Legal AI went mainstream in 2026. What a law firm should automate first.
Everlaw’s 2026 report found that 69% of legal professionals now use generative AI at work. The surveys agree on where: research (about 80% of AI users), document review (74%), summarisation (73%), memo and brief drafting (59%). Agentic tools — multi-step, acting across systems — are still in under a fifth of firms, though about half are considering them.
Start where the hours are high and the output is easy to review: summaries, first-pass review, research with verified citations. Then drafting from your own precedent. Then intake and time capture. Leave strategy and judgment with people.
Step one: summaries and review
Deposition summaries, first-pass document review, data-room reading. A person always checks the output, mistakes are easy to spot, and the hours are large — a 300-page transcript is a day of paralegal time.
Step two: research and drafting
High value, higher risk. Do it only with retrieval-first research and verified citations, and with drafting that works from your firm’s own precedent and style.
Step three: the front and back office
Client intake answered in minutes instead of the next business day. Time captured from activity rather than memory. Neither touches legal judgment; both touch revenue.
What to leave alone
Case strategy, settlement advice, anything a client is paying for a partner’s judgment on. AI can prepare the material. It should not make the call.
A worked example: a 40-lawyer litigation firm
Quarter one: deposition summaries and first-pass document review — about 170 paralegal hours a month back, with every summary linked to page and line. Quarter two: research with verified citations and drafting from the firm’s own briefs. Quarter three: client intake answered in minutes and time captured from activity. Each step pays for the next, and none of them puts a client outcome in the hands of software.
How to pick your first workflow
Choose one with three properties: high hours, output a lawyer checks anyway, and a number you can measure before and after. Deposition summaries usually win on all three, which is why we often start there.
Common mistakes
Buying a platform before a workflow. Licences for every lawyer, with no workflow owner, become shelfware.
Measuring adoption instead of hours. Logins are not results. Track hours saved and quality on real matters.
Eight legal agents, shown working — research, review, depositions, drafting, intake, diligence and time capture — each with before-and-after numbers, price and delivery time.
Frequently asked questions
What do law firms use AI for most in 2026?
Surveys put legal research first (around 80% of AI users), followed by document review (74%), summarisation (73%) and memo or brief drafting (59%).
What should a law firm automate first with AI?
Start with high-hour work that lawyers check anyway — deposition summaries, first-pass review and research with verified citations — then move to drafting, intake and time capture.
What should law firms not automate?
Case strategy, settlement advice and other judgment calls clients pay a lawyer for. AI can prepare the material; a lawyer should decide.
Sources: Everlaw 2026 Legal AI Adoption & Impact Report; 2026 legal AI statistics.