My research examines how information systems shape the work of knowledge workers, and thereby the value organizations create, as that work grows more complicated, less predictable, and increasingly delegated to generative and agentic AI. Three questions organize it:

How deterministic enterprise and AI systems differ in what they do to knowledge work. Drawing on complexity science, I separate complicatedness, the interconnected components a worker carries, from the complexity that can emerge among them. The assumption I question is that value comes from imposing structure, since complicatedness brings valuable variety as well as demands, and requisite variety holds that varied work calls for varied responses. I ask what imposed structure does for a worker carrying a complicated portfolio, and what generative AI does in its place.

How innovation and reliability trade off against one another. Complexity science treats continuous feature innovation as an accumulation of components and interdependencies in platforms whose reliability people already depend on, and IT failure as what emerges from it. The assumption I question is that IT failures destroy value, since the workers who meet a disruption cope with it and adapt around it. I ask when these failures destroy value, and when they prove serendipitous.

How work environments should be designed for agentic workflows to add value. Design science treats the work environment as the object to be built and evaluated. Knowledge workers delegate search, reading and composition to agentic AI, and the passage a claim rests on sits inside a session that closes. The assumption I question is that evidentiary standing is settled at composition. I contributed to a framework and a system for durable provenance, and I propose that a lasting address for each passage makes verification affordable, curbs sycophancy, and lets delegated knowledge work create value.

I established the industry relationship behind this work and continue to maintain it directly: a partnership with a U.S. insurance CRM platform, with access to its proprietary system, clickstream-trace and real user monitoring data. I study these records through quasi-experimental designs, and the partnership supports survey experiments, positioning the work I plan next to move from observation to intervention. I also work with the co-founders of an AI startup on the design of agentic work environments. I see building and sustaining partnerships of this kind as part of the research itself.