There have been lots of doomsday predictions about the impact of AI on the labor market that have failed to come to fruition. Core to these predictions is a fundamental misunderstanding about what many jobs actually entail. Many roles, especially white-collar jobs, are functions of general intelligence and trust. Because of the responsibility element, those jobs are more difficult to automate than one might initially think.
Behavioral economist and advertising executive Rory Sutherland has used the example of hotel doormen to illustrate how easily we misunderstand the true utility of certain occupations. A consultant working with a “mechanistic model of the world” might define a hotel doorman’s function as simply opening the door, concluding that an automatic door can perform the same task more efficiently. But that misses much of what a doorman actually does.
Beyond opening doors, a doorman provides a visible human presence, assists guests, monitors activity around the entrance, and discourages disruptive behavior. Those functions may be difficult to quantify, but they are nevertheless part of the value that the job creates.
Take my own work at a think tank. From the outside, it may appear that my job is simply to produce research. This may lead some to ask, “Aren’t you worried about losing your job once AI can write and think just as well?” But that concern is predicated on the same oversimplification of occupational function.
Working at a think tank involves far more than just producing written analysis. It also means representing an organization well, communicating research to important stakeholders, building relationships, and continually adapting to an ever-evolving policy landscape.
Another example is sales. The job of a salesperson is not simply to convey information, but also to develop relationships, understand human behavior, and negotiate in real time based on non-verbal cues. Salespeople can be described as “pushy” at times, but that is, admittedly, a very human quality.
Many jobs also have uneven schedules of value production. Rather than requiring consistent outputs, some jobs require performance during very specific moments. In financial reporting for example, most of workers’ visible output is based around quarterly and annual reporting deadlines.
For some jobs, the value production schedule can be unpredictable. Someone hired for exceptional communication skills, for example, may not demonstrate that value every day, but when those skills are needed, their importance becomes immediately apparent. To put it in extreme terms, laying someone off to save a corporation $100k per year may cost the firm $5 million when they don’t have that same individual around to negotiate and land a critical deal.
All of this is to illustrate that jobs are complicated. We live in a messy world, yet predictions about technology-induced unemployment often rely on simplistic assumptions about what workers actually do. If a job is reduced to its most visible component, then automation appears far more likely than it actually is. As of 2026, the data reveals that AI is not having a substantial impact on the labor market, and the biggest reason for that is that jobs are more complex than people imagine.
