AI agents that handle intake questions, scheduling, matter status and routing for law firms, with a firm boundary against legal advice and a handoff to people.
Agentforce lets law firms and legal service organizations deploy AI agents for the administrative side of client service. Agents can collect intake details from prospective clients, schedule consultations, answer status questions about existing matters, explain firm procedures such as how to upload documents, and route requests to the right practice group. They are grounded in the firm's CRM data and approved knowledge, and they hand off to a person whenever a question turns legal. They never give legal advice, assess the strength of a case or suggest a course of action. That boundary is designed in, not left to chance.
Legal is one of the most constrained settings for AI agents. Only licensed attorneys may give legal advice, a careless answer can create an implied attorney-client relationship, and confidentiality obligations cover even the fact that someone consulted the firm. Agentforce fits because its agents work within defined topics and actions, so a firm can permit scheduling, intake collection and status lookups while blocking anything that looks like advice. The Einstein Trust Layer governs how firm data moves through the model and what is retained. Where it does not fit is substantive legal work: research, drafting and case strategy belong with attorneys and specialized legal tools. Firms should see Agentforce as a way to reduce administrative load and response delays, not as a lawyer.
An agent on the firm's website asks structured questions about the type of matter, jurisdiction, parties involved and contact details. It records answers in Salesforce, flags the matter for a conflicts check, and tells the visitor clearly that no attorney-client relationship exists yet. Intake staff review the record rather than starting from a voicemail. Urgent situations, such as an approaching deadline, trigger immediate handoff to staff.
Once intake is reviewed and conflicts are cleared, the agent offers consultation times based on attorney availability and practice area. It sends confirmations, reminders and rescheduling options and records the appointment on the matter. Staff spend less time exchanging messages to find a slot. The agent can also explain what to bring to the consultation and collect engagement paperwork links approved by the firm. If a prospective client asks substantive questions while scheduling, the agent notes them for the attorney rather than answering.
Verified clients can ask about upcoming dates, outstanding document requests or who is handling their matter. The agent answers from approved fields in the CRM, such as next hearing date or documents received, and passes anything interpretive, such as what a ruling means, to the responsible attorney or paralegal. Clients get answers outside office hours about routine items, and staff receive fewer calls asking whether a document arrived. Every status answer is logged, so the firm can review what was shared and correct fields that clients find confusing.
Emails, portal messages and chat requests arrive with mixed purposes: billing questions, new matters, document submissions and urgent deadlines. The agent classifies each, routes it to billing, intake or the matter team, and escalates time-sensitive items to a person immediately, with a summary so nobody has to reread the thread. Routing rules can reflect practice groups, offices and conflicts walls, so a request never reaches a team screened from that matter. Response times by category help managers see where requests wait longest.
Before building a legal agent, the firm answers a few design questions. What data the agent may read: typically contact, matter status, key dates and assigned staff, but never privileged notes, work product or documents. How identity is verified before any matter-specific answer is given, since confidentiality attaches even to whether someone is a client. And how topics are defined, with explicit topics for intake, scheduling, status and routing, and an instruction that any legal question triggers handoff. Prospective client records should stay separate from active matters until conflicts are cleared.
Matter numbers, key dates, responsible attorneys and status fields sync to Salesforce, giving the agent accurate, approved information to share with verified clients.
Names and parties collected at intake feed the firm's conflicts process, and the agent does not schedule a consultation until conflicts are cleared.
Attorney availability and court deadlines inform scheduling, so the agent never books consultations over hearings or filing dates the docketing team has recorded.
Agents must not interpret law, predict outcomes or recommend actions. Define topics narrowly, test with questions designed to draw out advice, and require immediate handoff when a request becomes legal. Review transcripts regularly with a supervising attorney, and adjust instructions whenever an answer comes too close.
Intake collects names of adverse parties and related entities that must be checked before the firm learns more. The agent should gather only what the conflicts process needs, avoid inviting detailed facts, and hold scheduling until a person confirms the check is clear.
Keep privileged communications, work product and client documents outside what agents can retrieve. Verify identity before sharing matter details, disclose that the user is talking with an AI agent, and confirm with your ethics counsel how transcripts are stored and whether they could be discoverable.
No, and it should not be configured to. Agents handle administrative tasks such as intake, scheduling, status updates and routing. When a user asks what they should do or what the law says, the agent explains that it cannot provide legal advice and connects them with the firm. Topics and instructions enforce that boundary. Transcripts are reviewed regularly.
Limit the data the agent can read, verify the user's identity before discussing any matter, and exclude privileged notes and documents entirely. The Einstein Trust Layer helps with data handling around the model, but access design is the firm's responsibility. We test with scenarios where someone tries to learn about another person's matter. Those tests repeat after every change.
That is a professional responsibility question for your firm's counsel, not something software settles. Firms typically have the agent state clearly that no relationship exists until an engagement is confirmed, collect only limited facts before conflicts clearance, and keep prospective client records separate. We build whatever disclaimers and limits your counsel approves. Your counsel should review the wording.
It can be, for internal requests. Agents can take intake for contract reviews, NDAs or policy questions from employees, collect required details, route to the right lawyer and report status. As with firms, the agent does not interpret contracts or advise; it organizes the work so attorneys receive complete requests. Legal operations teams often lead these projects.
One onshore team with 150 Salesforce certifications, a Salesforce Consulting Partner since 2017.
A monthly brief for the people who own Salesforce, AI and revenue technology
What changed in Salesforce and AI this month, and what to do about it.
One email a month. Written by the consultants who deliver the work, not by a marketing team, for the leaders who make the technology decisions.
Consultant analysis, not vendor recaps. One click to leave.