Do You Still Need a Lawyer?
Michael McCready was still in his kitchen, coffee not yet poured, when the newsletter hit. Anthropic had released a legal plugin for Claude. He read the headline twice. By the time he reached his desk, Slack had twenty-seven unread messages.
By 7 a.m., his screen showed RELX down 8 percent.
Thomson Reuters negative 10.
By 8 a.m., RELX was down 14 percent.
By noon, $285 billion had vanished from the sector.
McCready had spent two decades building McCready Law in Chicago, the kind of solo practice where you either grow or disappear. He’d bought into LexisNexis tools and adopted contract review software.
His phone buzzed with messages from law firm partners, all asking some version of the same question: what happens now? McCready later told reporters what he’d been thinking that morning: “This is not a matter of people fooling around with ChatGPT or asking queries; this is actual agentic AI built specifically for law and built specifically for certain tasks.”
The market had just priced that reality in real time.
When $285 Billion Vanished Before Lunch
The stock tickers told the story. RELX fell 14 percent, its worst single day since 1988. Thomson Reuters dropped 16 to 18 percent. Wolters Kluwer sank 13 percent. The Goldman Sachs software basket fell 6 percent, the steepest decline since April’s tariff shock. The financial services index lost roughly 7 percent.
Total damage: a $285 billion rout across software and financial services.
Any company selling knowledge work automation faced sudden scrutiny. Investors were rethinking how much of the legal market could be automated. If Claude could review contracts at software subscription costs instead of law firm hourly rates, the economics of knowledge work had just shifted.
The irony was immediate.
Artificial Lawyer noted that “Claude’s legal plug-in has nothing to do with legal research, which is the core value proposition and wide-moat foundation of Thomson and RELX’s legal businesses.” Thomson Reuters gets 45 percent of its profit from Westlaw, its legal research database. RELX derives only 10 to 13 percent from legal operations. Wolters Kluwer is the same.
“This is not a matter of people fooling around with ChatGPT or asking queries; this is actual agentic AI built specifically for law and built specifically for certain tasks.” - Michael McCready
All three fell between 13 and 18 percent anyway.
The market was not pricing current disruption. It was pricing trajectory. If Claude could handle contract review today, what surfaces next? What capability comes in version 4.7? The selloff showed investors believed AI was improving faster than law firms could adapt. The threat was not to lawyers themselves, but to the high prices firms could charge.
And the selling did not stop on February 3. Ten days later, RELX was down 26 percent from its pre-announcement price, nearly double the initial drop.
The Plugin That Changed the Math
The legal plugin handles specific, high-volume tasks that used to consume paralegal time. Contract review and risk flagging. NDA triage against standard templates. Document summarization for depositions and case files. Compliance checking against regulatory standards. Nothing it does requires a law degree, but all of it used to require billable hours.
Forrester Research, partnering with LexisNexis, found that paralegals saved 50 percent of their time on administrative work when using AI tools. Routine document review that took four hours dropped to two. The gains were real and immediate.
Claude Opus 4.6 achieved 90.2 percent accuracy on legal reasoning benchmarks, the kind of performance that gets quoted in earnings calls. The 1-million-token context window means it can ingest an entire contract, all related NDAs, prior correspondence, and case precedents in a single request. What once required uploading documents in chunks and cross-referencing tabs now happens in one pass.
The economic disruption was surgical. Anthropic open-sourced the plugins. Harvey AI, the legal tech startup RELX acquired in 2024, charges $1,200 per seat per year. Claude operates on standard subscription pricing. The legal plugin is included in enterprise access, not sold as a separate premium product. This turned legal work that LexisNexis and Thomson Reuters had sold at premium prices into a low-cost commodity.
But the plugin comes with one line that matters. Anthropic’s documentation states: “The plugin assists with legal workflows and does not provide legal advice. AI-generated analysis must be reviewed by licensed lawyers before being relied upon for legal decisions.”
That disclaimer is not boilerplate. It is an admission of risk the market had not yet priced in.
The 69% Problem
Stanford researchers quantified the risk. Large language models make up false answers 69 to 88 percent of the time when asked legal questions: fabricated cases, misstated statutes, invented precedents that read like real law but are not. The 90.2 percent benchmark measures structured legal reasoning. Real-world queries are messier. A lawyer asks a question no benchmark anticipated, and the model invents a confident, authoritative, completely wrong answer.
Courts have begun responding with sanctions. In February 2026, a Kansas federal judge imposed $1,000 to $5,000 fines per attorney for submitting fake citations generated by AI. In 2025, an attorney in New York faced a $15,000 personal sanction for hallucinated briefs. Another case required $26,100 in court reimbursements and $5,000 in opposing counsel fees.
The sanctions create a floor of human demand. Someone must check the AI’s work. That someone needs a law degree and liability insurance.
And the liability question is still open. Some malpractice insurers have begun explicitly excluding AI-related claims from coverage. If a lawyer relies on AI output that turns out to be wrong, who carries the loss? Not Anthropic, whose disclaimer says it does not provide legal advice. Not the insurer, if the policy excludes AI errors. The lawyer, alone, absorbs the risk of a tool they did not build and cannot fully verify.
Beyond error rates, entire categories of legal work resist automation. AI cannot appear in court. It cannot read the pause, the tone shift, the body language that tells a lawyer whether opposing counsel will budge on a settlement term. It cannot advise a client on risk appetite or warn them that a deal aligns with the contract but not with their strategic position. It cannot assume ethical responsibility for the advice it gives.
Bloomberg Law captured this: trust is “an ineffable amalgam of expertise, experience, and emotional intelligence.” The plugin has none of these.
The adoption data supports this tension. The American Bar Association reported in late 2025 that AI had moved from experiment to infrastructure. Adoption reached 79 percent of law firms, up from 19 percent two years earlier. Over 30 states have AI-specific guidance for attorneys.
Yet MIT Technology Review found in December 2025 that the legal sector had not experienced major layoffs despite this surge. Roles are changing. Work is not disappearing.
Michael Bennett, law professor at UIC, said it clearly: “The most expert legal practitioners and advisors who have deep skill are going to benefit from this in the short term.”
The benefit flows to those who can use AI. Not those who get replaced by it.
Two Markets, One Profession
The market panic on February 3 was overblown about replacement. It was accurate about division.
Routine work that can be standardized is leaving the profession. Harvey charges $1,200 per seat per year for contract review as a standalone product. Claude includes it as one feature inside a general enterprise subscription.
The legal work is the same. The pricing model is not.
When contract analysis stops being a premium product and becomes a bundled feature, the revenue model underneath it collapses. That work still happens. It just stops generating law firm fees. In-house legal teams will handle it with AI assistance, and the $300/hour outside counsel call becomes a last resort instead of a default.
This creates two tiers. Elite lawyers with specialized expertise and client relationships will thrive. Their work amplifies, letting them handle more matters with deeper analysis. Junior associates who spent years on document review face a shrinking career path. The pipeline that fed law firms for fifty years (hire at 25, make partner at 45) is breaking.
That pipeline matters more than it looks. Document review is not just labor. It is how lawyers learn. A second-year associate reading five hundred contracts does not just review them. She starts to notice which indemnification clauses blow up, which vendors walk deals, which patterns signal trouble three months before it arrives. That pattern recognition is what makes a senior lawyer worth $800 an hour. If AI handles the reps, the training ground disappears. The profession keeps its experts but loses the mechanism that produced them. Nobody has an answer for what replaces it.
Every wave of legal technology, from Westlaw in the 1970s to email in the 1990s, expanded the work rather than eliminating it. But those tools automated lookup and communication. Claude’s plugin does actual analytical work: flagging risks, suggesting alternative language, comparing clauses across precedents. That crosses a line previous technologies never reached.
The legal tech market reached $26.7 billion in 2024, projected to hit $46.8 billion by 2030. Law firm tech spending rose 9.7 percent in 2025. Money is flowing into legal AI, not away from it. The profession is not dying. It is splitting into two professions wearing the same name.
McCready, Two Weeks Later
McCready pulled up the legal plugin in mid-February. Less than two weeks after his stocks crashed, the tool that spooked the market sat in his browser. He ran a vendor agreement through it. The AI flagged three risky indemnification clauses and noted an inconsistent liability cap. It suggested alternative language. Four hours of work compressed to twenty minutes.
He read the output. The AI had treated a force majeure clause as standard boilerplate, but McCready knew this particular vendor had used that exact language to walk away from a deal two years ago. The clause was technically fine. Strategically, it was a trap. He picked up the phone.
The negotiation happened the way it always did. Back and forth. Give and take. Reading the pause on the other end of the line. Deciding which risk to accept and which to push back on. The AI had done the mechanical work. The judgment remained his.
What the February 3 selloff got right: the economics of legal operations are broken. Document-review shops will not survive pricing at software margins. What it missed: the lawyer did not disappear. He changed what he does.
McCready was not replaced by the tool that panicked the market. He was reorganized by it. The $285 billion selloff was a misreading of the profession. AI does not replace lawyers. It sorts them.
Sources: