Three months of using an AI writing assistant improved patent drafts in a field experiment, but junior lawyers showed no average gain when later working without it. The result puts a useful wrinkle in the promise that AI can help people learn while they work: better output with a tool does not automatically mean better independent performance.
Watch Desk analysis
What happened
PsyPost reports on a preregistered, three-month experiment involving 133 US patent lawyers. Lawyers given access to an unreleased Google Labs writing tool produced drafts that scored about a third of a standard deviation better than the control group’s, and finished the first drafting task around 10 minutes faster on average.
On a later redlining task completed without AI, the AI group again did better overall, but that advantage came from senior lawyers. Junior lawyers showed no average improvement over junior lawyers in the control group. The study authors suggest that existing expertise may affect whether AI-assisted work builds skills, but the experiment did not establish why the groups differed.
Why it matters
AI assistance can make the work in front of someone better without making that person better at the work alone. That distinction matters to employers weighing productivity against training, and to anyone assuming that access to a capable assistant automatically develops expertise.
The study is evidence about patent lawyers and specific writing tasks, not a verdict on every profession. It was funded and conducted by Google, which developed the tool, and the paper has not yet been peer-reviewed. Those limits belong in the conversation, not in the bin.
Our read
The useful takeaway is not “AI makes juniors worse” or “AI trains seniors”. Neither conclusion is established here. It is that assisted performance and independent skill are different measures, and both deserve a place when organisations evaluate AI at work. Employers should be wary of counting polished output as proof that people have learned the craft.
What to watch
- Whether the findings hold in larger studies and professions beyond patent law.
- Whether later research identifies what separates AI use that builds expertise from use that merely improves the immediate draft.
- How employers measure learning and independent performance alongside speed and output quality.
Discussion spark: When workplaces adopt AI, should they prioritise better work now or evidence that staff can still do the work well without it?
Sources and evidence
- Prior expertise may determine whether AI builds skills or substitutes for them, field experiment suggests (7 November 2026, 14:00 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.