What is an AI agent for a lawyer?
An AI agent for a lawyer is software that carries out defined pieces of practice work on its own - reading and filing email, qualifying a new inquiry, tracking deadlines, assembling a bill - and brings every client-facing or high-risk step back to the lawyer for approval. The difference from a chat assistant is the verb. An assistant answers questions; an agent does work inside your systems and shows you the result before it counts.
If you are thinking about several agents working across a whole practice, AI agents for law firms covers the firm-level picture. This page answers the smaller, more personal question: what changes in one lawyer’s working day.
What does a day with an AI agent look like?
Here is a Tuesday for a solo lawyer running Referent’s agents.
7:40 a.m. - triage. Overnight, the email agent has read the inbox, routed each message to its matter, and drafted replies where replies are due. You read six drafts over coffee, approve four, rewrite two. Nothing has gone out under your name without your sign-off.
10:15 a.m. - intake from the courthouse hallway. A prospective client submits an inquiry while you are in a hearing. The intake agent captures it, checks it against your qualification criteria, opens the matter, and prepares a proposal with a payment link. In the hallway afterward, you review on your phone and approve. The prospect hears back within the hour instead of on Thursday.
1:30 p.m. - follow-ups. Three matters have been waiting on other people for over a week. The email agent has the nudges drafted; you approve all three in one sitting. Walking back from lunch, you tell Referent by voice to set a callback with the expert witness, and it becomes tracked work on the right matter rather than a note in your pocket.
4:00 p.m. - deadline watch. A hearing date moves. You change it once; the deadline agent syncs both ways with your calendar, so the matter and the calendar never disagree - and the date never lives only in your head.
Month end - billing prep. The billing-prep agent assembles your time and activity into billing-ready form. The evening you used to spend reconstructing the month from sent mail and calendar entries becomes a review pass.
Notice what the agent never touched: the legal judgment. It handled the recurring work between the legal work, which is where solo and small-firm days quietly go.
What makes a legal-grade agent different from a general assistant?
Four things: matter context, an approval loop, an audit trail, and confidentiality you can explain to a client.
Matter context means the agent knows which of your matters “Re: the agreement” belongs to, who the parties are, and what has already been said - without it, drafts are generic and filing is guesswork. The approval loop is the trust model: AI prepares, the lawyer approves every client-facing or high-risk action. Legacy software records the work - Referent executes it, and the lawyer signs. The audit trail logs every step, which turns your duty to supervise work product from a feeling into something you can inspect. And confidentiality obligations follow client data into any tool that touches it; Referent’s answer is per-firm isolated workspaces, AES-256 encryption at rest, TLS 1.2+ in transit, and no training on firm data - details on the security page.
General-purpose agents are worth taking seriously. OpenClaw is free, open source, and the fastest-growing open-source project in GitHub history, with 347,000+ stars as of April 2026; it runs on your own hardware around the clock with persistent memory. Hermes Agent, released by Nous Research in February 2026, writes its own skill documents as it works and shipped a desktop app in June 2026. Both are local-first, model-agnostic, and capable of real work. The line for legal practice: neither ships with an approval workflow, an audit trail, or legal-specific guardrails, and self-hosting means you own confidentiality, security patching, and supervision. Those duties stay with the lawyer either way; the question is whether your tools carry part of the load. Referent is the AI-native pick here because the approval loop and the audit trail are the product, not something the operator assembles.
Both OpenClaw and Hermes speak MCP, though - which opens a door in the next section.
How do you get an AI agent for your practice?
Two paths, and they start at the same door.
The first is Referent itself: AI-native practice management for solo and small firms, with the intake, email, deadline, and billing-prep agents built in, plus mobile apps for iPhone and Android. It can connect to your existing system of record over API or migrate your data in. Cohort 1 of the beta is full - 450+ applications in under four weeks for 20 seats, with onboarding starting August 10, 2026 - and the Cohort 2 waitlist is open. For the thinking behind the product, start with what is an AI-native law firm or the AI-native hub.
The second is bringing the AI app you already use. Referent runs an MCP server - MCP is the open standard connecting AI assistants to tools - at https://mcp.referent.law/mcp. In your app’s connector settings, whether that app is Claude, ChatGPT, Gemini, Perplexity, OpenClaw, or Hermes, add the server URL and sign in with your Referent account. Your assistant can then read and act on your clients, matters, and documents with the same permissions, the same approvals, and the same audit trail as the Referent app. Setup notes for each client are on the MCP page. Since MCP access requires a Referent account, both paths run through the same beta.
Cohort 1 filled its 20 seats in under four weeks. If the working day above is the one you want, join the Cohort 2 waitlist - every applicant gets founding-firm perks during the open beta.
Frequently asked questions
Is an AI agent the same as ChatGPT or another chat assistant?
No. A chat assistant answers questions in a window; an agent carries out work inside your systems - filing email to matters, opening a matter from an inquiry, preparing a bill. A legal-grade agent adds the parts practice requires - matter context, an approval step before anything reaches a client, and an audit trail of every action. The interface may look similar, but an agent's output is work done, not text to copy out.
Will an AI agent send anything to my clients without my approval?
Not in Referent. The trust model is that AI prepares and the lawyer approves every client-facing or high-risk action - draft replies, proposals, and payment links all wait for your sign-off. Every step is logged in an audit trail, so you can see what the agent did and when. General-purpose agent frameworks do not include this loop by default, which is the main reason to distinguish general from legal-grade agents.
Do I have to replace my current practice management software?
No. Referent can connect to an existing system of record over API and work alongside it, or migrate your data into Referent when you are ready. Solo and small firms rarely get to pause practice for a software change, so both routes are supported. Which one makes sense depends on how much of your current system you still use.
Is client data safe with an AI agent?
Confidentiality obligations apply to any tool that touches client data, so this is the right question to ask of every vendor. Referent stores data encrypted with AES-256 at rest and TLS 1.2+ in transit, keeps each firm in an isolated workspace, does not train models on firm data, and controls access by role. SOC 2 and ISO 27001 are in preparation, and the platform is CASA verified. The full picture is on our security page.
Can I use a general agent like OpenClaw or Hermes for my practice?
You can, and both are serious tools - OpenClaw had 347,000+ GitHub stars as of April 2026, and Hermes shipped its desktop app in June 2026. Out of the box, though, neither has an approval workflow, an audit trail, or legal-specific guardrails, and self-hosting means you own confidentiality, security patching, and supervision of everything the agent does. A middle path exists - both are MCP-compatible, so they can connect to Referent's MCP server and act on your practice data with Referent's permissions, approvals, and audit trail applied.
When can I get Referent's agents?
Cohort 1 of the beta is full - 450+ applications in under four weeks for 20 seats - and onboarding starts August 10, 2026. The Cohort 2 waitlist is open, and every applicant gets founding-firm perks during the open beta. Joining the waitlist at referent.law/apply is the fastest route to a seat.