"What do I do with the AI laggards?": the power law inside companies
"What do I do with the AI laggards?": the power law inside companies
A head of AI at a large company says 10% of staff produce 90% of the AI productivity gain, and training has not changed that.
Rufus's note (2026-10-28): This is skill-biased technical change on steroids, and it matches my experience so far.
Links
- Post: https://x.com/businessbarista/status/2115145399396671997
- X summary: https://x.com/i/trending/2115206039461388541
Overview
Alex Lieberman reports a conversation with the head of AI at "a huge company". Its biggest problem is "the AI laggards". A "Super Pareto Principle" holds: 10% of employees account for 90% of the gain in productivity and token use, from software engineering to non-technical knowledge work. The company has spent heavily on enablement, from LLM basics to training for each function, and "the power law of productivity increases hasn't budged."
The questions they are left with:
- Are 90% of people overpaid and 10% underpaid?
- Should tokens go to people in line with the power law, and how without killing the culture?
- Are big layoffs warranted, given they would not cut output, and what about the backlash?
- How does a non-tech company hire more "power law people"?
- Is the training and enablement worth it?
- How do they stop the top 10% leaving to work with other AI-native people?
Why interesting
Economists call it skill-biased technical change when a new technology raises the value of some workers far more than others. Here the gap is not between jobs but between people doing the same job, and training has not closed it. That changes the jobs question: not "which jobs does AI replace?" but "which people inside each job pull ahead?" It is one anecdote, but it fits what we see.