"operating on AI."

Law firms use AI for legal research (78% adoption), document drafting (65%), document review and e-discovery (45%), and client intake and communications (30%). These aren't experimental pilots anymore — they're daily production workflows at firms ranging from solo practices to the Am Law 200.

The shift from 2024 to 2026 has been from "trying AI" to "operating on AI." The firms leading adoption aren't just using AI tools — they've restructured workflows, reassigned roles, and built measurement systems around AI-assisted production. Here's what that actually looks like in practice.

Legal Research: 78% Adoption

Research is the most widely adopted AI use case because the ROI is the most obvious. What firms actually do: attorneys describe a legal question to CoCounsel, vLex Vincent, or Claude and receive a research memo with cited authorities in minutes instead of hours. The AI searches case law, identifies relevant statutes, and synthesizes analysis across jurisdictions. Real workflow example: a litigation associate at a mid-size firm receives a motion to dismiss. Instead of spending 6 hours researching the standard of review and relevant precedent, they prompt CoCounsel with the specific legal issues. The AI produces a cited research memo in 12 minutes. The associate spends 90 minutes verifying citations and refining analysis. Total time: 2 hours instead of 6. The verification step is non-negotiable. AI research tools hallucinate less than they did in 2024 (CoCounsel reports 95%+ accuracy), but "less" isn't "never." Every citation gets checked against the primary source.

Document Drafting: 65% Adoption

Drafting is the fastest-growing AI use case because the time savings are immediate and visible. What firms actually do: attorneys use Claude, Harvey, or ChatGPT to generate first drafts of briefs, motions, contracts, demand letters, client correspondence, and internal memos. The AI produces a structured draft; the attorney edits, refines, and finalizes. Real workflow example: a family law attorney needs to draft a property division motion. They provide Claude with the case facts, applicable statute, and desired arguments. Claude generates a 12-page first draft in 3 minutes. The attorney spends 45 minutes editing — adding case-specific details, adjusting tone, and verifying legal standards. Total time: 50 minutes instead of 3 hours. The quality gap between tools matters here. Claude consistently produces the cleanest legal prose with the best structure. ChatGPT is strong for shorter documents and client communications. Harvey's drafting agents, trained on firm templates, produce the most firm-consistent output at enterprise scale.

Document Review and E-Discovery: 45% Adoption

AI-powered document review has been used since 2012 (predictive coding/TAR), but generative AI has expanded what's possible. What firms actually do: litigation teams use Relativity, Everlaw, or Reveal to run TAR 2.0 (Continuous Active Learning) on document collections, automatically classifying millions of documents as relevant or non-relevant. Newer AI features identify privilege, extract key facts, and generate document summaries. Real workflow example: a commercial litigation matter produces 3.2 million documents in discovery. Instead of 40 contract reviewers spending 6 months on linear review ($4.8M), the firm deploys TAR 2.0 with a senior associate training the model. The AI classifies the collection in 3 weeks, with human review of the top 200,000 documents and quality-control sampling of the rest. Total cost: $800K instead of $4.8M. Recall rate: 82% vs. the estimated 60% for human-only review.

Client Intake and Communications: 30% Adoption

Client-facing AI is the newest adoption frontier and the most sensitive. What firms actually do: chatbots and AI-powered intake forms on firm websites qualify leads and collect case information 24/7. AI generates personalized follow-up emails, appointment confirmations, and case status updates. Some firms use AI to draft initial case assessments from intake data. Real workflow example: a personal injury firm's website uses an AI-powered intake form that asks dynamic follow-up questions based on the injury type. At 2 AM, a potential client describes a car accident. The AI intake system collects accident details, insurance information, and medical treatment status — then routes the qualified lead to the appropriate attorney with a pre-drafted case summary by morning. The ethical line: AI can collect information and draft communications, but it cannot provide legal advice to potential clients. Intake AI must clearly disclose that it's not an attorney and that no attorney-client relationship exists until a human lawyer engages.

The Adoption Gap: What Separates Leaders from Laggards

The 78% research adoption number hides a significant disparity. Leading firms (Am Law 200, tech-forward mid-size): use AI across 3-4 workflows, have formal AI policies, measure ROI, and iterate monthly. Their attorneys save 10-15 hours per week on production work. Middle-of-the-pack firms: use 1-2 AI tools informally. No policy, no measurement, no firm-wide adoption. Individual attorneys experiment but the firm doesn't operationalize. Savings: 2-5 hours per week, inconsistently. Laggard firms: no AI adoption, no policy, active resistance. These firms lose competitive position monthly as AI-enabled competitors produce more work at lower cost. The gap isn't about technology — it's about management commitment. The firms winning with AI have managing partners who mandated adoption, funded training, and built measurement into the workflow. The firms struggling have managing partners who said "use it if you want" and walked away.

The Bottom Line: Law firms in 2026 use AI for research (78%), drafting (65%), document review (45%), and client intake (30%). The leaders save 10-15 hours per attorney per week. The difference isn't which tools they use — it's whether management committed to structured adoption with policies, training, and measurement.