MIT’s David Autor warns AI may widen the skill gap for Gen Z workers

Gen Z was handed a tidy promise about AI: give a 24-year-old a capable enough assistant, and their work starts to look like a veteran’s.

New research co-authored by MIT economist David Autor suggests that promise may run in reverse. AI made everyone’s drafts better. It did not make everyone better.

The working paper, published by the National Bureau of Economic Research in September 2026 as Working Paper 35720, followed 133 patent lawyers from 11 US intellectual property firms through a three-month randomized controlled trial. The most important result only appeared after the AI was switched off.

What the patent lawyer experiment found

Participants worked with a custom Google AI drafting assistant called InFlow. The tool has since been folded into Gemini Notebook.

With the assistant in hand, drafting quality rose quickly. Output improved by 0.34 standard deviations after 10 days and by 0.38 standard deviations at the 90-day mark.

Junior lawyers looked like the clear winners at this stage. They saw the biggest improvements in quality and efficiency, finishing tasks 18 minutes faster on average.

In an unaided test of independent judgment, lawyers who had used the AI beat the control group by 0.32 standard deviations overall. Senior lawyers with AI experience outperformed their peers by 0.45 standard deviations.

Junior lawyers showed no average improvement.

The flat average concealed something odder. Among juniors, the middle of the distribution hollowed out: mediocre results became less common, while both strong and weak performances became more frequent.

The equalizer theory meets the expertise problem

The researchers concluded that foundational expertise is what allows people to get lasting value from AI. Without that base, the tool can raise the quality of the work product without necessarily raising the skill of the person producing it.

The findings slot neatly into Autor’s broader body of research on how technology affects skills and labor markets unevenly. New tools rarely land on everyone the same way, and this paper adds generative AI to that list with unusually clean experimental evidence.

Google Research published its own announcement about the project on October 7, 2026. Co-authors on the paper include Tanya Rodchenko and Josh Martin.

What this means for Gen Z and the firms hiring them

The core tension the study exposes is between short-term productivity and long-term skill building. For employers, a junior lawyer finishing tasks 18 minutes faster with better output looks like a pure win on a timesheet. The unaided test suggests the dashboard may not capture whether that associate is actually becoming a better lawyer.

The polarization finding cuts both ways for Gen Z workers themselves. Some juniors in the study came out of the trial performing better on their own. Others came out performing worse. AI did not set a ceiling or a floor; it seems to have stretched the range.

There are clear limits to keep in mind. The trial covered one profession, patent law, over three months, and the paper is a working paper rather than a final peer-reviewed publication. Whether the same pattern holds for coders, analysts or writers is an open question the study does not answer.

The useful question is no longer only whether AI makes junior workers faster. It is whether the way they use it today leaves them more capable, or more dependent, when the tool is not in the room.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.