Farnam Mohebi

I study how technology changes the way organizations divide professional work, and what that shows about expertise. In my dissertation, “Where to Draw the Line: AI, Delegation, and the Reorganization of Expert Work,” I ask how delegating peripheral tasks to AI affects professionals’ core work. My research is grounded in organizational ethnography, complemented by quantitative and computational analysis where the question calls for it. Papers and working papers are available on request: farnam.mohebi@berkeley.edu.

Job Market Paper

Where to Draw the Line: How Delegating Peripheral Tasks Disturbs Professionals’ Core Work

Solo-authored. Under review, Administrative Science Quarterly

Dissertation proposal: Winner, 2026 AI Dissertation Proposal Competition (Management category), Mays Business School, Texas A&M University

Nominee, Best Student Paper Award, 2026 Fisher AI in Business Conference, Fisher College of Business, The Ohio State University

Presented at AOM 2026 (OMT; abstract in AOM Proceedings), AOM Expertise Symposium 2026 (invited), ISA RC52 2026 (invited), UC Berkeley AI Leadership Accelerator 2026 (invited talk), ASA 2025, Organization Theory in Healthcare Association 2025, Duke Strategy PhD Conference 2026, and Bay Area Management PhD Conference 2026

Abstract

Organizations commonly delegate peripheral tasks to lower-cost workers or to AI so that professionals can focus on their core work, the work that requires the expertise only they have. I ask how delegating a peripheral task affects the core work it is meant to support, and I find that it can undermine it. In an 18-month ethnography, I studied radiation oncologists, whose core work is to treat cancer. Their work is a bundle of tasks, including outlining the tumor on patient images, which they consider a core task, and outlining the healthy organs nearby, which they consider a peripheral task. Because organ outlining was considered peripheral, the department brought in AI to produce outlines of the healthy organs. The AI outlines of the healthy organs did not always arrive, mainly because of how the software was set up, so I could compare the same physicians drawing organs and reviewing the AI outlines. Following Abbott (1988), I coded two kinds of judgment that physicians voiced or enacted about the patient: findings that changed what to treat, such as the tumor boundaries, and decisions that steered where the radiation dose went. Both came up more often when physicians drew the organs than when they reviewed the AI outlines. I theorize that doing some peripheral tasks takes professionals through the case details their core work depends on, and that this connection runs through the person doing both tasks. I call these connected peripheral tasks. Though the AI outlines may meet the department’s accuracy standards, physicians who review them lose occasions to enact their knowledge, which I call expertise unenactment. This study shows that when professionals shed peripheral tasks, they can shed part of their core work, and that delegation can undermine expert work before any deskilling.

Working Papers

Same Threat, Different Futures: Diverse Professional Responses to Anticipated AI Encroachment

With Ambar La Forgia and Eliza Brown. First author. Working paper

Presented at ASA 2026, AOM 2026 Health Innovation PDW (expert presentation), the NBER Digital Economics & AI Tutorial, the MIT Rising Scholar Conference, and the UCSB Health Innovation Working Group

Abstract

When technological change threatens a profession, it forces professionals to confront what makes their expertise indispensable: which tasks do they hope to continue to perform themselves and which do they hope to delegate to technology? AI is the latest disruption creating uncertainty about the future of work. Research suggests professionals will respond by defending their jurisdiction over specialized analytical work. However, within a profession, experts might not find the same work meaningful. Through 72 interviews with physicians, we found they diverged in what they sought to protect, scientific clinical reasoning versus healthcare provision, and how they sought to protect it, through personal execution versus oversight of technology. These dimensions produced four orientations: Augmentation (performing analytical work, with AI enhancing execution), Verification (ceding execution but overseeing AI's analytical outputs), Connection (delegating analytical tasks to deliver relational care), and Dissemination (overseeing AI's analytical outputs to focus on population health). Because professionals vary in what they find meaningful, they hold different visions of how to protect professional jurisdiction from technological encroachment.

Concede and Claim: What Professionals Keep as the Human Part of Their Work When Machines Take Their Tasks

Solo-authored. Manuscript in preparation

Abstract

As machines take over tasks in professional work, researchers have long asked what remains as the human part of the work. Sociology of automation has often answered with what machines cannot do, such as tacit knowledge or connection, a capacity that people have and machines lack. This view has merit, since machines tend to take well-defined tasks first, and professionals themselves often name these capacities as human. However, research on professions suggests that professionals also take part in defining the human part of their work, because they claim the right to their work, answer for its results and keep ties with the people they serve. I take up this tension in my study and ask how professionals define the human part of their work when machines take some of their tasks, and how definitions differ within one profession. I use interviews with 72 U.S. physicians in three groups whose tasks artificial intelligence (AI) takes differently, and 133 documents, most of them comment letters that physician organizations sent to federal regulators about AI. I found that even when physicians conceded capacity to the machine, they kept a part of the work as a position: most granted that AI could do a task as well as they could, yet nearly all claimed the decision on its output, and most claimed to answer for it. AI can neither examine a patient nor know one over years, so the capacity view would expect every physician to claim these as human. Yet physicians claimed them only where their work keeps ties to patients: physicians who follow patients over years most often claimed the relationship, image readers who examine patients claimed examining and being with them, and most readers who rarely see patients claimed neither. What physicians called human rested not only on what machines cannot do but also on who answers for the work and who stays tied to the client. Unlike a capacity, a position and a relation do not shrink as machines improve, and professionals, not only machines, shape them.

The Unargued Good: How Machines Bring Efficiency into Professional Work Without an Argument

Solo-authored. Manuscript in preparation

Abstract

Machines now enter professional work with values attached, and tools that write notes and read images are sold on the time they save. Sociologists of professions expect professionals to defend their own values when an outside value arrives, and physicians long resisted efficiency that payers and managers pressed on them. However, research on technology suggests that a machine can carry a value while those who want it stay out of view. I ask how professionals take up a value built into a machine, and when they contest it. Using interviews with 72 U.S. physicians about artificial intelligence (AI), physician organizations' letters to federal regulators, and residency standards, I find that physicians treated the efficiency AI promised as a virtue, a mark of good physicians and of care for waiting patients, and often called it good without saying why. They contested the same value as an injury when administrators, payers or schedules demanded it, and again when they pictured who would spend the time AI saved. Because the value went unargued, few asked whether the tools saved time at all. Whether professionals contest an outside value depends less on the value than on whether they can see who demands it.

The Digital Escalator: How Physicians on Social Media Navigate Competing Forms of Credibility

With Eliza Brown and Colby Fortin. Manuscript in preparation

Seventy-eight physician interviews paired with computational analysis of more than 1,300 physician-influencer accounts, on how physicians build, convert, and defend credibility as expertise becomes public content. Social media separates a physician’s personal credibility from the institution’s and opens a second route to standing. The physicians the institution served least can climb it.

Mundane Touchpoints: How Automating Scheduling Messages Erodes Nurses’ Knowledge, Authority, and Sense of Completion

With Julian Hong, Simone Zhang, and Haley Choe. First author. Early stage; data collection ongoing

An ethnography of an AI tool that routes a clinic’s scheduling messages away from nurses, using its staggered rollout across departments, paired with analysis of nurses’ notes in the electronic health record.

Papers at Interdisciplinary Journals

Papers from my clinical collaborations, at medical and health-informatics journals.

The Impact of AI on Clinical Decision Making in Healthcare Teams

With Eliza Brown, Michelle Zhao, Sarah J. Stephens, Yvonne M. Mowery, Manisha Palta, and Julian C. Hong. First author. Revise and resubmit, npj Health Systems (a Nature Portfolio journal)

A qualitative follow-up to a randomized trial of a predictive AI tool on hospital care teams. The tool increased authority without taking it from any role. Some team members held formal standing but lacked predictive expertise, others had the expertise but lacked standing, and the tool supplied what each group lacked. We call this latent authority: authority that belongs to a role but goes unused until a technology supplies what its holders lack.

Presented at UCSF Bakar Institute 2026 (invited), UCSF Computational Cancer Community 2026 (invited), and Research AI Day 2026 (poster)

From Moral Agents to System Clerks: The Hidden Cost of Ethical AI for Physicians

With Julia Adler-Milstein, Margaret Eby, and Julian C. Hong. First author. Comment, under review, The Lancet Digital Health

A Comment arguing that the ethics rules meant to make clinical AI safe can hollow out physicians’ moral agency, turning doctors who exercise judgment into clerks who sign off on a recommendation.

Invited: TETHICON 2026, Stanford University

From Data to Patient: How Clinical Information System Design Shapes Patient Knowledge Construction

With Keefer Chen, Ryzen Benson, and Julian C. Hong. Co-corresponding author. Under review, JAMIA Open

Think-aloud sessions with 14 oncology clinicians showing that re-presenting the same record as a patient timeline, rather than by originating department, changes what physicians see in a case: whose observations count, and whether the chart serves clinical reasoning or institutional compliance.

Public Writing

Rebooting One’s Professional Work: The Case of French Anesthesiologists Using Hypnosis

The ASQ Blog, April 2026

An author interview with Nishani Bourmault and Michel Anteby, conducted with Leticia Smith, on how physicians reinvent deeply technical work.

Enhancing Research Transparency Inspired by Grounded Theory

UC Berkeley D-Lab, April 2024

A methods essay arguing that transparency is not a checklist of shared data and code but an account of the whole research process, and that grounded theory’s discipline of documenting each analytic move can bring that kind of openness to quantitative work as well.

From Ideas to Streamlined Research: The Benefits of Full-Cycle Methodology

UC Berkeley D-Lab, December 2023

A practical guide to carrying a research idea from its first formulation to a finished output, treating each phase as a chance to revise the question rather than a step to be executed in order.

This research has been supported by nearly $40,000 in competitive grants and awards, including grants from the Berkeley Economy and Society Initiative, the Institute for Research on Labor and Employment, the Berkeley Center for Workplace Culture and Innovation, and the Fisher Center for Business Analytics at UC Berkeley.

Cartoons by Mark Hill.

Page updated

Google Sites

Report abuse