Major corporations betting on artificial intelligence have encountered a paradoxical situation: almost all productivity gains come from just a small share of employees. This is the "super Pareto principle," in which 10% of workers generate 90% of the results and token consumption.

The classic Pareto rule states that 20% of effort produces 80% of the result. In the case of AI, the skew is far more radical. Tokens are units of text processed by a language model; their consumption shows how actively a person uses AI at work. It is this metric that exposes the gap between a small group of advanced users and the bulk of employees.

Just 10%

Most companies investing in AI find that 10% of employees account for 90% of productivity gains and token consumption. The gap is observed across all areas — from software development to non-technical knowledge work.

At the same time, corporations invest considerable time and resources in training. Employees are taught how large language models underlying modern AI assistants work. Many departments receive training tailored to their tasks. Despite these efforts, the share of active users has not shifted.

Six uncomfortable conclusions

Such results force companies to reach harsh conclusions:

  • the company is overpaying 90% of employees and underpaying the remaining 10%;
  • it would be more sensible to distribute tokens in proportion to output, but it is unclear how to do so without harming corporate culture;
  • large-scale layoffs are probably justified and would not hurt overall results, but companies fear a negative reaction from the workforce;
  • it is unclear how companies outside the tech sector can hire more people who, with the help of AI, work many times more productively than others;
  • there are doubts about whether efforts on training and implementation pay off at all;
  • there remains the risk that the best 10% will leave for places where they can work among equally advanced AI users.

It is precisely the fate of lagging employees that has become the number one problem for executives. While some are rewriting workflows around AI, others remain at the same level of efficiency — and this is turning into a strategic challenge for the entire business.

Cryptalist analysis: The polarization of AI skills is forming a new class of corporate inequality that directly affects the assessment of returns on technology investment. For the market, this is a signal that capital will flow to companies capable of scaling competencies, not just purchasing models. Investors should more carefully evaluate not the size of AI budgets, but the real depth of penetration of tools into workflows.