They are suffering from delusions of grandeur

Aaron Levie, founder and CEO of Box, has realized something: AI is causing some managers to suffer a certain disconnection from real work and believe that AI does things that in reality (for now) can’t do. for him it’s clear that what the CEOs of technology companies are experiencing is an “AI psychosis.”

AI myths and realities. This year we are experiencing frenetic movements in the technology industry. Stock market valuations of technology companies skyrocket, but at the same time Mass layoffs accelerate. There seems to be an explanation gaining momentum in Silicon Valley: those in charge of technology companies are suffering from what Levie calls “AI psychosis.” There are in these moments some cognitive disconnectionand CEOs and senior officials believe that AI can do tasks today that in reality still require being under human control and expert judgment.

There is a long way from saying to doing. According to the CEO of Box, managers are sensitive to this “delirium” because “they are sufficiently removed from that last stretch in which the work is done.” That is to say: a CEO sees a prototype of an AI model that generates a contract or a line of code and believes that is enough to declare that the work is done. However, it is not these managers who have to review that code in search of flaws or analyze contracts in search of misleading or false clauses that the AI ​​has invented.

100x Organizations. There is a particularly surprising case in this area. Zeb Evans, CEO of the project management startup ClickUp, recently said in X that had laid off almost a quarter of its employees after deploying 3,000 AI agents to do their job. According to him, the human employees who have remained in the company simply have to supervise the machines, forming what Evans called a “100x organization.”

What the CEO sees vs. what he should see. The triumphalist messages of some companies and CEOs like Evans can be quickly contrasted with the data we have today. The decisions being made – for example, in the area of ​​layoffs that are often hidden behind the adoption of AI – should be based in improved productivity which at the moment does not exist. Some studies made it clear:

• A study from the University of California at Berkeley evaluated several investigations in this regard and concluded that “there is no robust relationship between the adoption of AI and an aggregate productivity gain.”
• Other investigation of the National Bureau of Economic Research (NBER) indicated that AI had indeed improved productivity, but found “a productivity paradox, whereby perceived productivity gains are greater than measured productivity gains.”
• Finally, MIT researchers They created thousands of agents to work on various tasks and concluded that in many cases they did not perform those tasks with human quality. According to their estimates, AI models will be able to complete many tasks “with 80-95% success in 2029 with adequate quality,” but they will not yet surpass human workers.

The new bottleneck. The danger of this “psychosis” is that by automating the production of content or code the problem does not disappear. It just moves. If everyone uses AI to produce more things, the bottleneck is precisely the managers who must manage and control the review of a volume of data that did not exist before. It is in fact just what Harvard Business Review denounced in a recent analysis. Levie made it clear: CEOs must “go down into the mud” and see what AI can and cannot do, because otherwise what they will end up having is true organizational chaos.

Image | Hunter Race

In Xataka | It is normal that CEOs of technology companies like AI. They are using it to “clone” themselves and not go to work