Samuel Puchala will speak on building reliable, repeatable AI workflows at the Canadian Legal Summit
Most lawyers experimenting with legal AI are making one of two mistakes – and Samuel Puchala, president and general manager of LexisNexis Canada, sees both clearly from where he sits. One group briefly tries a tool, gets a disappointing result, and shelves it for another day. The other dives in, gets impressive early returns, and trusts the output more than it warrants. “There’s a happy middle ground where there’s really a very significant amount of efficiency and proficiency which you can gain from these tools,” Puchala says. “But you have to keep your professional judgment.”
From pharmacology to legal tech
Puchala’s path to running one of Canada’s leading legal technology businesses was anything but direct. He studied pharmacology and toxicology at the University of Toronto before pivoting to law school at Western University and completing an MBA at the Ivey Business School. He then spent time at Bain & Company, a stint in pharma at Merck, and 16 years as a partner at Satov Consultants before eventually landing at LexisNexis. “I started in law school after a pharmacology and toxicology degree and then did almost everything else before finding my way back into that legal ecosystem,” he says.
A steady-state business no more
When Puchala joined LexisNexis as managing director of its large law and print businesses in December 2020, the company occupied a stable but incremental position in the market. By the time he stepped into the president’s role in June 2024, something more disruptive was well underway. That shift crystallized when large language models became publicly available in late 2022. “As soon as I opened my first ChatGPT window, I knew the world [was] going to change very quickly for us,” he says. LexisNexis had been working with AI well before that public moment, and the company has since repositioned itself – a shift Puchala describes as “really a transformation from being a content-led set of online tools for lawyers to an AI technology-first business.” Content remains embedded in its competitive position – LexisNexis still sells legal books, and its case law database anchors its AI products – but the strategic direction has changed.
That pace of change has created a new problem for law firm buyers. A LexisNexis Canada survey of 610 lawyers found that 93 percent are already aware of generative AI. However, Puchala says it is likely to be essentially universal by now, with adoption rates accelerating year over year, as Canadian Lawyer’s coverage of legal AI tools and adoption consistently shows. Yet the speed of product development has made traditional sequential evaluation cycles obsolete. “By the time you’re looking at that second, third competitor, the first one’s already moved several generations ahead,” Puchala says, meaning firms that spend months on structured pilots risk making decisions on tools that have already advanced beyond what they evaluated.
Accuracy is not just about hallucinations
The concern lawyers most often raise about AI is accuracy, and Puchala separates two problems that are frequently conflated. The first is hallucination – fabricated citations or invented authorities with no basis in actual law. “Accuracy is fundamental to our value proposition,” he says. Within LexisNexis’s platform, the AI is grounded in verified Canadian legal content, with citations linking back to the original source material, and human expertise is built into the review process. “Within our ecosystem, you have very little chance of a hallucination because everything is cited back, is linked back to the actual content,” Puchala says. But the second problem is subtler and persists even when citations are accurate. “A real and accurate citation that’s not a hallucination doesn’t automatically mean that analysis is correct,” he says. A verified case still needs to support the proposition and fit the facts at hand – and that judgment call remains with the lawyer, not the tool.
When lawyers get AI wrong
Two root causes account for most of the frustration lawyers report when AI tools fail to deliver. The first is using a product for a task it was not built to handle – expecting a capability it does not yet have. “Telling people what not to do is just as important as selling on all the great things that the product is able to do,” Puchala says, a principle he considers central to how LexisNexis introduces its tools to new customers. The second is imprecise prompting. Vague or ambiguous instructions leave the system without the information it needs to produce a useful output. “If you’re working with an articling student and you send them out there with some ambiguous instruction, there’s a good chance they’ll run around confused,” he says. The leverage is real, the capability is genuine, but oversight is not optional – outputs can be brilliant, draft-quality, or completely off base, and the only defence is professional review.
What agentic AI actually changes
The shift to agentic AI represents the most meaningful change now underway in legal practice. “In a basic chat interaction with an AI, you ask a question, and you receive an answer. With agentic AI, you can give it more of an objective, and it can work through the multiple steps towards that objective rather than have you direct its every move,” Puchala says. The practical illustration he offers is brief preparation: an agent can coordinate the organization of source documents, analyze facts, draft a memo, and check the output – work that previously required manual direction at each stage. LexisNexis is building out customizable skills – pre-built workflows that lawyers can modify or write from scratch for their specific practice area – and collaborative workspaces where counsel and clients can work together inside the AI environment. It has also announced a recent data alliance with Harvey.
Puchala will explore these topics further as a panellist at the Canadian Legal Summit taking place on October 14, 2026, in Toronto, where the conversation will centre on building reliable, repeatable AI workflows for legal practice.
This article is based on an episode of CL Talk, which can also be found here:
The episode is also available on our CL Talk podcast homepage, which includes links to follow CL Talk on all major podcast platforms.
