The list at the centre of the Law Society's report on agentic AI in the justice system reads to me like a working week. I sat for four years as a fee-paid judge in the county court, where small claims and debt claims arrive every day. Published on 5 October, the report names small claims, debt recovery, road traffic penalties, employment tribunal claims and probate as the places where autonomous systems will first make it easier for people to bring a case.
Agentic AI differs from the chatbots most firms know. A generative tool writes, while an agentic system takes a goal, plans the steps, adapts them and carries them out without a person approving each one. The Law Society says fully agentic AI is not yet in use in the justice system but looks likely to emerge, and that it will arrive through a series of procurement and design choices rather than in a single moment. The work is a foresight study built on desk research, interviews with around 30 experts and discussions with members, so it describes where things are heading and does not predict dates.
Easier to start a claim is not the same as justice
The report's sharpest warning is that agents help people into the system without improving what happens once they arrive. A tool that issues a claim and assembles the supporting papers does nothing for the fairness of the hearing or the quality of the decision. The Law Society expects more claims and longer, more complex bundles, with courts answering the extra load through automation of their own. Its press release notes that some court hearings are already listed as far ahead as 2030.
For a practice that does litigation, debt or employment work, the effect lands on your desk first. The letter before claim, the particulars or the witness statement from the other side may come from an agent working for a litigant in person, and it will be long, confident and aimed at one outcome. Budget fee earner time to check every authority in that material, and prepare the proportionality arguments before the first hearing rather than at it.
An agent acting for your client is not an adviser
The report draws a line that every solicitor will recognise. An agent built to achieve the user's objective has no duty to the court, no sense of proportionality and no reason to point out the weak parts of a case. A client who reaches you after a tool has run the opening stages may arrive with expectations that tool set, and the Law Society expects lawyers to spend more of their time validating, correcting or reframing AI-driven decisions. Brett Dixon, the Law Society vice president, put the principle in five words: “Legal judgement cannot be delegated.”
Add a question to your intake for these areas of work, asking what any AI tool has already sent, filed or agreed in the client's name, and record the answer on the file before you give advice.
When your own tools act in a client's name
The Law Society calls for clear accountability, transparent AI decision-making, meaningful human oversight and public debate before widespread use. Those calls go to government and the regulators, yet each has a counterpart inside your firm. The report warns that oversight can exist on paper and vanish in practice when a system is too complex, too routine or too convenient for the person meant to check it.
If you run software that sends chasers, issues claims or files documents, map each step where it acts for a client, name the person who approves that step, and test whether the approval is a real check or a reflex click. The NCSC advice on agentic AI covers the security side of the same question. Then read the consumer summary of the report with whoever leads your litigation or debt work, and agree this month where your firm will draw the line between what a tool may do alone and what a person must decide.
The Law Society's report, The future of agentic AI in the justice system, is open to any reader on lawsociety.org.uk, with two summaries and the full report as free PDF downloads and no sign-up.
If you want a clear rule on where your firm's AI tools may act alone and where a person must decide, start with a conversation.
