The Review Queue Needs an Admission Policy

When drafting becomes cheap, the scarce resource is the reviewer's willingness to read. A firm that lets every generated alternative enter the partner's queue can become slower while every associate becomes faster. The first remedy should be to decide what deserves admission to that queue.

Why is final review the wrong place to make decisions?

Review is often treated as a final inspection of finished prose. That is an expensive place to discover that no one chose the theory, resolved the factual assumption, or decided which alternatives mattered. AI makes it easier to defer those choices by producing a complete-looking document for each possibility. The partner receives pages that represent unresolved decisions, disguised as completed work.

What did the court call AI slop?

A recent Florida appellate order shows this problem from the other end of the review process. In Lisandrillo v. Palozzi, No. 4D2026-2262 (Fla. 4th DCA Sept. 16, 2026), the Fourth District Court of Appeal addressed what it called "AI slop": lengthy, unfocused filings of convoluted and frivolous arguments with suspected AI-generated components, a problem it expressly distinguished from hallucinated law or false citations. The court ordered counsel to show cause why sanctions should not be imposed. Whether AI produced the filings was not the point; counsel remained responsible for independent professional judgment. The lesson for workflow design is narrower: accuracy is not enough if cheap production lets unfinished reasoning reach the court in polished form.

What should a document need before it enters review?

More drafts do not make the reviewer faster. So manage unfinished work. Before a document enters review, its owner should identify the decision requested, the material questions still open, and the sources needed to evaluate them. A second version must offer a consequential alternative, not the same sentences rearranged.

This changes what AI should produce. Measure the product by how much it reduces the effort of deciding, not by how much it puts in front of the decision-maker.

How should review scale with risk?

Routing helps, but it creates its own judgment problem. A model's confidence is no substitute for the consequences of error. A deadline or disclosure that looks routine may deserve more scrutiny than a difficult but reversible stylistic choice. Scale review to materiality, detectability, and reversibility, and spot-check lower-risk work so systematic failures do not disappear into the fast lane.

Does fast-track review let partners stop teaching?

Smarter AI that reduces choices and fast-tracks the decision queue should not let partners avoid teaching. Supervisors still owe reasonable supervisory efforts under applicable professional rules (see ABA Model Rule 5.1). Some reviews should deliberately build a junior's judgment. Those need time set aside for explanation instead of being buried in a queue built for fast approval.

What should firms measure instead?

Measure time to a usable decision and total correction effort, not draft turnaround. AI can multiply drafts indefinitely. A competent organization needs a reason for each one it asks an attorney to read.