The transformation happening in legal services is not the one the breathless predictions promised. Artificial intelligence has not made lawyers obsolete, nor has it democratized access to justice in any meaningful way. What it has done, quietly and consequentially, is redistribute the cognitive labor within law firms in ways that will reshape the profession for decades.
The work that once fell to junior associates and paralegals — document review, contract analysis, case law research, the grinding hours of reading that built the foundation of legal expertise — increasingly happens in seconds. A task that might have occupied a first-year associate for a week can now be completed, with reasonable accuracy, before lunch. The question is not whether this is happening, but what it means for how lawyers are made.
The apprenticeship problem
Law has always been an apprenticeship profession. Junior lawyers learned by doing the unglamorous work: reading thousands of documents in discovery, drafting routine motions, researching precedents that senior partners would later synthesize into strategy. The hours were brutal, but they served a pedagogical function. You cannot advise a client on contract risk if you have never read contracts at scale.
AI disrupts this model at its foundation. If the reading is done by machine, who develops the judgment that comes from having done the reading? Large firms are already grappling with this question, though few discuss it publicly. Some have begun rotating associates through "traditional" review projects specifically for training purposes — an artificial preservation of inefficiency in service of professional development.
Where the machines actually excel
The honest assessment of legal AI is more modest than the marketing suggests. Current systems are genuinely useful for pattern recognition across large document sets, for identifying relevant clauses in contracts, for summarizing depositions and case files. They are less reliable for nuanced legal reasoning, for understanding context that requires knowledge of a specific client's business, for the judgment calls that distinguish competent advice from excellent advice.
The technology works best, in other words, at precisely the tasks that once trained new lawyers. It struggles with the tasks that experienced lawyers perform. This is both reassuring and troubling — reassuring for partners who fear obsolescence, troubling for the pipeline that produces partners.
The access question
Proponents of legal AI often invoke democratization: if routine legal work becomes cheap, perhaps legal services become accessible to those who cannot afford them. The evidence so far is mixed. Consumer-facing legal AI tools exist, but they tend to address the simplest matters — basic wills, straightforward contracts, parking ticket disputes. The gap between what AI can handle and what most people actually need from a lawyer remains vast.
Meanwhile, the efficiency gains accrue primarily to large firms serving large clients. Corporate legal departments are pressuring outside counsel to use AI tools and pass along the savings. The result may be that sophisticated clients get cheaper, faster service while ordinary people see little change.
Our take
The legal profession is experiencing something more interesting than disruption: it is experiencing a slow-motion crisis of reproduction. The machines are good enough to eliminate the work that made new lawyers, but not good enough to replace the lawyers that work produced. Law schools continue to graduate students trained for a model of professional development that is quietly disappearing. The firms that figure out how to train lawyers in an AI-augmented environment will have a significant advantage. The rest will discover, in a decade or so, that they have a partnership track with no one qualified to walk it.
