AI agents for law firms are software workers that run the operations around legal work: answering new enquiries day and night, gathering the details for a conflict check, collecting documents from clients, telling clients where their matter stands, keeping deadlines in front of the right lawyer, sorting the inbox and capturing billable time. They don't give legal advice, sign anything or decide whether to take a case. For a firm of 5 to 100 people, a first agent usually costs $5,000 to $15,000 to build, runs for a few hundred dollars a month and goes live in three to six weeks.
This post is part of our guide to rolling out agents, which starts at wizeb.com/blog/how-to-implement-ai-agents-in-business. If you haven't picked a starting point yet, read wizeb.com/blog/first-ai-agent-for-business alongside this one.
The Problem: The Work Around the Work
Most of the talk about AI in law is about research and drafting, and tools such as Harvey, CoCounsel and Lexis+ AI now cover a lot of that. But ask a managing partner where the week goes and the answer is usually somewhere else. A potential client calls at 7pm about a car accident, gets voicemail and signs with the firm that picked up. A paralegal spends Monday chasing bank statements for three divorce files. Reception answers "any news on my case?" a dozen times a day. Associates reconstruct Friday's time entries from their sent folder.
None of that needs a law degree, and most of it lands on people whose time you bill at $150 to $500 an hour, or who should be doing work you can bill. That gap, between who is doing the work and who needs to, is where agents earn their keep in a law firm.
7 Law Firm Jobs an AI Agent Can Take Over
- 1Intake and lead response. A web chat, SMS or phone agent answers every new enquiry within a minute, at any hour. It asks your intake questions for that practice area (what happened, when, where, who else is involved, any deadlines already running), screens out matters you don't handle, and books a consultation into the right lawyer's calendar. Everything goes into your practice management or intake system, so the lawyer opens the consult with notes instead of a name and a phone number.
- 2Conflict check prep. The agent pulls every party name from the intake (the client, the other side, related companies, spouses, insurers), normalises spellings and runs them against your conflicts database. It hands a lawyer a short report of possible hits. The lawyer clears the conflict; the agent only does the searching and the paperwork.
- 3Client document collection. Once a matter opens, the agent sends each client a personal list of what the file needs: medical records authorisations for a personal injury case, financial disclosure for a divorce, passports and certificates for an immigration petition. It follows up on a schedule, checks uploads are the right document and the right dates, and tells the paralegal when the file is complete.
- 4Matter status updates. Clients want to know what is happening, and failing to keep them informed is one of the most common sources of bar complaints. An agent answers "where is my case?" from the matter record in plain language: documents received, waiting on the other side, hearing date set. Anything that asks what the client should do next goes to the lawyer.
- 5Deadline and docket reminders. Your docketing staff still enter and verify court deadlines. The agent reads the calendar and makes sure nobody misses one: it reminds the responsible lawyer at set intervals, flags matters with no activity for 30 days, and escalates anything due within a week that has no draft in the file.
- 6Inbox triage and reply drafts. The agent sorts incoming email by matter and intent (client update, opposing counsel, court notice, new enquiry, billing) and files it to the right matter. It drafts routine replies, such as appointment confirmations or document requests, for a person to approve. For the general pattern, see wizeb.com/blog/ai-customer-service-agent.
- 7Time capture and billing follow-up. From calendars, sent email and document activity, the agent drafts daily time entries for each lawyer to review and post. Later, it sends polite reminders on unpaid invoices and answers simple billing questions, with any dispute routed to the billing partner.
Where to start
Pull last month's enquiry log and check how many came in outside office hours and how long they waited for a first reply. If it's more than a handful, start with intake: a missed consult is lost revenue you can count. If intake is covered, start with document collection, which needs only read access to matter checklists and one sending address.
What It Connects To
A useful first build usually touches three systems: your practice management platform (Clio, MyCase, PracticePanther, Smokeball or Filevine in the US; Clio, LEAP or Actionstep in the UK), your intake or CRM tool if you run one separately (Lawmatics, Clio Grow), and email, texting or the phone line. Later builds add the document management system (NetDocuments, iManage), e-signature and billing. Several of these platforms now ship their own AI features, so check what you already pay for before anyone starts building.
If you want the agent to answer from your own intake scripts, fee terms and FAQs, we cover how to do that safely in wizeb.com/blog/train-ai-agent-on-company-data. For a phone agent, the running costs are broken down in wizeb.com/blog/ai-voice-agent-cost.
The Ethics Rules You Have to Build In
Lawyers carry duties most businesses don't, and an agent has to respect all of them. In the US, the ABA's Formal Opinion 512 on generative AI (americanbar.org/news/abanews/aba-news-archives/2024/07/aba-issues-first-ethics-guidance-ai-tools/) sets out how the existing Model Rules apply. Your state bar may have its own guidance too, so check it before you build.
- Competence and supervision. Lawyers must understand the tools they use and supervise them the way they would supervise a non-lawyer assistant. In practice: a named lawyer owns each agent, reviews a sample of its work every week, and can switch it off.
- Confidentiality. Client information is protected. Use providers whose terms say your data isn't used to train their models, keep data in approved regions, and get the data-use and retention terms in writing. Opinion 512 says some uses need the client's informed consent, so decide where your engagement letter covers this.
- No legal advice. The agent never tells a prospective or current client whether they have a case, what it's worth or what to do. It gathers facts and books time with a lawyer. This matters for unauthorised practice rules, and it also keeps you from creating a lawyer-client relationship by accident.
- No made-up citations. An operations agent shouldn't be writing legal argument at all. Courts have sanctioned lawyers for filing AI-invented cases, starting with Mata v. Avianca in New York in 2023. Keep research and drafting in tools built for it, with a lawyer checking every citation.
- Advertising and outreach. Agents should reply to people who contacted you, not cold-call or cold-text prospects. Solicitation rules and, for texts and calls, the TCPA both apply.
- UK and EU. Firms need a lawful basis under UK GDPR or GDPR and a data processing agreement with every provider, and SRA-regulated firms keep their duties on confidentiality and client care. The EU AI Act requires you to tell people they are talking to an AI from August 2026, which is good practice everywhere.
Give the agent the narrowest access that does the job. An intake agent needs to create leads and read calendars. It doesn't need the document system, trust accounting or other matters. Our security guide at wizeb.com/blog/ai-agent-security-guide covers permissions in detail.
Built-In Features or a Custom Agent?
If your whole problem sits inside one platform, such as summarising a matter in Clio, switch on the vendor's AI first and see what it gets wrong. A custom agent makes sense when the job crosses systems (intake that answers the phone, runs a conflict search and books a consult), when each practice area needs its own questions and rules, or when you want one agent across web chat, SMS and phone. You own the prompts, workflows and test sets, and you can change vendors without starting over. The full decision framework is at wizeb.com/blog/build-vs-buy-ai-agents-2026-decision-framework.
What It Costs and How Long It Takes
- One job, two or three systems (for example, after-hours web and SMS intake into Clio): about $5,000 to $15,000 and three to six weeks.
- A phone intake agent or several jobs, with actions such as conflict searches, booking and document checks: about $15,000 to $40,000 and six to ten weeks.
- Running costs: model usage, hosting, telephony, messaging and monitoring usually come to $200 to $1,000 a month for a firm with a few hundred enquiries a month.
- Your team's time: a few hours a week during the build to agree intake questions and review test outputs, then about an hour a week reviewing escalations.
Those are WizeB's working ranges for growing firms in the US, UK and Europe, not a price list. The full breakdown is at wizeb.com/blog/ai-agent-cost-guide.
How to Roll It Out Safely
- 1Write the rules first: which practice areas the agent screens for, the questions for each, what it must never say, and who it hands off to. Get a partner to sign it off.
- 2Collect a test set: 150 to 200 past enquiries or client questions for the job you picked, with the outcome your best intake person reached.
- 3Score the agent against them before any client sees it. Agree the pass mark up front: right practice area, complete conflict details, zero legal advice.
- 4Run two weeks in shadow mode. The agent drafts, a person sends or calls back. Track how many drafts go out unchanged.
- 5Go live on the lowest-risk slice first, such as evening and weekend web enquiries for one practice area, then widen once the numbers hold.
Example: A 14-Lawyer Firm Losing Evening Calls
This is an illustrative example, but the pattern is common. A personal injury and employment firm in Texas with 14 lawyers gets about 400 new enquiries a month. Its phone logs show that 35% arrive after 5pm or at weekends, and that those callers wait until the next morning for a call back. By then, many have already spoken to another firm.
The firm starts with intake only. A voice and SMS agent answers out-of-hours enquiries, asks the screening questions for each practice area, collects the names for a conflict search, and books a consultation into the next free slot. Anything urgent, such as an accident that happened today, goes to the on-call lawyer's phone. After two weeks in shadow mode, the agent goes live for evenings and weekends. Three months later the firm adds document collection for new personal injury matters. Both phases cost about $22,000 to build and run at around $700 a month including telephony. The lawyers start each morning with booked consultations and complete intake notes instead of a list of voicemails.
How WizeB Builds Agents for Law Firms
WizeB builds custom AI agents, workflow automations and voice agents for growing businesses in the US, UK and Europe. For law firms, we start with your enquiry and matter data and pick the one job where an agent wins the most clients or frees the most staff hours at the lowest risk. We connect it to your practice management system, calendar and phone line with the narrowest access that works, write your ethics rules into its instructions and tests, and test it against your past enquiries before a client sees it. You own the prompts, workflows, test sets and accounts. Details are at wizeb.com/services/ai-agents, and phone intake is covered at wizeb.com/services/voice-ai.
Plan your first law firm agent
Book a free scoping call at wizeb.com/services/ai-agents. Bring last month's enquiry log with arrival and first-reply times, and the practice management system you use, and we'll tell you which job an agent should take first, what it will cost to build and run, and how fast it can go live. Not sure you're ready? Take the free AI-Native Index at wizeb.com/ai-native-index.
