AI Didn’t Eat Law. It Fed It.
Picture a partner at 7:00 a.m., coffee going cold, eyes on a memo that did not exist last night. The firm’s model sifted the cases, assembled arguments, mapped likely counterpoints. What would have taken a week is now a morning read. The question is no longer “Can we afford to take this on?” but “How many more like this can we run?” When the cost of first drafts and first passes falls toward zero, the floor of “viable matters” drops with it. The ceiling, inconveniently, rises too.
Damien Charlotin’s argument in the Washington Post deserves to be read as a field report, not a provocation. Lawyer headcount is up. Law school seats are full. Recent grads are working. The data are whispering something unfashionable in a world that loves the clean decapitation story: law is not a stack of detachable chores, it is a braided practice. Pull on one thread with automation and the rest tightens.
The bundle that refuses to unbundle
Jobs are often described as bundles of tasks, and AI, we are told, excels at splitting bundles apart. That’s true for loose bundles: discrete, repeatable fragments you can peel off and ship to the machine. But much of lawyering is a tight bundle. Research shades into strategy; drafting forces value judgments; evaluation is entangled with client risk tolerance, opponent psychology, and the unwritten norms of a jurisdiction. When you automate the “easy” strands, the remaining ones do not simply shrink. They get tougher, more contextual, more consequential. The memo appears faster, which means the team can now consider three strategies instead of one, pressure-test each against fresh authority, and negotiate more aggressively because they know where the cliffs are. The unit of work expands as the unit cost falls.
This is why the fantasy of a lawyerless legal market keeps dissolving on contact with reality. The machine can fetch, summarize, and draft. It cannot decide, in the thick sense of decision that clients actually pay for: which trade-off to live with, which proof to chase, which regulatory hair to split, which silence in the record to lean on. Automating the prelude shifts weight onto the coda. Judgment accumulates value precisely because everything that precedes it has become cheap and abundant.
When cheap law makes more law
We have been here before. E-discovery did not wipe out document reviewers; it detonated the volume of discoverable material and institutionalized new workflows. The per-gigabyte price of insight crashed and the total spend climbed because parties could plausibly go deeper. That was Jevons in a pinstripe: cheaper capability drove broader use. Generative systems are widening the mouth of the funnel again. If you can explore more theories before lunch, you will. If a startup can get a credible compliance baseline in a day, it tries new markets. If a plaintiff’s firm can triage ten thousand potential claims with good precision, it files more, better cases. AI does not just accelerate answers; it lowers the activation energy of disputes and deals.
The downstream effects are not just tactical. Lower research and drafting costs alter who can buy what kind of legal help. Mid-market companies that used to ration counsel can now ask for scenario analysis, policy comparisons, and playbooked negotiations. Governments that struggled to keep up with industry filings can run their own models to vet and shape rules. The pie does not stay the same size when knives get sharper.
The junior lawyer, remixed
Entry-level work will mutate first. The classic apprenticeship—hours of cite-checking and first drafts—was a training tax paid by clients and subsidized by margins. Models swallow parts of that instantly. Yet firms still hire juniors because they need surge capacity, because clients read team size as commitment and coverage, and because no partnership survives without a pipeline. The tasks change: less mechanical extraction, more model supervision; less rote drafting, more adversarial testing of the machine’s claims; less lonely memo-writing, more orchestrating multi-tool workflows that stitch retrieval, synthesis, and human judgment into something auditable.
If that sounds like engineering, it is, but of a legal kind. The craft migrates from typing toward system design: which authorities to privilege, which facts to feed, how to build defensible prompts, when to distrust a fluent answer, how to capture rationale so a regulator—or a jury—can follow it. Apprenticeship shifts from stamina to calibration. The skill ceiling rises, not falls.
Billing the machine
The economics will not sit still either. If an hour swells into a day’s worth of output, the hour stops being a meaningful unit. Expect more fixed fees, subscriptions, and productized services—brief banks with living commentary, compliance monitors that emit alerts with annotated reasoning, negotiation copilots tuned to a client’s risk posture. Firms that treat their models as billable colleagues will lose; firms that treat them as infrastructure will find new margins in reuse, versioning, and speed.
For clients, the experience improves and complicates at once. Turnarounds compress. Coverage broadens. But the volume of “what we could do” expands faster than budgets, forcing explicit prioritization. The firms that win will not be those boasting the biggest model, but those that help clients choose which problems to solve, and which to ignore, with a paper trail that survives cross-examination.
Regulation begets representation
Even if AI destroys jobs elsewhere, law tends to swell when the rulebook changes and the stakes climb. Every model that touches consumers, safety, finance, health, labor, or the environment drags new rulemaking behind it and litigation beside it. Agencies will write and revise. Companies will comment and comply. Cities will improvise. Plaintiffs will discover new theories of harm, and defendants new defenses. This is expansionary terrain for a profession that translates uncertainty into structured commitments.
The profession’s guardrails will tighten too. Malpractice carriers will demand logs and supervision protocols. Bars will clarify what counts as competent use of AI and where delegation crosses into unauthorized practice. Privilege will be tested by chat windows. Discovery will include prompts. The firms that thrive will treat model governance as part of professional responsibility, not a marketing slide.
The quiet conclusion
If you came for a clean before-and-after, the legal market refuses to provide one. The story is messier and, for lawyers, more optimistic than the headline doom loops predict. AI is stripping cost and latency out of the front end of legal work and pouring complexity and consequence into the back end. That tends to create more law, not less; more lawyers, not fewer; different ladders to climb, not no ladders at all.
On this site we track replacements. Here, the replacement is subtle. AI is replacing the excuse to do less law. It is replacing the waiting room with a war room. The firms that understand that—who convert speed into scope without losing rigor—will pull away. The rest will discover that the only thing worse than being replaced by a machine is being outflanked by a human who learned to use one.