Robert Ambrogi Blog
Yesterday, Thomson Reuters made two announcements. It opened early access to the next generation of CoCounsel Legal, the most substantial reworking of its flagship legal AI assistant since the company acquired Casetext in 2023, and it released its Future of Professionals Report 2026, which warns of a widening gap between AI adoption and the value organizations are getting from it.
After those announcements, I had the opportunity to speak with Thomson Reuters President and CEO Steve Hasker about both, in an interview in which he also talked about the company’s development of its own LLM, called Thomson. Below is a transcript of that conversation, which I have edited for length, style and consistency.
Ambrogi: I have a couple of questions about CoCounsel and a couple of questions about the report you put out this morning. Regarding CoCounsel, could you talk more about the “why now.” You’ve accelerated the release of this. What’s changed that’s given you the confidence that this is ready to put out there?
Hasker: A couple of things, Bob. When I step back from it, this is the first time since we acquired Casetext that we’ve truly unlocked the value of our content — so Westlaw, Practical Law and so forth — and, importantly, the expertise of 2,700 attorney-editors in training the agent. The prior version of CoCounsel didn’t do that. So in our view — but, more importantly, in the view of the customers that have been in the beta now for a month or so — this is a huge step forward, and it’s a huge step forward for the industry relative to anything else they’ve seen or been looking at.
The reason that now [is the moment] is twofold. One, this onset of agentic AI, agentic tools, has been, I think, as big a step forward as the early generative AI breakthrough. And [two], the reason we’re able to move quickly and get it into the limited access starting this week is that it’s built with the same architecture and thinking behind Westlaw Advantage — and, Bob, that’s been a big success with our customers. So it’s the same technologists, researchers, scientists, lab folks, in many cases the same product and engineering people. They had a successful foray with Westlaw Advantage that launched last summer, and we’ve … ported [largely the same team] across to rebuild CoCounsel.
The reason we’ve moved it forward is it’s just performing so well. … This is the most successful beta trial I’ve ever been a part of, by a long way, in terms of the feedback on the utility of the product, the accuracy of the product, and also the reliability. … So that was the basis on which we said, “Let’s go.”
Ambrogi: Sounds like one of those medical trials where the results are so great that they say, “Let’s stop the trial and put it out there.”
Hasker: Not to confuse what we do with saving or prolonging lives … , but that’s a good analogy. It is a medical trial that’s gone so well we’ve said, “All right, let’s go.”
Ambrogi: How much of that has to do with partnering with Anthropic to use the Claude SDK to develop the product? How critical was that to bringing it to this level of viability?
Hasker: It’s a part of it, Bob. We’ve built our agentic AI products to be model-agnostic, so the idea is we’ll use the best available models — and the SDKs built thereon — to innovate our flagship products. We felt in the lead-up to Westlaw Advantage that the Claude models were the best for the things that we do, and, based on all the evals and testing that we continue to do on basically a weekly basis, in our view that continues to be the case for our specific purposes.
We’ve also found the Anthropic team, and particularly the Anthropic engineering team, to be very easy to work with, and the productivity has been high. As much as anything, that’s been chemistry between the teams. I’ve observed tremendous chemistry between the lead engineers there and the lead engineers here, and, as you know, when you’re inventing, that can be as important as anything else.
Having said that, we’ve now put out — or at least are in advanced testing of — the first version of our own Thomson model. A couple of years ago we acquired a research lab called Safe Sign Technologies, which was centered around Cambridge University. We’ve extended that to Imperial College in London, and we’ve built our own large language model for legal. It’s imaginatively called Thomson. And it’s starting to outperform all of the latest models for specific legal tasks.
We haven’t decided what we’ll do with that yet, Bob, but we think it affords us a degree of flexibility. So we may have some or all aspects of CoCounsel running on Thomson and others running on the Claude SDK. We just think it’s important to have a great working relationship and partnership with Anthropic, but also the ability to exercise some flexibility should we need [it].
Ambrogi: I wanted to ask you about this concept of “fiduciary-grade AI.” That was a big part of the presentation this morning, and it’s also discussed in the report. The concept is clear, and I get it. But do you think it is something that distinguishes you from other legal products on the market? Is fiduciary-grade AI a distinguishing feature, or is it something that should be common to all legal products on the market?
Hasker: I think it is something that should be common to all legal products but is not currently. Bob. I’m sure you keep tabs, as we do, on the public cases of hallucination. We’ve had three, just in my recent memory, in the last couple of months. Extraordinarily successful and well-regarded firms using either a frontier model or a startup built on a frontier model, without access to the authoritative content, without access to the in-house domain expertise, hallucinate. And in at least one or two of the examples that I’m thinking of, the opposing counsel had access to our tool and were able to run the brief or the opinion … through our tool and identify the problems, and then make the judge aware of [them]. So the answer is that anyone who’s in and around the legal AI space should meet the fiduciary-grade AI standard, however the industry decides to define that. But the fact of the matter is that today, that is not happening.
And the reason it’s not happening is [that] without native access to the authoritative content and the in-house expertise to train the agents, the risk of hallucination goes up, and therefore the risk of these problems becomes greater. And that, in our view, does not constitute fiduciary-grade.
The other two things that we have that we think are distinctive, Bob — one [is] our data-governance standards. You might have heard us say that the proposition we put in front of our customers is that their input will not become part of our AI output. In other words, we’re not training our tools or our model based on our customers’ input. Now, that’s a little bit antithetical to the Silicon Valley ethos.
The other one that’s antithetical is our customer support. If you’re using Westlaw Advantage [or] CoCounsel [and] it produces [an] answer that you don’t understand and you’re up against [a] time crunch, you can pick up the phone and speak with a reference attorney, and that person can run a mirror image of the query in the tool and explain exactly how it’s reached [its] conclusion.
One of the things that’s striking when you look at CoCounsel is that it’s extraordinarily transparent as to the steps it’s undertaking in sequence and in parallel, and extraordinarily transparent in terms of the source for each step. If it runs a particular check and reaches a specific conclusion, it will cite exactly where from [among] the information resources it used to reach [that] conclusion. So it’s about as far from a black box as we could currently design.
We think that’s important for two reasons. One, it helps gain confidence around this fiduciary-grade AI standard. And the second is, one of the things that we hear about … is what happens to the apprenticeship model for young lawyers, whether they’re coming up through a law firm or even going earlier into a general counsel’s office? How do they learn the craft, and how do they become lawyers with lots of miles on the clock and judgment? We think that a tool that is the opposite of a black box — that actually walks them through all of the steps, with the citations built in and all of the sources [and] reference material — could be an incredibly important part of the next generation of an apprenticeship for lawyers. And that’s why we’ve built it the way we’ve built it.
Ambrogi: One other quick question on CoCounsel: the competitive positioning. Obviously Harvey and Legora have gained huge traction over the last year in particular. As the market looks at CoCounsel against that backdrop of Harvey and Legora and other competitors, what do you see as the key differentiators, the key qualities, that position your product as unique in the market?
Hasker: We think that ours at the moment is the only one that reaches this fiduciary-grade standard. The reason for that is the native access to Westlaw, Practical Law and all of the reference attorneys and attorney-editors. And we think that, for the first time, [the] CoCounsel … output makes that clear, both in terms of its accuracy and in terms of the complexity of task it’s able to understand. As I said, it’s the first time we’ve really unlocked the value of the content and the expertise for the benefit of the … tool.
And that’s been the basis of the feedback we’ve gotten from customers — so this isn’t just Steve saying this. This is beta-test customer after beta-test customer coming forth and saying, “I haven’t seen something that’s able to perform this level of task, this level of sophistication, at this level of accuracy.” I’m not here to denigrate Harvey, Legora or anybody else, but the fact of the matter is those players do not have access to those content sets or that in-house expertise.
Ambrogi: I want to turn to the report. One of the key findings seems to be that AI adoption is no longer the issue — it’s more the fact that the organizational strategy at law firms, legal departments and other professional organizations is not reaching the day-to-day work. Based on the data and talking to customers, what do you think separates the organizations that are effectively operationalizing AI from those that are merely experimenting with it?
Hasker: I don’t think this should come as a big surprise, … change management’s always the hardest thing. Getting young executives, whether they’re lawyers or any other form of knowledge worker, to play around with a cool tool that helps them produce some work product in a fraction of the time and opens up some new avenues of thinking is not that hard. And so it has proven this time — you would argue even more so this time, because of the power of the frontier models. So I’m not all that surprised that firms are using a bunch of different tools, that attorneys of various tenures are using a bunch of different tools. And I’m also not that surprised that the underlying change management required to get a return on that firm-wide is lagging.
I had one managing partner of a major firm here in New York say to me, in answer to this question, “I think the tools are 18 months ahead of the change management.” In other words, the process of rethinking the workflows, rethinking potentially roles, rethinking how performance is measured — doing that hard work of rewiring rather than just adding a tool on top of an existing process — is, in some way, shape or form, yet to come.
Now, the firms that are ahead — and we’re working with a number of them — have actually taken the last 12 months to start that process. Within a particular practice area, they’ll be saying, “Here’s how much of the underlying initial work — whether it’s an initial draft [of] a motion or whatever it might be — could be produced by the tool. And here’s how we think about the first time [an] associate gets involved, [the] interaction with a more senior attorney. At what point is it ready for distribution to a client? How much transparency and labeling are we providing in terms of how much was produced by AI, and where the team’s involvement has kicked in?” I think the firms [that] are a bit further ahead are those that have started to think that through, practice by practice, whereas the majority of firms are still in that, “Let’s make the tools available and see what happens” [mode].
Ambrogi: What about the client side of this equation? The report says that 78% of corporate clients say AI-enabled quality improvements are very important or essential, but only 6% say that their outside counsel are delivering on that. What do you make of that? I’m assuming this is saying that AI capability is now a key procurement issue — somebody made that point on the LinkedIn Live this morning. Why aren’t firms getting this message, or why aren’t they responding to it?
Hasker: I think it’s still relatively early days. You have a portion of clients who are saying, “We forbid you from using AI in the generation of work product on our matters,” because they’re in highly regulated industries or have a particular approach to it. We’ve got to respect the fact that there are still some industries and some firms in that category. It is by far and away a … minority, for sure, but they exist.
The key point, though, is that both the general counsel and the firms, by and large, have access to the same tools. So the in-house teams are saying, “If we’re accessing these tools and starting to see some benefits, or at least we’re pushing into this, then our expectation is the firms will do that themselves.” And I think they see the same thing that you and I do, which is that the firms are using these tools, experimenting with different tools. What the finding says is a version of what we just talked about, which is the clients are saying, “You’re using these tools. Where’s the benefit to me, in terms of turnaround time? Where’s the benefit in terms of quality? And where’s the benefit in terms of cost? And what’s the path to my seeing those [benefits]?”
I don’t know whether you’ve noticed this, but we recently completed our 21st year of the Legal Executive Briefing at Pebble Beach, where we had 125 managing partners attend, and before that we had 75 general counsel. And I noticed a pretty marked uptick in assertiveness from the general counsel. The majority of general counsel in the prior year — the mood music was, “Look, I deal with important transactions, important litigations and other things. And when I go to the outside market, I’m looking for the best result. I’m the chief risk manager of my corporation. So if it’s Latham & Watkins or Sullivan & Cromwell or Pinsent Masons or something else, that’s [who] I’m going out to.”
This year I noticed much more of, “No, I’m going to consolidate the number of firms that we use. If we used 100 firms last year, we’re going to consolidate down to a panel of 10 or 20, and we’re very thoughtful about which ones qualify for different types of areas and matters. And we expect the benefits of AI to kick in [in] terms of the billing. And we’re a little tired of the rate increases that we’ve seen.” So I sense — frustration is the wrong word — but a bit more assertiveness from the general counsel. And you may say, “Well, it’s not a statistically robust [sample of] 75 …” But I reckon they’re starting to say, “These AI tools — given that we have the same access that the firms do, by and large — we need to start to see the benefits come through in terms of quality, speed and cost.”
Ambrogi: That sounds like an insightful observation. I’m certainly hearing and seeing similar themes out there. I thought it was really interesting that the report outlines three different future paths: AI to elevate, AI to scale, and AI to reimagine. I’m curious what your thoughts are as to whether there is a particular path that firms should be gravitating toward [among] those three.
Hasker: A couple of things. We’re not in the business of advising law firms how they should manage their business and think necessarily multiple years ahead. But what we were trying to frame up was a couple of different ways of thinking about the way in which the tools might [be] use[d]. I thought the team did a good job of framing it up. What I would say, though, Bob, is that I’m in the optimistic camp here.
It seems to me that the legal industry has created this incredible profession, and this means of economic activity, by figuring out how to [put] people in the context of a particular client matter — and that this technology, like those that have come before, will be in service of that. So the firms that win will be those that say, “We have a very clear view as to how we are going to continue to put the best people in front of our clients at the right time for specific matters, and the technology will help those people be more productive, be more cost-effective, and move with greater speed.” I think [turning] it the other way around, from a law-firm perspective, risks taking [firms] into territories that they’re not necessarily equipped to [enter].
The example I would use: On the tax side of our business, we also serve the big four tax and accounting firms, and the number of times where they’ve said, “We’ve built this new technology,” with or without Thomson Reuters’ information or software, and then a year, two, three years later we get a call saying, “Hey, would you like to buy it? Because we’re a professional-services firm and that’s what we’re world-class at. We’re not a software-development firm, and we’re sure as hell not a software-maintenance firm.” I’m not here to predict that will happen, [but] the big four are bigger, with more in-house tech talent, than the law firms — even the largest law firms. So it might be part of the future.
Secondly, I think that AI will lead to an uptick over time in demand for legal services. If you think about business formation, restructuring, litigations around copyright and all of the different issues, I personally think this will be like every other big tech breakthrough, where you have a pretty marked increase in the level of activity. The question is how much of that is performed by machines supervised by people, and how much is performed by people. So the P&L of a firm may change: their tech budget may go up, their people budget per dollar of revenue [may] come down, and their real-estate costs per dollar of revenue will go down.
Those are two points of optimism, because it’s not hard to find the doomers who say, “This is the end of the glory days of law firms,” and so on. Notwithstanding the economic cycle, I think AI is going to be highly accretive in terms of demand for legal services. And I think the firms that win will be those that assemble the best people and use the technology to support that process.
Ambrogi: With CoCounsel available today, are those firms going to need fewer people involved?
Hasker: It certainly enables any particular practice team or individual lawyer to get to an extremely accurate answer in minutes. So it’ll enable that lawyer to do more work. The question is, what does the demand curve look like? There are a lot of lawyers, Bob, as you know, who don’t get home at night at a reasonable hour. Maybe it’ll enable them to do that first.
Ambrogi: Steve, thank you. I really appreciate your time.