019re in any kind of knowledge work role (and particularly if you are in a leadership role), it\u2019s time to start thinking about how you can augment yourself with AI.<\/p>\n If you are a board member, it\u2019s time to start thinking about how your C-suite is engaging with AI\u2014and whether the organization should monitor it. Should your C-suite be experimenting with digital clones of themselves? What guardrails do you want in place? Should the ability to collaborate successfully with AI be a competency you explicitly evaluate for your C-suite?<\/p>\n Across all these roles, one pattern is clear: Wage gains from AI are accruing disproportionately to workers who can leverage these tools effectively. The premium isn\u2019t for human judgment and problem solving alone\u2014it\u2019s for the capacity to leverage AI judgment effectively. Keep in mind, as Riedl & Weidmann (2025) noted, the ability to collaborate with AI is distinct from solo problem-solving ability. Using a 667-person interactive benchmark across math, physics, and moral reasoning tasks, the authors show that the capacity to work effectively with AI is a separate, measurable skill\u2014not merely a byproduct of being good at the underlying task. Collaborating with AI is a trainable, separable capability.<\/p>\n The Bottom Line<\/p>\n Your next boss probably won\u2019t be an AI. But it will likely be a human augmented by AI capabilities. And, in the long run, true AI CEOs might be coming.<\/p>\n \u201cShame on me if OpenAI is not the first big company run by an AI CEO.\u201d \u2014Sam Altman, Conversations with Tyler podcast, November 2025.<\/p>\n And aside from automation bias (our tendency to over-rely on technology) and the results of several studies showing that AI+Human often underperforms AI alone\u2014what could possibly go wrong?<\/p>\n \u201cAI will probably most likely lead to the end of the world, but in the meantime, there\u2019ll be great companies.\u201d \u2014Sam Altman, Airbnb Open Air Conference, June 2015<\/p>\n Skynet<\/a>, Daleks<\/a>, and Colossus the Forbin Project<\/a>\u2014here we come!<\/p>\n What\u2019s your experience working with AI? Are you already augmenting yourself\u2014or your team? If you are a board member, or an advisor to a board or CEO, are you thinking about creating AI clones of your C-suite? What are your policies for C-suite use of AI? I\u2019d love to hear in the comments.<\/p>\n