The legal profession has always been a pyramid scheme of sorts—not in the fraudulent sense, but in its architecture. At the base, armies of junior associates and paralegals perform the grinding work of document review, contract analysis, and legal research. They bill hundreds of hours, learn the craft, and eventually some climb toward partnership. This structure has persisted for a century. It is now being quietly dismantled.
Large language models can now perform in minutes what once required days of billable paralegal time. Contract review, due diligence summaries, initial research memoranda, deposition preparation—these tasks that once formed the foundation of legal training are increasingly handled by AI systems that never tire, never bill overtime, and never make partner.
The economics are brutal
A mid-sized law firm might employ a dozen paralegals at salaries ranging from modest to comfortable, plus benefits, office space, and training costs. An enterprise AI subscription costs a fraction of a single salary. The math is not subtle. Major firms have already begun reducing paralegal headcount, though few advertise this publicly. The reductions are framed as "efficiency improvements" or "restructuring," but the pattern is unmistakable.
More troubling for the profession's long-term health is what happens to the pipeline. Junior associates historically learned by doing tedious work—reading thousands of contracts teaches you what a problematic clause looks like. If AI handles the volume work, how do young lawyers develop judgment? The profession has not answered this question, largely because acknowledging it would require confronting uncomfortable truths about the business model.
The work that remains
AI cannot yet appear in court, negotiate across a table, or counsel a client through a divorce. It cannot exercise the judgment that comes from understanding a particular judge's temperament or a jurisdiction's unwritten norms. The human elements of legal practice—persuasion, empathy, strategic thinking under uncertainty—remain stubbornly human.
But these skills were traditionally developed through years of grunt work. The partner who can read a room learned that skill while reviewing her ten-thousandth document. Strip away the apprenticeship phase, and you may produce lawyers who can prompt an AI effectively but cannot think independently when the AI fails or the situation is genuinely novel.
Our take
The legal profession is conducting an uncontrolled experiment on its own future. By eliminating the entry-level work that trained generations of lawyers, firms are optimizing for quarterly profits while potentially hollowing out the expertise pipeline. This is not unique to law—medicine, accounting, and journalism face similar pressures—but law's explicit apprenticeship model makes the stakes unusually clear. The partners making these decisions will be retired before the consequences fully manifest. Their successors may find they have inherited a profession that forgot how to train its own.
