Can law firms use AI agents before staff are trained?

Yes, under supervision and only at read and draft depth. A law firm can use agents for document review, research summaries and chronologies while staff train. Nothing the agent produces should be filed, served or sent to a client until a trained lawyer has reviewed it. Professional duties to clients and courts do not pause for a pilot.

Why does the profession raise the bar?

Lawyers owe duties of competence, confidentiality and candour to clients and courts. A research summary citing a case that does not say what the summary claims is not a technology problem; it is a failure of the lawyer who relied on it. Supervising a junior's work has always been part of practice. Supervising an agent's work is the same duty with a different subject, and it requires the supervisor to know how the agent fails.

Confidentiality adds a second constraint. The agent must work matter by matter, with no leakage between clients, and every document it reads should be logged.

What can agents do from day one?

• First-pass document review, tagging relevance and issues with reasons (depth 2).
• Research summaries, with every authority linked to its source so it can be checked (depth 1 and 2).
• Chronologies and dramatis personae from the file, each entry linked to its page (depth 2).
• Clause extraction for due diligence schedules (depth 2).
• First drafts of standard documents from the firm's precedents (depth 2).

What must wait for trained lawyers?

• Anything filed with a court or registry.
• Anything sent to a client, the other side or a third party.
• Privilege decisions.
• Advice, in any form.
• Use of the agent on matters with heightened confidentiality, until access controls are proven.

Many firms will keep filing and client communication held permanently, whatever the agent's certification.

What does supervised early use look like?

Take a commercial litigation team at a mid-sized firm. It starts an agent producing chronologies and first-pass review on two matters. A senior associate reviews every chronology against its sources.

In week two, the associate finds that the agent attributed an email to the client's finance director when it was sent by an assistant on the director's behalf. The distinction matters to the case. The associate corrects it, writes the reason, and the brief is updated to record the sender and the named signatory separately. Over the same weeks, every lawyer and paralegal on the team completes foundational training, and the associates and senior paralegals who review agent work most often work towards operator certification. Expansion to further matters waits for that. Property firms face a similar choice with very different stakes, set out in whether property management firms can use agents before staff are trained, and accountancy practices face the closest parallel, covered in whether accountancy firms can.

Which controls fit a law firm?

• The firm's own limits cap agents at depth 2 and hold every external send.
• Matter-level scope, with access logged per document.
• Every authority in a research summary linked to source, and checked before use.
• Reasons written for every correction and filed with the decision.
• A named supervising lawyer for each matter where the agent works.

Where a firm operates in the UAE, the federal Personal Data Protection Law shapes how client data may be handled, discussed in UAE Personal Data Protection Law and training staff who use AI agents. Firms also tend to make a set of predictable errors, such as training partners last, covered in what law firms get wrong about AI agent training.

Where the Academy fits

Every fee earner and paralegal can take Pryme Intelligence Academy's free Foundations track, which covers the decision chain, depth and certification, and ownership, data and trust. Associates and paralegals who review agent drafts daily take the Operator track, with module tests set in legal services. The topic guide on training teams for AI agents covers who to train first.