Wexler's head of commercial strategy Cail Wyn Evans joined Philip Young of Garfield AI, Philip Roberts KC of One Essex Court, Warren Little of BT Group, and Oran Gelb and Joanna Munro of BCLP to discuss ‘AI in Litigation, beyond the hype: What works, what doesn't, and where are we heading?’ Here are seven of the most interesting ideas that came up.
Will AI reduce litigation?
"Users tell us they're issuing ten or twenty times more claims than they used to, because it's now economic for them to do so," said Philip Young of Garfield AI, about the debt recovery work his platform handles. The mechanism is Jevons paradox, which Cail Wyn Evans set out: when steam engines became more efficient in the nineteenth century, coal consumption rose rather than fell. As the cost of bringing a claim falls, the number worth bringing rises. Cail said Wexler is seeing more matters created across its users.
Which AI tools are working?
The reliable ones stay close to the evidence. Cail said Wexler blocks internet access so that every assertion traces to a document you've uploaded, and he separated fabrication, an assertion with no basis in evidence, from the different problem of applying accurate information incorrectly. "We're not a legal research platform," he said.
Philip Roberts KC of One Essex Court made the same point from the bar, endorsing walled-garden tools confined to your own documents or to court decisions and legislation. Young's platform sits at the other end of the workflow, taking small debt claims from chasing letters through to default judgments and trial bundles.
Who's pushing AI adoption?
Increasingly, the client. Warren Little of BT Group said BT were early adopters of Microsoft Copilot, now use Wordsmith configured to take BT's view of its market, and are looking at specialist litigation tools rather than general business AI, so the team can do more of the work itself. Clients like BT now expect their external counsel to use AI, because they use it internally, and from the firm side the list of clients refusing AI is shrinking.
Where does it go wrong?
In predictable places. Roberts noted the sycophancy built into general models, which will tell you your question was the most insightful anyone has ever asked. Young pointed out that debtors are defending claims with consumer-grade models that miss their best arguments, weakening their own position. And Little described the downstream harm: a claimant often discovers the case is weak only at trial, after the time and money are gone.
What still needs a lawyer?
The elements that were never only about documents. Cail pointed to the client relationship as work that conducting litigation only partly describes. Joanna Munro of BCLP made the related point that translating an AI output into a client's commercial reality depends on knowing the business and its strategy. Little doubted that the late nights juniors once spent paginating built the skills that matter, and suggested the real lesson was watching senior lawyers read clients and commercial situations. Verification still sits with the lawyer too, though Roberts offered a caution against treating the human alternative as flawless. "Have you appeared in front of a human judge?" he said.
Does AI cut costs?
The room split on it. Some argued that new tools tend to add process rather than remove it, and that litigation costs more than it ever has. Young put the counterview on certainty rather than headline price. When you can constrain the cost variables in a matter, you can quote fixed fees for defined stages, which moves a client from an open-ended estimate to a defined range. That changes which cases get run: Young recalled meritorious claims worth several million pounds that previously couldn't justify their own costs and now can, and litigation funders, whose central problem is uncertainty over budgets, can back cases they'd otherwise decline.
What's holding legal AI back, and where will it go next?
The courts, more than the technology. Young said firms still serve documents to each other by email in 2026, while Singapore and the United States already run shared digital service, as Little and Roberts noted. He proposed putting an experienced software engineer on the committees that revise the Civil Procedure Rules, which were never written to be captured in code, and pointed to the county court bulk issuing centre in Northampton, which already operates on its own rules inconsistent with the CPR. Looking further out, Young expected the profession to split, with high-volume work moving to technology and complex work staying with lawyers who sit on top of it, and he confirmed Garfield is building an API, during a discussion of agentic tools raised from the audience. On disclosure, Roberts said current Bar Standards Board guidance asks for transparency where AI materially affects the scope of your services and stops short of a duty to declare every use.
The overarching theme was more matters, more documents, and courts with less room to absorb either. The premium sits on knowing the facts early and being able to show where each one comes from - and that's what we built Wexler to do.