What a law-firm comparison reveals about making AI worth exploring

Two divisions at one law firm received the same drafting AI, called LawBot by the researchers. Where managers offered a route into more interesting work, paralegals used it 58% more and experimented 70% more than their peers. Stanford HAI’s 7 October 2026 account makes the result relevant to a practical question: what gives employees a reason to learn an unfamiliar tool?[1]

The underlying work predates that account. Stanford had already recognised the working paper by Arvind Karunakaran, Roshni Raveendhran and Tami Kim; Karunakaran describes a programme combining experiments with ethnographic fieldwork on meaningful tasks and skill development. The October article is a new explanation of that research, not evidence that the field study was just conducted.[2][4]

The enrichment group also received training, mentoring and protected exploration time. This observational comparison does not isolate the pitch. HAI does not give division headcounts, fieldwork dates or detailed measurement definitions.[1]

The opportunity has to change the working day

A promise about better work becomes concrete when it answers three questions. Which activity will become easier? What will the employee gain the opportunity to do? Who will help them develop that ability? A tool purchase addresses only part of that arrangement.

Take an illustrative operations role. Suppose an employee spends much of the week assembling recurring reports. AI could help prepare the material, while a manager gives that employee responsibility for investigating one persistent source of customer delays. The new task provides a reason to understand the underlying records, check the generated analysis and discuss the findings with colleagues. The value of learning becomes visible in a responsibility the person can actually exercise.

The same software could instead be introduced with an unchanged role and a higher reporting quota. Both choices might increase the number of reports produced. Only the first explicitly makes room for a broader skill. This is an illustrative design comparison, not an additional result from the law firm.

What “better work” can mean

There is a useful older framework for understanding this. In their 2001 paper on job crafting, Amy Wrzesniewski and Jane Dutton describe employees changing the activities they perform, the relationships around those activities and how they understand the purpose of their work. They also emphasise that opportunities depend on the surrounding job and organisation. The framework is conceptual; it did not test generative AI.[3]

Applied to an AI rollout, those three dimensions suggest different kinds of opportunity. A worker might take ownership of a new analysis, collaborate with a specialist they previously supported only indirectly, or gain a clearer view of the decision their work informs. These are changes in the experience and scope of a job, even when its title stays the same.

That is why a generic invitation to experiment may be insufficient. An employee needs a plausible destination for the experiment. A small, relevant responsibility with feedback can provide one; an open-ended instruction to use AI more often does not specify what success would look like.

A practical way to put the finding to work

For the employee, the corresponding proposal is specific: identify the repetitive work the tool could help with, then name the more valuable work the released time could support. For example, completing a monthly report might become an opportunity to explain one customer problem and recommend a tested improvement. The argument connects personal development with something the organisation needs.

Report an error or suggest a correction