The legal AI delegation problem.

Most law firms treat AI adoption as an individual skill gap, assuming that if attorneys just learn better prompting, efficiency will follow. That approach usually fails.

In their research on AI fluency, professors Rick Dakan and Joseph Feller break down human-AI interaction into four competencies (known as the four Ds): delegation, description, discernment, and diligence. Three of these are operational habits that lawyers can build with practice. They learn to write clearer instructions (description), check output against primary sources (discernment), and stay within security guidelines (diligence).

Delegation is different. Delegation determines which aspects or steps of a legal matter go to a software model and which stay with an attorney, and how the two should pass work back and forth. These are decisions that ideally should be made well before a specific matter appears.

Delegation as judgment

When delegation is left to individual discretion, firms confront two predictable failure modes. Some lawyers hand off tasks that require deep context or nuanced legal judgment, risking quality and confidentiality. Others insist on manually drafting routine documents that a model could handle in seconds, erasing any potential efficiency gain.

Firms cannot resolve this with a single, off-the-shelf policy. The right boundary line depends entirely on how the firm operates. A boutique litigation group with a higher risk tolerance requires a different workflow than an M&A practice that relies on routine due diligence to train first-year associates. Matter types, partner compensation, and talent development all dictate where, and how much, human oversight should remain part of the process. Two firms using the same tools should use them differently, drawing the line for delegation in different places.

The limits of AI

Software tools will continue to improve at description, discernment, and diligence. Models will become easier to instruct, better at spotting their own mistakes, and safer to use out of the box. But no software can decide the core operational question - how a firm ought to run its practice. Delegation must be an institutional choice and a human decision, and getting it right at the organizational level is what separates real efficiency from the firms that never see the true benefit of their AI tools.

← Back to insights Read the full four-Ds framework →