The One Job AI Is Not Allowed to Touch

Added: 2026-05-10

[00:00:00]In the last 3 years, AI has replaced customer service workers, junior developers, uh copywriters, parallegals, data analysts, even the people who took your drive-thru order, the people who read your medical scans. And here's the thing, they were all told the same thing. The technology can do what you do, but just faster, cheaper, without the healthare plan. And in boardrooms across the world, the people making that decision, yeah, they nodded, they signed off, and they went back to their $14 million compensation packages. Okay, but here's the question. Nobody in those boardrooms has been asked out loud. So, if AI can replace the people at the bottom, why can't it replace the people at the top? The answer, yeah, it has nothing to do with capability and everything to do with something much older than AI. This is Barely Human Labs. We investigate modern life in the age of AI so that you don't have to.

[00:01:02]Today, the one job that never appeared on any McKenzie list of roles vulnerable to automation. The one desk that has remained completely untouched, while of course, every other desk in the building got a new AI colleague or a pink slip.

[00:01:17]The CEO's desk.

[00:01:20]Yeah. So, we're going to look at whether AI could actually do the job, what the research says, and of course, most importantly, why the answer to whether it happens literally has nothing to do with whether it should. Let's start with an honest question, okay? One that rarely gets asked honestly. What does a CEO actually do all day? Not the LinkedIn version, not the, you know, TED talk version, the actual version.

[00:01:47]Research from Harvard Business School found that CEOs spend the majority of their time in meetings reviewing information prepared by other people, uh, making decisions based on that information and then communicating those decisions downward. That's the job description. Okay? And here's the uncomfortable part. Reviewing information prepared by others is summarization.

[00:02:09]AI does that. making decisions um that are based on structured inputs in pattern recognition and decision support. AI does that. Communications to large groups is language generation. AI does that. The structured information in decision out nature of a significant portion of executive work is not the frontier of AI capability. It is squarely in the middle of it. Okay. So then why is nobody in the boardroom talking about this? We'll get there. But first, let's look at the track record of humans currently in the role. So now let's look at the actual decisions CEOs have been making, okay, with their irreplaceable human judgment. So in 2022 and 2023, multiple major tech companies went on enormous hiring sprees based on growth projections that didn't materialize. They then conducted mass layoffs framed publicly as AIdriven efficiency that caused billions in severance and destroyed employee trust across entire industries. The humans who made the overhiring decisions also made the layoff decisions. But here's the thing, those humans did not get laid off. Claro publicly fired 700 customer service staff and celebrated the AI that replaced them. then quietly rehired because the AI couldn't do the job adequately. So the CEO who made that call received no consequence. The 700 people who lost jobs for less than a year and came back with less negotiating power. They received the consequence.

[00:03:44]Boeing the decision that led to two fatal crashes were made by executives.

[00:03:48]The engineers who raised concerns were overruled. The workers who reported the problems, yeah, they were just ignored.

[00:03:55]The CEO who presided over that period left with a package worth tens of millions. The people on those planes, yeah, didn't receive a package. And here's the thing, this is not a list of exceptions. This is a pattern. And the pattern has a structure that AI replacement would theoretically fix.

[00:04:14]Which brings us to the actual case for putting AI in the corner office. Okay, let's make the argument properly because yeah, it deserves to be made properly.

[00:04:23]The case for AI decision-making at the executive level, it's not a joke. It is a researched legitimate argument. First, cognitive bias. So, human executives carry the full catalog of human cognitive biases. Overconfidence, sunk cost thinking, ingroup favoritism, the tendency to make different decisions depending on whether it's before or after lunch. These are documented.

[00:04:50]Seriously, they're measurable and they scale badly when the person experiencing them is making decisions that affect thousands of people. AI systems have their own biases inherited from training data, but those biases are in principle auditable. You can look at them, you can measure them, correct them. A CEO's cognitive biases are not on the quarterly earnings call. They just show up in the outcomes. Second, consistency.

[00:05:18]Human judgment is variable, right? So research on judicial decisions a reasonable analog to executive decisions, it shows that outcomes vary depending on factors that have like nothing to do with the actual case. What time it is, whether it's before a break, personal circumstances on that day. An AI system applies the same logic to the same inputs every single time. for decisions where consistency matters like resource allocation, risk assessment, hiring frameworks, that is not a weakness. It is precisely what you want.

[00:05:52]And third, information. A CEO can process a limited amount of information.

[00:05:57]An AI system can process everything simultaneously.

[00:06:01]Every market signal, every internal data point, every comparable decision made by every competitor. A well-run study by MIT and Stanford found that AI decision support at the executive level performed comparably to human executives on structured decision tasks. So the category where AI underperformed most was middle management, reading a room, judgment calls, uh navigation ambiguity.

[00:06:28]So in other words, the people most protected from AI replacement by capacity are middle managers. The people actually being protected though, yeah, they're the executives. And here's where the argument stops being about capability because the capability case is made. What comes next is why none of it even matters. Barely Human Labs, two videos a week, Tuesday and Saturday. If this is landing, then subscribe [music] and turn on notifications. Membership in the description if you want to support directly. It keeps these coming. Back to the corner office. Okay, so here's the central problem with every argument in part three. The people who would implement AI replacement of executives are the executives. I mean, that's the whole point, right? The decision to replace workers with AI was made by CEOs. The decision to replace CEOs with AI would have to be made by Yeah. CEOs or by boards of directors who are appointed largely based on relationships with CEOs and who are also not on any list of roles. vulnerable to AI displacement or by shareholders who can theoretically demand anything but in practice defer to management on operational decisions.

[00:07:42]When a junior developer is replaced by AI, it happens in a quarter. Decision made, headcount cut, savings appear in the next earnings report. When a CEO is proposed to be replaced by AI, it would require the board to decide the CEO is replaceable. the board that yeah the CEO helped appoint using criteria the CEO's organization would define implemented by the executive team that responds to the CEO. It is not going to happen. Not because the capability argument fails.

[00:08:14]We just made that argument, right? But because the incentive is missing for the only people with the power to make it happen. Let's put the salary number on the table because it earns its place here. So, in 2023, the average CEO of an S&P 500 company earned approximately 272 times the median worker salary at their company. That's a documented figure from the Economic Policy Institute.

[00:08:39]272 times. The argument used to justify replacing workers with AI is they cost money, they make mistakes, the technology can do the job. On cost, a CEO at $15 million a year costs more than $272 workers earning the median salary. The ROI case for AI replacement is arithmetically stronger at the top than anywhere else in the building. On mistakes, we just went through the track record, right? overhiring, mass layoffs, products that killed people, strategies that literally cost billions with essentially no personal consequence for the decision maker on the technology doing the job. We just made that case, too. Every single part of the replacement argument applies more powerfully to the corner office than to the people who got replaced. And now, here's the part that reveals what this was really about all along. So when AI makes a mistake at the worker level, there's a human in the loop who takes the blame. The human in the loop as legal shield. You approve the output.

[00:09:46]The output was wrong. The consequence lands on you. When a CEO makes a mistake, the consequence lands on the workers, the customers, and the shareholders. The CEO lands on a board somewhere or gives a TED talk. The decision maker and the consequence receiver are in most corporate structures different people. AI replacement has been applied exclusively to the consequence receivers, never to the decision makers. That's not a technology story. That is a power story.

[00:10:17]And it has been a power story for a long time. If you're feeling something specific right now and you'd like to feel it again next week, subscribe. New video every Tuesday and Saturday.

[00:10:28]membership in the description if you want to support the channel directly.

[00:10:32]Back to the experiment. Let's be honest about what has actually been tried because this area attracts a lot of noise and deserves precision. A Polish drinks company appointed an AI system called Mika as its CEO in 2023. The coverage was significant. The reality was that Mika handled specific operational decisions, things like uh logistics, supply chain, while human executives continued to make all strategic decisions. It was interesting.

[00:11:03]It was not a replacement. It was design support software with a name and a press release. Netragon Weboft, a Chinese company, appointed an AI as its official CEO in 2022. Their stock outperformed the market in the period following.

[00:11:19]Whether that was, you know, because of the AI, the underlying business, or because the story attracted investor attention. Yeah, that's genuinely difficult to separate, right? Both experiments are real. Both were carefully constructed to never actually threaten the humans running the company.

[00:11:35]The AI CEO in every case was a tool the executives used, not a replacement for them. Companies are willing to use AI CEO language for the press coverage.

[00:11:46]They're not willing to use AI CEO reality for the accountability, which tells you exactly what you need to know about how seriously they take the idea. And here's the honest conclusion that the AI replacement conversation has been avoiding. The decisions about what gets automated and what doesn't, they're not being made based on what the technology can do. Yeah. They're being made based on who has the power to make the decision. Workers get automated because the people deciding have every reason. Cost savings, efficiency metrics, stock price to do it and no personal consequence if it goes wrong.

[00:12:24]Executives don't get automated because the people deciding are the executives.

[00:12:29]If the logic were applied consistently, if AI can do the job genuinely drove replacement decisions, the boardroom conversation would look very different.

[00:12:38]But of course, it doesn't because the logic was never the driver. The power structure was the driver. The logic was the press release. This is Barely Human Labs. We run these experiments so you don't have to. New video every Tuesday and Saturday. Subscribe. Membership in the description. Now, the ending this deserves. The AI revolution as it is currently being run is a revolution directed by the people at the top for the benefit of the people, you guessed it, at the top. The workers at the bottom experience it. The executives at the top design it. AI didn't create that structure. That structure is old, much older than AI. What AI did was give it a better story. The story is the technology is neutral. The displacement is inevitable. Nobody chose this. It's just what progress looks like. None of that is true. Someone chose this.

[00:13:33]Someone signs off on every replacement decision. Someone receives the bonus when the headcount number drops. And that someone is not on any list of roles vulnerable to automation. This is Barely Human Labs. Subscribe because the next experiment is already running and the people who designed it are not in it.