A day at a law firm or an in-house legal team is full of work that doesn’t need a lawyer but ends up with one anyway. Reading court notices, checking deadlines, chasing clients for documents, drafting a first version of a standard contract, answering “how’s my case going?”. All of it is necessary, all of it repeats, and it eats hours that could go into case strategy.
AI agents can take on a good share of these routines. The care lies in drawing a clear line where the agent’s work ends and the lawyer’s begins. This guide lays out that line for law firms and legal departments.
What goes to the agent
The rule of thumb is simple: the agent prepares, organizes, checks and flags. Examples of routines that work well:
- Court notices and filings. Reading court dockets and e-filing systems, matching each notice to the right matter, summarizing it and alerting the responsible lawyer.
- Deadline tracking. Calculating a suggested deadline, accounting for business days and holidays, and adding it to the calendar for the lawyer to confirm.
- New client intake. Taking the first contact, asking the initial questions, requesting documents and giving the lawyer an organized summary of the matter.
- First drafts. Producing the first version of simple motions, notices and contracts from the firm’s own templates.
- Triage of incoming contracts. Comparing against the company’s standard terms and flagging clauses outside acceptable limits, such as penalties, jurisdiction or limitation of liability.
- Client updates. Letting clients know when their case moves, in language they understand, with text reviewed by the lawyer where agreed.
- Organization. Filing documents by client and matter, keeping records up to date and chasing overdue invoices.
In-house legal teams can add routines like taking requests from other departments, classifying them by type and urgency, answering recurring questions based on internal policies and tracking contract signatures.
What stays with the lawyer
The agent never signs, never files on its own and never gives an opinion on the firm’s behalf.
Anything involving legal judgment, professional responsibility or a commitment to the client stays with a person:
- Setting case strategy and the argument to make
- Reviewing and approving every document before it’s filed
- Confirming the final deadline, even when the agent calculated a suggested one
- Giving the client legal advice on what to do
- Accepting or declining a new matter and setting fees
- Negotiating settlements and contract terms
- Deciding on any situation involving a conflict of interest
This separation protects the client, the firm and the lawyer. Professional responsibility still belongs to whoever signs. The agent increases the team’s capacity, but it doesn’t change who answers for the work.
Made-up citations and other known risks
The most talked-about AI risk in law is the made-up citation: the model produces case law with a number, a judge and a summary that look real and don’t exist. There have already been public cases, in several countries, of filings submitted with exactly this kind of error.
A well-built agent reduces that risk in a few ways:
- It only uses as sources the firm’s templates, the case law databases it searched and the documents in the matter
- It shows where each citation came from, with a link or reference to check
- It clearly marks what the lawyer needs to verify
- It doesn’t fill gaps by inventing: if it didn’t find something, it says so
Even so, human review remains mandatory. The agent saves the time it takes to assemble the document, and the lawyer spends part of that time reviewing it carefully.
Other risks worth watching: a deadline calculated with the wrong local holiday, a notice matched to the wrong matter because of a similar name, and one client’s document showing up in another client’s context. All of them are preventable with clear rules, review and strict separation of data by client.
Confidentiality, professional rules and privacy
Client confidentiality is one of the pillars of legal practice, and bar associations have started publishing guidance on generative AI. In the US, the ABA’s Formal Opinion 512 (2024) focuses on competence, confidentiality, supervision and transparency with clients; bar regulators in other countries have issued similar guidance. It’s worth reading the rules that apply to you before you start.
In practice, these are the safeguards for the agent:
- Data in the firm’s environment. Documents and conversations stay in accounts and systems the firm controls.
- Contracts with AI providers. A written commitment that data isn’t used to train models and is deleted within a set period.
- Separation by client. The agent never mixes documents or information from different clients in the same task.
- Role-based access. The agent respects the same ethical walls as the lawyers, including in confidential or conflicted matters.
- Everything logged. What the agent read, produced and sent is recorded, so the firm knows exactly what happened in each matter.
- Transparency with clients. If a client is talking to an agent, they know it.
Privacy laws such as GDPR and Brazil’s LGPD apply too, and firms often handle data about people beyond their clients, such as opposing parties and witnesses. Some matters involve sensitive data, like health information in disability or employment cases. That data falls under the same rules of minimum access and retention.
Integrating with legal systems
An agent is only truly useful when it works in the systems the team already uses: the firm’s or department’s practice management software, court e-filing systems, email, WhatsApp or Teams, the e-signature tool and the document repository.
Before you start, it’s worth mapping out:
- Which systems the chosen routine uses, and which have integrations available
- How court notices arrive today: a docket monitoring service, email or manual checks
- Where the document and contract templates live, and whether they’re up to date
- Who has access to what, so the agent follows the same rules
How to get started at a firm or legal department
The first agent usually pays off most in a routine with high volume and clear rules. Court notices and deadlines, new client intake and standard contract triage are good candidates. An approach that works:
- Pick one routine and one responsible lawyer, with time set aside for review
- Measure the before: time spent, weekly volume, errors and rework
- Gather the firm’s templates and rules, including the exceptions only the veterans know
- Run the agent in parallel, comparing what it would do with what the team did
- Expand what it does gradually, always with review at the points that involve judgment
Bring the team in from the start. Paralegals, legal assistants and junior associates know the routines in detail and will work alongside the agent. With the repetitive part handled by the agent, they can take on work closer to the actual practice of law.
The agent handles the volume. The lawyer handles the judgment and answers for what they sign.