AI5 min read

What an AI agent is and how it really helps a law office

The word “agent” is used so often that it now sounds like a marketing label. Behind it, though, is a concrete idea that can save hours of routine work in a law office — if it is introduced carefully. In this article we explain what an AI agent actually is, where it helps in legal practice, where you should not trust it, and how to get started without unnecessary risk.

Chatbot versus agent: what is the difference

An ordinary chatbot answers a question. You type, it returns text, and that is it. An agent has a goal and can decide on its own which steps to take to reach it. To do so it uses tools: searching documents, reading emails, filling in a spreadsheet, checking a calendar, creating a draft.

The simplest definition: an agent is a language model connected to tools that works in several steps and checks its intermediate results. If you give it the task “sort this week’s new enquiries by type of case and urgency”, it will open the mailbox, read the messages, classify them and show you a list — but it should not reply to clients by itself.

This is where the key concept of a human in the loop comes in. A good agent prepares; a person approves. In legal work this is not a nice-to-have but a requirement.

Real uses in a law office

Here are tasks where agents already work well, because the result is easy to check and a mistake is not fatal:

  • Intake triage. The agent reads incoming emails and forms, identifies the type of matter, deadlines, parties and missing documents, and suggests which colleague should take the case.
  • Document summaries. A long contract, case file or minutes are condensed into one page with the key obligations, deadlines and penalties — with references to the specific clauses so they can be checked quickly.
  • Deadline tracking. From court notices and letters the agent extracts dates and proposes them for the calendar. A person confirms every date.
  • First drafts from templates. A power of attorney, a formal notice, a request for documents, a standard letter to a counterparty — the agent fills in your template with data from the file.
  • Searching your own archive. “Have we had a similar dispute over a commercial lease?” The agent searches the firm’s old opinions and notes and returns the excerpts with their source.

What these tasks have in common is that the agent works with your data and templates, and the final text goes through a lawyer. The agent does not give legal advice to clients and does not sign anything.

Invented citations: the best-known lesson

Language models can generate convincing text that is completely wrong. In law this is especially dangerous, because it looks like real case law.

The best-known example is Mata v. Avianca in the United States (2023). Lawyers filed a brief citing court decisions that ChatGPT had invented. The court found that the decisions did not exist and sanctioned the lawyers. The lesson is not “AI is bad” but that every citation must be checked against the primary source — the official database of decisions or the text of the law.

Professional secrecy and GDPR

A lawyer keeps the client’s secrets. When you put a document into an online service, you need to know where it goes, who has access and whether it is used to train the model.

  • Check where the data is processed — inside or outside the EU, and on what basis it is transferred.
  • Require a data processing agreement with the provider and an explicit term that your data is not used for training.
  • Use business versions of the services, not employees’ personal free accounts.
  • Keep data to a minimum: if the task is to check the structure of a contract, the names of the parties can often be replaced.
  • Keep a list of which tools are allowed in the office and for which tasks.

More on the practical side of data protection on a small website is in our article GDPR and cookies: a practical checklist.

The European AI Act

Since 1 August 2024, Regulation (EU) 2024/1689 on artificial intelligence (the AI Act) has been in force. Its obligations apply in stages and depend on the risk of the system and on your role — whether you are a provider or only use a ready-made system.

For most law offices that use agents for internal help, the most practical consequence is good organisation: knowing which systems you use, training staff to work with them, and making sure people can check and override the result. For the specific obligations in your case, consult a specialist.

How to start: a short plan

  1. Pick one recurring task with a clear result, for example summaries of incoming contracts.
  2. Describe how it is done today: how long it takes and who checks the result.
  3. Choose a tool with business terms, a data processing agreement and processing in the EU where possible.
  4. Run a pilot with anonymised or non-confidential documents for 2–4 weeks.
  5. Compare: time saved, number of errors found, convenience for colleagues.
  6. Write short internal rules: what is allowed, what is not, who approves.
  7. Only then expand to the next task.

Do not start with “an agent that does everything”. Small, measurable steps bring real benefit and do not put clients’ trust at risk.

In summary

An AI agent is a useful assistant for preparatory work: it reads, sorts, summarises and proposes drafts. It does not replace legal judgement and carries no responsibility. A firm that introduces it with clear rules, a check of every citation and care for client confidentiality can free up time for the real work with people. If you are considering such a tool for your team, you can write to us — we will tell you honestly whether your task is a good fit for an agent.