We Should Replace CEOs With AI
Possibly the only automation argument that makes sense. Or does it?
The big promise of AI is that it will decouple productivity from labour, enabling rapid, infinite growth. But even when we gloss over the fact that infinite growth within a finite system is literally the definition of cancer, AI is still not meeting this promise. Study after study has shown that it is too inaccurate to augment jobs, let alone replace them (read more here). These studies also show that, in the vast majority of cases, AI is way too expensive to have a positive effect on anyone’s bottom line (read more here). What’s worse, more and more research is suggesting that, even with the trillions of dollars being poured into AI development, these models aren’t going to get much better than they currently are (read more here). Thanks to the rule of diminishing returns, AI is about as good as it will ever be. But what if I told you this is only because we have approached AI the wrong way? In reality, AI can seriously disrupt the labour market, heavily boost productivity, improve workers’ lives, and make a killing while doing so. It just can’t do this from the bottom-up, as we have been trying to do. But it can be done from a top-down approach, and in fact, it makes a whole lot of sense. This is why we should replace CEOs with AI.
AI Automation Limitations
AI hallucinates (or more accurately, gets things wrong) constantly. In fact, OpenAI’s latest research paper found that there is no viable way to reduce the current model’s rate of hallucinations (read more here). This severely limits which tasks and jobs AI can automate. Basically, it needs to be a task/job where accuracy is not important. But you still need a human expert to oversee this output in order to catch, correct and mitigate mistakes. Ideally, this oversight will be inherently structured around the task/job; that way, oversight will not have to come from an additional hired position.
Even when a task/job meets these requirements, there is a “cost floor”. AI models are extremely expensive to run, let alone build and develop. There is a reason OpenAI and its cohorts are light-years away from profitability. As such, for AI suppliers to be sustainable and the companies using AI to automate jobs to generate a return from their AI use, the most expensive tasks/jobs should be the ones automated first. This gives the best chance of the astronomical cost of AI deployment being less than the money it saves, ensuring the AI supplier and user aren’t experiencing a net loss.
So, AI can automate extremely well-paid jobs that don’t depend on accuracy and that involve inherent human oversight.
Notice how that is the total opposite of how AI is currently being deployed? From report writing to bookkeeping and data entry, and even coding, AI is being used to replace relatively low-paid jobs that depend entirely on accuracy and have very little inherent human oversight. This is a total arse-about-face way to deploy AI.
Okay, so do we know of a job that doesn’t depend on accuracy, has inherent human oversight and is extremely well paid? Yes, we do. CEO.
The Case For CEO Automation — Economics
The average pay for a US CEO in 2024 was $22.98 million. That makes being a CEO the highest-paid “job” around. In 2024, the average US CEO earned 281 times the salary of the average worker beneath them. Almost all S&P 500 companies cluster around this ratio, but some outliers pay their CEOs significantly more. For example, Starbucks’ CEO made 6,666 times more than the average Starbucks worker in 2024.
On a totally separate note: Join a union.
As a result, automating the single job of CEO can save most companies as much as, if not more than, automating the entirety of their low-paid admin staff. Take S&P 500 company Welltower—its CEO earned $17,199,891 in 2023, while its 533 employees earned, on average, roughly $100,000 each. If Welltower automated its CEO position, it would save as much as automating a third of its workforce! And Welltower is not an outlier here.
Furthermore, automating a single role with a single AI is far easier and cheaper to deploy, as well as far easier and cheaper to manage, than automating hundreds or thousands of different interconnecting jobs with AI.
In terms of the economics and practicality of deploying AI automation, the CEO position makes the most sense.
The Case For CEO Automation — Accuracy
But surely the job of CEO needs 100% accuracy, right? They are literally running the ship. Well, truth be told, the public perception of what a CEO does and what a CEO actually does are miles apart. CEOs might consider themselves the captain, but they are more like a floating busybody—at least according to the data. And guess what? Busybodies don’t need to be accurate.
Okay, so what data supports this?
