Meet two experts who engineer reliability into every AI output at Thomson Reuters
Highlights
- Thomson Reuters experts engineer reliability into CoCounsel Legal through human verification and trusted databases.
- Professional-grade AI balances cutting-edge innovation with accuracy standards demanded by legal practice.
- Agentic AI systems represent the next generation, offering autonomous decision-making within user-controlled parameters.
When a legal professional uses AI to draft a brief or analyze research for a case, the hours of work saved are substantial. The accuracy of their system’s information, however, is another matter.
As a lawyer, do you feel assured that a brief you submit doesn’t contain hallucinated case citations that will earn a reprimand from a judge? Are you sure that the AI-generated contract you send to a client includes all essential clauses and employs the proper terminology?
It’s an exciting period for legal technology, with expansions and developments in AI happening by the month, if not the day. Yet any push for innovation in AI, as important as it is to legal tech, should never come at the expense of the product’s core reliability.
Security is essential. Lawyers using AI need to be confident that any piece of data they reference — and any information their system provides — has been verified by knowledgeable, veteran and, most importantly, human legal professionals.
Jump to ↓
Product leaders prioritize trust and accuracy over speed
From a “nice-to-have” to a “must-have”
Why professional-grade AI is in a different class
Building trust through transparency
Balancing innovation with reliability
Working for you, not the other way around
Product leaders prioritize trust and accuracy over speed
Two product leaders at Thomson Reuters drive CoCounsel Legal’s development with a focus on reliability and user trust. Matthew Heagerty, Vice President of Product for CoCounsel Skills, has been with the product since day one, guiding its evolution from startup innovation to enterprise-grade legal AI. Tyler Alexander, Director of AI Reliability, ensures that every output meets the rigorous standards legal professionals demand. Together, they work to balance cutting-edge innovation with the accuracy that legal practice requires.
“Lawyers are responsible for their clients’ life, liberty, and property: decisions with real consequences,” says Alexander. “If I’m a lawyer making that type of decision, I need to know that my system is reliable, that [the outputs] exist in reality. That’s one of the biggest issues with using something like consumer-grade AI — you often have no idea where its advice is coming from.”
A fundamental question for lawyers in the age of AI is, as Alexander frames it: “Are you really going to risk someone else’s skin because you received a ‘this sounds good’ output, but don’t know what information it’s based on?”
From a “nice-to-have” to a “must-have”
AI usage in the legal world is no longer optional for many firms. “At this point we are pretty firmly in the ‘must-have’ territory for AI,” says Heagerty. Even two years ago, quite a few law firms still regarded as AI as a technology with potential but felt no urgency to adopt it. Today, that’s no longer the case.
“The efficiencies gained from using the right type of generative AI (GenAI) technology are simply too much to ignore now,” Heagerty says. It is “freeing up time so that lawyers can focus on higher-value work, and they’re able to deliver more to their clients. If you compare a lawyer who isn’t using GenAI, versus one who is, the difference is becoming night and day as to what one is able to accomplish over the other.”
Client expectations also drive this shift. More and more clients want their counsel to use some type of AI, in part because they regard the technology as a way to reduce expenses for standard-issue tasks. During litigation, the stakes become even higher. If an opposing counsel is using AI, they would gain a substantial advantage regarding research capabilities and speed if their counterpart is using outdated methods.
“Whatever resistance still exists to AI is primarily centered around the issues of trust and reliability,” Heagerty says. That’s where the importance of having professional-grade legal AI comes into play.
Why professional-grade AI is in a different class
A professional-grade AI solution like Thomson Reuters’ CoCounsel Legal has a number of benefits for its users. None is more important than its ability to provide multiple layers of accuracy and security.
CoCounsel’s outputs are thoroughly verified by teams of professionals with real-world experience in the legal sector. Further, outputs are built upon the substantive legal databases compiled by Practical Law and Westlaw over decades.
We have to hold ourselves to a higher standard. Our users demand it. We’re often dealing with issues that may end up in court. There are confidential communications to clients. It’s not just a consumer-grade AI assistant.
Vice President of Product for CoCounsel Skills
Building trust through transparency
AI providers continually enhance their systems with better search, broader information, and faster performance. Yet these moves can’t be made at the expense of accuracy.
“One of the things we take very seriously is making it very straightforward for our attorney users to verify where any information is coming from,” Heagerty explains. “There’s so much that’s on the line, so it’s important to give our users the ability to make sure it’s trustworthy.”
Any CoCounsel user should be able to easily locate the source of any piece of information in a system output. This could be a Practical Law or Westlaw-derived case citation or a link to an internal document that the user has uploaded.
Alexander’s team focuses on continuously evaluating outputs prior to launch as part of an iterative process, working to ensure outputs provide the level of verification that CoCounsel users have come to expect. To that end, the team works closely with Product and Engineering colleagues to ensure CoCounsel’s backend produces outputs reliable for legal practice; free of harmful hallucinations and grounded in verifiable sources from Thomson Reuters’ databases and user-provided content.
Tone matters as much as accuracy. CoCounsel is designed to have an “objective” tone when responding to queries. This stands in marked contrast to some public AI systems, where the tone of the AI “voice” has been criticized for being overly confident, regardless of the accuracy of the system’s output.
For CoCounsel, should there be any doubt about information in a particular case, the system is meant to voice this uncertainty, asking the user for more information and specificity, not papering over any gaps and offering an unverified answer.
That said, flexibility remains important. “There are going to be use cases, such as in the drafting area, where you might want to choose a tone of voice that will be more persuasive, or one that will take a more adversarial position,” Heagerty says. Having CoCounsel employ a more “oppositional” tone is useful, for instance, if a lawyer is using the system to write a letter threatening legal action, or drafting deposition questions.
Balancing innovation with reliability
The legal technology sector must continue maintaining the pace of innovation while ensuring absolute reliability. “We definitely want to make sure that we are delivering cutting-edge technology,” Heagerty acknowledges. “At the same time, it’s incredibly important that we make sure that what we are providing is accurate and meets user needs and not just jump to the newest model immediately without going through that verification process.”
This approach may mean moving “slightly slower” than some competitors, but Heagerty believes this deliberate pace is essential. “Coming from Thomson Reuters and the trust that we have, it’s very important that we make sure that we maintain that confidence in what we’re putting out.”
We need a high degree of confidence in what we ship is suitable for legal use. At TR, that means extensive evaluation by technical experts on the machine learning and applied science side, as well as the hundreds of experienced attorneys and attorney editors who work on Thomson Reuters’ products.
Director of AI Reliability
Building a path ahead
There’s great potential for the development of agentic AI systems. Here, the goal is to evolve a law firm’s AI from being regarded as an office assistant to become more of a true “co-counsel.”
An agentic AI system should be able to quickly grasp where a set of queries is leading and respond in turn. The aim is for the system to help a lawyer make more informed decisions by offering suggestions or tactics that they might not have come up with on their own.
Alexander describes agentic systems as being the “next generation” of AI, “where the systems are capable of making decisions autonomously with minimal instruction, within parameters capable of being set and controlled by the user.” As always, this development will have to be grounded within an even greater level of internal verification standards, so that users remain confident in the outputs that an agentic AI gives them.
“We want to preserve that explainability,” Alexander says. Users “want to understand the decisions the AI is making, and what information it’s relying on. It’s important to know what steps it’s taking and what it looked at to reach its conclusions.”
Working for you, not the other way around
Another goal for CoCounsel teams is to make the system even more intuitive. A greater fluency of use is essential to encourage more widespread adoption of AI technologies within the legal workspace.
“There’s a lot of things we’re doing to have CoCounsel be as intuitive as any consumer-grade application,” Heagerty says. “We’ve built custom workflows to allow for more specialized outputs: things that, in the past, might have required a more expert level of familiarity with the product. We wanted to open that up to any user, even one who’s brand new.”
Training becomes more effective through personalization. CoCounsel teams continually refine the system’s prompt library, crafting prompts to guide users to information that they’ll need for any given task.
“From a training standpoint, one thing we’ve found that’s important is including examples and demonstrations that will match an attorney’s own practice,” Heagerty says. If the lawyers are in-house counsel at a real estate firm, their training should include multiple real estate-related examples, “which makes the system much more accessible.”
The ongoing development of CoCounsel is “really opening up a world of capabilities to all our users. If you can dream it, we can do it,” Heagerty says. “Some of that is relying on the advantages we have, [such as] using the expertise from Practical Law in crafting execution plans. It’s looking to the authority in Westlaw to know which questions [the system] should ask when documents are uploaded as part of a deal, or as part of a case file.”
As Alexander says, “the system adapts to lawyers, and not vice versa. Instead of you having to learn how to use CoCounsel, we want it to understand how you work.”
The future of legal AI
Alexander predicts that the upcoming year will be “a watershed moment,” both for CoCounsel and for overall developments in AI.
“We’re removing barriers, making it a lot easier to access the technology and use it to accomplish things that, in the past, weren’t easy to accomplish without needing structured and expert prompting,” he says. Given the promise of agentic AI, “it’s going to be an exciting time for legal technology.”
Indeed, AI technology is evolving so swiftly that even medium-term planning is becoming a challenge in legal tech. “Not that long ago, building a roadmap for the next year or two was pretty much par for the course,” Heagerty says. “But the pace of innovation is such now that it’s become more difficult to do that, to make predictions. I think people are going to be continually surprised as to the extent of what’s possible.”
Learn more about CoCounsel Legal and experience professional-grade legal AI built for your practice.
CoCounsel Legal
Work smarter with CoCounsel, seamlessly integrated with Westlaw, Practical Law, Microsoft 365, and DMS partners
Go professional-grade AI ↗
