Running on Enthusiasm

A recent study followed employees at a law firm and an academic medical center as they worked on AI applications for their colleagues. Both organizations offered training and a secure environment for experimentation. Over two years, researchers watched what happened as employees tried to turn their early ideas into something the rest of the organization could use.

At the law firm, identified as LegalCo, the work became increasingly difficult to fit around existing responsibilities. More than 80% of the participating experts eventually stepped away. At the medical center, NE Health, participation continued with ongoing support and recognition for the work. Researchers Arvind Karunakaran, Katherine Kellogg, and Batia Wiesenfeld examined how those different experiences developed, and what they meant for the organizations’ AI efforts.

In both organizations, employees were being asked to follow through on their experiments. A promising application needed more testing and review with colleagues before it was ready for wider use. For the people involved, this meant taking on a continuing responsibility alongside their existing work.

NE Health gave employees ongoing technical support and clear criteria for judging whether their AI applications were working well. Their contributions also counted toward promotions and other career opportunities. LegalCo offered training at the start, but little ongoing help, and employees’ AI work often wasn’t reflected in performance reviews or pay. Over time, participation at the law firm declined while the medical center’s effort continued to grow. MIT Sloan reported 141 organization-wide AI applications in use at NE Health, compared with three at LegalCo.

The gap between 141 applications and three is striking, though it doesn’t tell us how valuable those applications were. With only two organizations in the study, there’s also a limit to how much the comparison explains. What the researchers describe in more detail is how employees kept participating when the work was supported, and gradually stepped away when it became too much to sustain.

It’s reasonable to let people try a few ideas before committing time and money to a project. Once something looks worth developing, though, their manager needs to make room for it in the workweek. Otherwise, the project continues as an extra assignment, even as more people come to depend on it.

Enthusiasm makes that arrangement easier to sustain for a while. Someone who wants a project to succeed may continue to make time for it without asking their manager what else should wait. From the sponsor’s perspective, progress looks encouraging. The extra hours are harder to see when the employee’s regular work is still getting done.

That’s why I’d be cautious about treating early participation as evidence that an AI program has the support it needs. Before expanding a promising pilot, sponsors need a realistic account of the effort behind it. If continuing the project means postponing other work, that tradeoff deserves an explicit decision. If nobody’s willing to decide what should wait, the employee shouldn’t be left to figure it out.

It’d be a shame to respond to falling participation with another campaign to get people excited about AI. Some of the people who’ve stepped back may have a very good idea of what it would take to finish what they started. They may also have learned that agreeing to continue means accepting work they no longer have time to do.

Algorithm and Blues publishes Sundays.

Research

Karunakaran, A., Kellogg, K. C., and Wiesenfeld, B. M. (2026). Experimentalist Intensification Governance: Managing Worker Negative Consequences Associated with Generative AI Innovation Work. MIT Sloan Working Paper 7382-26, posted July 2026.

For the participation and application counts, see MIT Sloan’s account of the study, September 9, 2026.