Public Defenders Map Where AI Belongs in Their Casework

A new interview study asks 17 public defenders across the United States a question most AI-for-legal conversations skip: not whether the tools work on a benchmark, but where in the actual job they would help. The answers, laid out in a paper on arXiv by Inyoung Cheong, Patty Liu, Dominik Stammbach and Peter Henderson, are unusually specific about where practitioners think AI belongs and where it doesn't.

The authors map public defense work into five pillars: evidence investigation, legal research and writing, client communication and support, courtroom representation, and defense strategies. Interviewees consistently identified evidence investigation, such as reviewing large volumes of digital records, as the area with the greatest potential for AI support. Legal research and client communication got a more limited read. Courtroom representation and defense strategy were rated least compatible with AI, which reads less as technophobia and more as an honest map of where judgment and relationship are the actual product.

The barriers list is worth sitting with. The paper reports that AI adoption is constrained by costs, restrictive office norms, confidentiality risks, and unsatisfactory tool quality. Public defense is a chronically under-resourced part of the justice system, so a category of software pitched as a great equalizer bumps into the same procurement, culture, and privacy walls that everything else does. The safeguards practitioners asked for were mandatory human verification, limits on over-reliance, and preservation of the relational aspects of lawyering.

The honest caveat is that this is a qualitative study of 17 practitioners, not a measurement of how much time any AI tool actually saves on a real docket. The paper does not put dollar figures on the cost barrier or name specific vendors, and it doesn't say which of today's shipping products, if any, clear the practitioners' quality bar for digital evidence review.

The authors' proposed way forward is a research agenda built around open science, domain-specific datasets and evaluation, and incorporating frontline practitioners' perspectives into system development. For anyone building legal-tech into this market, the useful signal is that the buyers who need the productivity most are also the ones most explicit about which parts of their work they don't want automated.