In a three-month experiment, AI improved the quality of patent drafts, but only experienced lawyers showed an average improvement when later tested without the tool. The findings raise a practical question for workplaces: can AI improve today’s output without weakening the learning that builds tomorrow’s expertise?
Watch Desk analysis
What happened
The study involved 133 lawyers at 11 US intellectual-property firms. Researchers randomly gave two-thirds access to a custom AI patent-drafting assistant; the others initially worked without it. After 90 days, the AI group produced drafts that scored 0.38 standard deviations higher, which the authors describe as an 11-percentile-point gain relative to control scores.
The researchers then tested lawyers on a patent-editing task without AI. Experienced lawyers who had used the assistant scored 0.45 standard deviations higher than experienced lawyers in the control group. Junior lawyers showed no average improvement. The study sorted participants by experience, not age: “senior” meant at least seven years in practice. Read Fortune’s account of the research.
Key findings
- Draft quality improved with AI access
After 90 days, the AI group’s drafts scored 0.38 standard deviations higher than the control group’s. - The unassisted result differed by experience
Experienced lawyers showed an average gain on the later AI-free task; junior lawyers did not. - The study measured a short-term, specific task
Participants drafted patents from simulated materials, then edited a hypothetical patent without AI. - The evidence has important limits
Only 91 of the 133 lawyers completed the final test, and the working paper has not been peer-reviewed.
Why it matters
The distinction is between doing better work with a tool and getting better at the work yourself. Fortune reports that experienced lawyers described using AI as a “logic auditor”, while some juniors liked skipping the blank-page stage and moving straight to review. But fluent assistance may be less useful when someone has not yet learnt how to spot a flawed argument.
This does not show that AI prevents junior workers from becoming experts over a career. The trial lasted three months, involved patent work for a single sophisticated client, and did not measure participants’ skills at the outset. The paper also says Google conducted the study and paid its direct costs; its six co-authors included Google employees.
Our read
The finding is neither “AI makes everyone an expert” nor “keep the new tool away from juniors”. It is a reason to take practice seriously. For learners, one practical approach suggested by David Autor is to frame an argument unaided, then use AI to challenge and test it. For employers, the harder job is to preserve guided practice rather than treating a better first draft as proof that training is no longer needed.
What to watch
- Whether further studies find similar differences beyond patent-law work and this short trial.
- Whether employers pair AI use with unassisted assessments and structured mentoring.
- Whether later research shows how these effects develop over years, not months.
Discussion spark: Should employers require workers to practise important tasks without AI before relying on it, or would that preserve outdated work rather than build useful expertise?
Sources and evidence
- Source update (10 October 2026, 12:49 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.