“Mere Verification”

I am concerned with a trend in the legal community: the notion that it suffices to merely verify the output of agentic systems, ensuring no hallucinations, errors, or material omissions. Is a brief that is technically correct necessarily rendered with requisite diligence? I argue no.

A risk-tiered approach is warranted. In highly formulaic, low-stakes contexts, verification of accuracy might suffice. But in high-stakes, complex, and bespoke matters, diligence requires more than basic correctness.

Consider dispositive briefing in litigation involving tens or hundreds of millions at issue: every word in a 30-page document matters. The phrasing of legal propositions, the framing and priority of cited facts, the selection of pincite ranges, etc.—these are not binary right/wrong choices. Their effectiveness exists on a spectrum of better or worse across persuasion, risk allocation, precedent weight, jurisdictional nuance, and client objectives. Agentic AI, however advanced, cannot fully replicate this layered professional judgment, at least according to today’s rules.

True diligence demands the lawyer’s active involvement at multiple stages, applying legal judgment to each material detail—shaping prompts, assessing intermediate outputs, characterizing propositions, prioritizing facts, and refining strategic framing. Anything less risks facilitating the unauthorized practice of law by nonlawyer technology companies.

At every point where legal judgment is required, a lawyer must be the one applying it. In high-value, high-complexity postures, where 30 pages can determine multimillion-dollar outcomes, studied human judgment remains indispensable, line by line.