Why AI can replace software engineers but not chief executives, according to chief executives
The automation debate reveals less about technology than about who holds power over how we value labour
Christine Carrillo runs a tech startup. She also coaches seven CEOs monthly, surfs daily, and cooks dinner every evening. How does she manage it? In April 2021, she explained on Twitter: her executive assistant handles sixty per cent of her work. The EA manages most emails, runs fundraising, executes sales campaigns, builds operational playbooks, updates financial models, recruits new hires, and onboards them. Carrillo's assistant is based in the Philippines.
The internet's response was swift and merciless. If someone else does sixty per cent of your job, perhaps they should earn sixty per cent of your pay. The technology journalist Christopher Mims suggested the incident made a good argument for universal basic executive assistants. The writer Sonia Saraiya was more direct: the most overvalued asset of a company is a CEO.
Carrillo had inadvertently surfaced a question most executives prefer to keep buried. If a chief executive's daily tasks can be delegated to someone earning a fraction of their compensation—or automated through the software tools Carrillo mentioned using—what exactly is the chief executive's job? And if the answer involves intangible qualities like leadership and strategic vision, why don't those same qualities protect workers in other roles from replacement?
What chief executives actually contribute
The empirical research offers a more modest picture than compensation packages suggest. Economists at Harvard Business School and the London School of Economics tracked over 1,100 CEOs across six countries and found that firms led by "leader-type" CEOs showed about seven per cent higher sales after controlling for inputs. Seven per cent. Not sevenfold. Not even double.
More importantly, CEO impact operates primarily through matching rather than through a hierarchy of executive quality. Different firms need different leaders. About seventeen per cent of companies in the sample had a CEO whose style was simply wrong for their needs. The brilliant turnaround artist might be disastrous for a company needing steady operational management. What matters is fit, not inherent superiority.
This undermines a common justification for stratospheric pay: that a tiny elite of exceptional executives are worth whatever the market will bear. The research suggests CEO effectiveness is largely horizontal. There isn't a clear ladder where the best CEOs are worth vastly more than the second-best. The variation is real, but it's mostly about pairing the right person with the right situation.
The compensation puzzle
None of this would matter if CEO pay had grown roughly in line with the value CEOs demonstrably add. It hasn't.
Since 1978, CEO compensation at large American firms has grown by 1,460 per cent. Typical worker compensation has grown by 15.3 per cent. The ratio of CEO to worker pay has expanded from about twenty to one in 1965 to nearly four hundred to one today.
If high CEO pay simply reflected a competitive market bidding for scarce talent, we would expect it to grow roughly in line with other highly paid professionals. It hasn't. CEO pay has outpaced even the top 0.1 per cent of earners—the surgeons, the partners at elite law firms, the investment bankers. The Economic Policy Institute calls the excess "economic rent": income that exceeds what would be necessary to attract someone to the role. CEOs are not paid what they're worth in any straightforward market sense. They're paid what they can extract.
The mechanism matters. Compensation committees, typically composed of independent directors, recommend executive pay. In theory, they provide objective oversight. In practice, their members are often current or former executives at other companies. The SEC requires disclosure of "compensation committee interlocks" when board members serve on each other's committees, but the broader cultural issue resists regulation.
Consultants benchmark pay against peer companies. A consultant who consistently recommends below-market figures will find themselves recommended less frequently. Each company benchmarks against others who are themselves benchmarking against the first. The result is a self-referential spiral, each round ratcheting the numbers higher.
The sophisticated justification
Tournament theory offers an elegant defence of the gap. Borrowed from sports economics, the argument holds that CEO pay functions as a prize, motivating everyone below to compete for advancement. Even if the current CEO doesn't directly earn their pay through productivity, the prize creates incentives throughout the organisation. In professional golf, larger purses correlate with better average scores.
But management isn't golf. In sports, we observe the competition. We know who played better because we watched. In corporate hierarchies, the tournament is invisible. The criteria for winning are set by the same people who award the prizes. And when researchers examined whether large executive pay gaps actually improve firm performance, the results disappointed. One study found that firms with larger pay disparities in top management exhibited more earnings manipulation and worse future performance. The tournament fosters destructive behaviour as often as constructive.
The automation question
This brings us to the question Carrillo's revelation forces into the open. If most of a CEO's daily tasks can be delegated or outsourced, and if the remaining value-add is primarily about matching the right leader to the right organisation, what happens when we apply the same automation logic to executive roles that we've been applying to everyone else?
The technology industry has spent years proclaiming that artificial intelligence will transform labour. Generative AI can write code, draft legal documents, analyse financial data, handle customer service. Entry-level positions are particularly vulnerable. Anthropic's chief executive Dario Amodei has warned that AI could eliminate half of entry-level white-collar jobs within five years. Amazon's Andy Jassy told employees his company's corporate workforce will shrink from AI. Shopify requires managers to prove AI can't do a job before posting a new position.
When engineers, lawyers, and content creators argue that their work involves intangible human elements AI cannot replicate, they're told this is wishful thinking. Tasks can be decomposed. What can be decomposed can be automated. But when the same argument turns toward executive roles, the response shifts. Chief executives provide unique strategic vision, we're suddenly told. They maintain crucial relationships. They inspire in ways no algorithm could match.
The irony is difficult to miss. If Elon Musk can serve as CEO of multiple companies while maintaining a prolific social media presence and attending to various personal pursuits, the cognitive demands of the role cannot be quite as singular as defenders suggest. The same executives who dismiss workers' claims about the irreducible human elements of their jobs invoke precisely those claims when their own positions come under scrutiny. Relationship management, talent attraction, strategic intuition—these are the defences offered for executive work. They are also the defences software developers and customer service representatives have offered for theirs, to considerably less sympathetic audiences.
The asymmetry
The CEO automation question is provocative but somewhat beside the point. AI won't literally run companies in the near future. The question's value lies in what it exposes about how we assign economic worth.
The same "intangible human skills" argument defenders of CEO compensation invoke could apply to many roles facing automation pressure. Teaching involves relationship building and individual attention. Nursing requires empathy and judgement no algorithm possesses. Creative work depends on human experience and cultural understanding. These arguments are dismissed when workers make them. They're treated seriously when executives make them about their own positions.
The difference isn't the nature of the work. It's the power to defend one's position in the labour market. CEOs sit on each other's boards. They hire the consultants who recommend their pay. They control the organisations that lobby for their interests. Workers in most roles lack these structural advantages. When automation pressure arrives, they face it individually, armed only with the argument that their work involves something irreducibly human.
This asymmetry explains why the CEO automation question generates such different reactions depending on who asks. When a labour-oriented publication raises it, the business world dismisses it: leadership can't be automated; the question misunderstands what executives do. When the same question targets software engineers or customer service representatives, the tone changes: adapt or die; the technology is coming.
The quiet part
Angela Monta, twenty-five, works through the night five days a week as a virtual assistant to a California venture capitalist. She earns $1,200 per month from her home in San Mateo, Philippines—work that would command several times that figure if performed by an American in an American office. She's not unhappy. The pay exceeds her former government salary, and she avoids the commute. But her position in the global labour market depends on the same decomposition logic AI now threatens to extend: if work can be done remotely, it can be done cheaply; if it can be done cheaply, it can eventually be automated.
The viral response to Carrillo's thread wasn't really about one CEO's surfing schedule. It was about recognition. The economic logic applied to labour displacement has always been applied selectively. The argument that tasks can be decomposed and automated is wielded against workers without power to defend themselves. When it's wielded against those who hold power, suddenly the irreducible human element becomes important again.
The question of whether CEOs could be automated is less interesting than the question of why we don't ask it more often. The answer has little to do with the actual complexity of executive work. The Harvard research suggests CEO impact is modest and depends heavily on fit. The compensation data shows pay growth has dramatically outpaced any measure of increased value. The automation evidence suggests the same logic used to justify white-collar displacement could apply, with equal validity, to executive roles.
What protects CEOs isn't the irreplaceable nature of their contribution. It's the power to define which contributions are irreplaceable. That power is not distributed equally. And the CEO automation debate, absurd as it might seem, is one of the few contexts in which this asymmetry becomes briefly, uncomfortably visible.