Is AI about to get too expensive for lawyers?

Hello and welcome to the Lawyer Podcast. I'm Catherine Griffiths, the Lawyer's editor in chief. And I'm Christian Smith, the Lawyer's litigation editor. Just as the world of law was getting carried away with the excitement of AI, legal tech firm Legora has brought us all thudding back down to earth. Jude Law's AI of choice has told his clients that it plans to move to consumption based pricing. We'll get on to what that is in a moment, but it's a decision that could fundamentally change how law firms use AI right as its role is becoming normalized. So on this episode of the Lawyer Podcast, we return again to the question of legal AI. And we're joined by our tech editor Ben Lucas, who broke the Legora story and our tech reporter Clementine Doyle to ask just how big a deal this is, what it means for the market and what it might mean for lawyers day to day jobs. Ben, you broke the story about Legora changing its pricing model. Talk us through what's going on. So I spoke to several innovation chiefs that said, Legora had spoken to their firms in recent months about the legal tech's intention to change its billing model away from seat based pricing, which sees firms charge based on the number of users towards consumption based pricing, which is based on usage. And this is like to happen when their contracts come up for renewal. Now when the lawyer went to Legora for comment, it confirmed those plans. And it said that the need to switch was being driven by more of its functionality being agente. Now this move comes as the AI giants, such as anthropic and AI, have also switched some of their products to consumption based pricing. Now the kind of the reaction in that story that I managed to pick up is that many innovation leaders that I spoke to were kind of accepting of the change, you know, expected that this was going to happen, you know, somewhat inevitable. Now the foundation models are priced in this way. But one or two have been kind of shaken by the way, I just wanted to read one of those quotes. This is one innovation head that was informed about Legora's intention to change its billing model said and I quote, "It hasn't been received well by the board members and it makes my job harder and it strains the partnership approach we all want with our tech providers." They went on to say quote, "We are not ignoring their cost pressures, but we are evaluating alternative providers and considering reprioritizing Legora." So it's really shaken some firms even if others are sort of accepting and ready for the change. So a little bit of a mixed reaction at best to Legora's move there Ben. You mentioned that it's switching its models because of agente AI. And I want to bring in Clem here because it is this whole issue of agente AI is what is powering an awful lot of this explain in not too techy language please. What is agente AI and actually why would this actually change its change Legora's and other techs pricing models? So there are a few factors at play here. Basically the subsidy period is ending. So for the past two, three years many AI companies in this goes beyond legal. They've effectively subsidized customers usage. The VC funding was abundant, we've seen that with Harvey and we've seen that with Legora and the priority was winning those enterprise contracts and building market share. But now the expectation will be to demonstrate sustainable gross margins and a credible path to profitability. Investors are asking whether revenue keeps pace with infrastructure costs. Unfortunately, as we've seen, the economics of AI have proven very awkward. Each question and each prompt sets in motion an immense volume of computation which requires an LLM large language model to perform billions and sometimes trillions of mathematical operations before a single sentence even appears on your screen. That work takes place on specialized processes known as GPUs, graphics processing units are many supplied by Nvidia. And these chips are scarce and they're expensive and a single prompt may occupy in the sliver of their capacity for a matter of seconds but multiplied across millions of requests a day which you can get in a law firm. They become one of the company's largest expenses or they can become one of the company's largest expenses. So when a lawyer asks an AI assistant to summarize a multi-hundred-page contract, the model must read the prompt and hold the relevant information in its memory and then generate each successive token of text in sequence until the answer is complete. None of that exists in advance so it is carried out a fresh for every single request. And as one might imagine, those calculations guzzle electricity really on a next ordinary scale and modern AI data centres have become some of the world's fastest growing consumers of power. I know Katty wrote a great horizon about our data centres but with each interaction it draws energy that has to be generated and also cooled. And so the cost rises with the ambition of the task, rewriting an email may only require relatively little computation but reviewing thousands of contracts or searching multiple document repositories or completing multi-stage agentec workflows can trigger dozens or even hundreds of separate model calls each consuming additional compute. So from a vendor's perspective every interaction carries a price. I've referenced agentec, you've referenced agentec, we are moving into the agentec era of AI and that becomes computationally intensive too and each of those capabilities requires substantially more inference than a simple chatbot conversation. What do we mean when we say agentec that refers to AI systems that can decide and execute a sequence of actions to achieve an objective. So we're shifting from reactive A.K.A. we put a request to a chatbot that we've been used to and you know it comes, the robot answers you. Agentec, AI or AI agents are much more proactive so you give it an outcome and it determines the steps it needs to take to get there. So we've gone from summarise this agreement to review this agreement and identify every clause that differs from our standard playbook, flag anything unusual, draft fallback language, compare it with previous deals, produce a report for the partnering question. So in that sense agentec shifts the lawyer's role to increasingly supervising and exercising judgement over an AI generated output rather than performing every individual task. So that's really interesting because I've sort of been wondering what agentec is for a while and you're sort of saying that it's just a significantly more complex way of the AI working and it's able to do multiple steps at a time rather than basically what I use it for which is planning holidays and alternative to Google. You've actually made me feel slightly guilty because now that you pointed out the shortage of chips and things like that I'm sort of wondering if I should be using AI so much to plan my holidays and maybe I should be leaving that for like vital NHS or law firm related activities but we'll figure that out another time I think it's fine. Ben Lagora has already started doing this in some respects. I think I suppose two questions is how long do law firms really have until this becomes the norm with Lagora and then also are we seeing it elsewhere? Is Harvey doing it? Are other legal texts planning on doing it? Doing it? What's the story there? How wide an impact will this be? So from the conversations I've had it seems to be that this change will be coming in when their contracts come up for renewal. Now with regards to other legal tech providers so it's heavily rumoured that Harvey could follow in Lagora's footsteps but we should say that the company declined to comment when we went to them. For comment if they were going to change their pricing model but I did think it was interesting recently that the CEO of Harvey Winston Weinberg revealed about their token usage over the last few months and they said in January it hit one trillion but in June they expected it to be 12 or 13 trillion so that gives you a sense of that expansion there and so just to be clear tokens are the unit of data AI models use which is relied upon to price AI usage but smaller legal techs have come out to kind of clarify their position I've been speaking to them over the past week. Law of view you know that's another legal tech it's got clients including PWC, Deloitte, Violea. They said they're going to be moving towards consumption-based pricing explaining that AI capabilities now increasingly tied to usage. Others have said they're sort of introducing a consumption-based layer alongside their other offerings. Some have been doing this for a while so it feels like there's a hybrid model emerging and meanwhile others are kind of keeping an eye on this trend. You know, but I think it was Thompson Reuters, which is the owner of Co Council Legal, used by firms like Moblebond Dickinson, Gibson Dunn etc. They're keeping a close eye on it. And can be like Clio, which is valued at I think five billion and it's latest funding round. It uses seat-based pricing at the moment, but their CEO said quite law firms should be very clear about what's coming. So this change in the economics of AI is coming and that's going to be impacts in these legal texts across the market. And interestingly, I mean, some legal texts have already been doing its irreloved to you, for example, already does consumption based pricing. So it's not completely unheard of. So I have a question for you both about what it means for, what it means behaviorally, so far in law firms. You know, up till now, sort of this in these early days of AI adoption, you know, you saw an awful lot of firms who are incentivising their staff to have a target number of prompts to use every year to absolutely embed it in their workflows. And it was really very much part of a firms learning development framework. It was absolutely about making sure that clients understood that the firms all over this new, wizzy tool. That presumably, in terms of absolute sort of volume, that's going to change a lot because, you know, I can imagine that law firm managing partners and in particular, the CFOs are going to be looking at this with a very, very concerned expression on their faces because this is going to add to the mountain of costs that's already facing law firms in terms of all of their tech use. So what do you think? What's your best guess as what's going to happen with the way that these that these LLMs are actually being used within organisations and actually how to change people's behaviour around how they use them and in terms of really good service to liberate clients as well? So I think it will have a very significant impact, but unevenly across your firms. And the impact will depend on how a firm has adopted AI and how much it ultimately delegates to it. The firms that rolled AI out to every person in the business will be the most exposed. Of course, under a flat per seat license, that strategy made financial sense, but a usage-based model changes the incentive. So it's ironic, really, the workflows where legal AI can deliver the greatest value can also be among the most expensive. Can we add here? We're just all desperate to point out the irony of law firms belly-aking about usage and our our-based pricing because they're getting a massive taste of their own medicine. I wonder actually, if long-term this will spur lawyers on to a more progressive view of actually what bullying could be for clients, but we'll park there there for the moment. Perhaps I'm sure many of our listeners will have very strong views on that. Ben, so yeah, I think Clem touched there on the sort of rising costs. They become harder to predict. I mean, people I've spoken to said they don't have much visibility over how many tokens and AI tools consumes at certain points and the level of transparency around that isn't necessarily there at the moment to manage those costs. So that's a big tricky problem. So you've all sort of saying that it's actually very hard for law firms to know how much AI they're using, so they could be, it's kind of like using data on your phone, isn't it? Like you could all of a sudden use five gigabytes when you're in Spain and be charged £100 million or you might use 50 megabytes and you just have no idea. Exactly, yes. Law firms don't have a lot of transparency over this and I'm sure their innovation managers will have an idea of how much the sort of the more recent large language models use versus the earlier ones, but on a more detailed level they probably they don't with the tools that are there. And that creates a real headache because then there are questions around managing AI use across the firm to keep those costs down, who manages that, who's responsible for that. And I think from people I've spoken to, it means that firms will now have to be more disciplined and selective about when they are deploying AI and what types of large language models they are deploying for certain tasks. The more recent ones that are more token hungry maybe for a bigger piece of work and then for those earlier models maybe for something that requires that. But I think you touched on an earlier cap that there's a real question around adoption of AI as well. We're still hearing that there are some lawyers that are resistant to it or still getting used to AI. But if they're now sort of being told that every prompt or activity you do is going to cost as money that could deter adoption and lawyers that are on 100 being encouraged to be more AI savvy, but then they might get told off or warned about how much they're using. So again, more challenges arising there potentially. Well, there are some even big law firms that still don't use it at all. And I know they're sitting here smuggling and saying I hate to say I told you so. I think, Clemen Ben, you were moderating panels at the Lawyers Inhouse Financial Services Conference this week. And it was very clear from all of the discussion in the room and by the way, those panels were not always about AI. But funny how the conversation drifted back to AI all the time. These are in-house lawyers at financial services, broadly speaking whether it's private equity, financial institutions, insurance companies and so on. And we were taking polls on the day on the expectations that clients have. And in particular the pressure that the GC's, the general council are being put under by their own boards to reduce legal spend and essentially forcing their external law firms to come up with AI solutions really, really fast. And of course to a large extent them themselves using AI internally as well. So you've also got, you know, this client push is not going to go away just because the law firms don't fancy the idea of, you know, hiked bills from Legora or any of the other LLMs that are on the market. That reminds me, Kat, of Revolut and their chief legal officer spoke to us a few weeks ago, saying that they had scrapped the traditional panel model in favor of something more dynamic where they can replace firms if they don't think they're up to scratch. And part of the reasoning for that was they want firms to prove their investment in AI. And if they don't, well, it's see later. Yes, Kat, I think one of the, yeah, the points you're picking up on there is the pricing conversation between clients and law firms. And as you say up until now it's been clients very much demanding their law firms use AI with an eye to reducing bills in the future. Now with this shift to consumption-based pricing, if all the sudden a piece of AI supported legal work by a law firm, you know, that's going to, and that skyrocket because it used, you know, so many tokens and it cost so much more, do they pass that cost on to the client or do they absorb it themselves as part of, you know, a competitive tender when it comes to calculating that. But that's a really tricky decision for the law firm. But on the other side of this, you know, we've got to remember that in-house teams are also using these tools. So they will also be dealing with the shift to consumption-based pricing. Some in-house teams do use Harvey or the Gore or some use other, you know, providers out there that will have the shift to consumption-based pricing. So they're also grappling with this dynamic as well. So whether there's some understanding or sympathy, they're, you know, it'd be really interesting to see how that conversation pans out over the next few months. Let's just broaden this out, I suppose. Is this a sort of market reser? I mean, has LaGoura and some of these AI companies potentially making a big mistake here that they might, as you said at the start, they're being. Some law firms are considering whether they want to continue with LaGoura. Obviously everyone does it, then it's a bit of a different story. But this could give opportunities, I suppose, to smaller legal texts. It could be a huge issue for the likes of boutique law firms that don't have the same ability to manage big changes and costs that other big firms do. What's the long-term impact of this? And I suppose as well, sort of at the top as well that many believe that this was coming or for some, everyone's sort of said, "Oh, this was always going to be a thing." So why has it taken so many as a surprise as well? So yeah, I think you make a really interesting point there. I spoke to David Wang, the Chief Innovation Officer at Cooley, and he seemed to suggest that, you know, if LaGoura and others fail to manage this shift properly, it could potentially cause problems, you know, a premature move to monetize things. That increases how law firms think about these AI tools. And, you know, he made the point that they're rapidly improving open source solutions out there, waiting in the wings. And law firms could turn to those. I say could because, you know, many law firms will have strict sort of data security and compliance standards that won't make a switch to these tools straightforward or easy by any means. But there is a world in which, you know, with this shift, which I think is going to affect all legal texts going forwards. I think, you know, we might see a greater diversity of tools and options being used across the legal market. So some people sticking with those sort of big legal AI tools, some shifting to the alternatives put out for example, by unthropic and opening eye-cloth for legal, for example, which we've seen really shake things up. And then maybe some people turning to open source things in the future. As I said, I think one of the things is this is a shift that's happening now, but these shifts and changes are happening so quickly all the time. So again, I do feel for law firms a little bit. If they react too quickly and too decisively now, they might end up being sort of stuck with a solution that they have to change in another six months time. So I think, again, some of the things I'm taking away from these conversations is that really this should be emphasizing, sort of, having a flexible approach in your innovation program and your innovation strategy, because these changes are just happening all the time, throwing up new headaches and challenges. And you've just got to be sort of nimble and ready to adapt when these things come up. The biggest law firms will probably absorb the cost, you know, a global firm billing $2,000 to $3,000 an hour for a part of time is unlikely to abandon AI, because it's technology-built doubles. But if AI saves even a fraction of a lawyer's time, the economics can still work. And the debate for those firms becomes one of profitability and not affordability. Mid-market firms may face tougher choices if a firm's margins are thinner and clients are more price sensitive, rapidly rising AI costs could become harder to absorb, more difficult to justify, and those firms will more urgently need to answer the questions, you know, does every lawyer need access to the most powerful model? Should only certain practice groups use a genetic? Can we use cheaper models for routine work? Is there a point at which the AI bill outweighs the efficiency gain? And in that sense, AI itself may become tiered. We might end up with basic AI for drafting and summarisation, premium AI for complex legal reasoning, agentec reserved for those really high value matters. But as we say, naturally, it all comes down to the client, right? You know, some may actively expect their firms to use sophisticated tech if it makes transactions faster. And I know some clients are willing to pay for that. I'm looking at Hogan-levels, legal engineering department. They want the Oatkajur, but other clients will refuse to pay those higher fees as we have seen. It seems to me that one of the biggest takeaways, Clem and Ben, is that anyone listening, anyone who is using tech in a particular way now is not going to be using it in the same way in six months' time, because their firms will say, no, no, no, we now have an entirely new policy, and the way that you're going to use it is like this, and the client wants it like that. So I think it's incredibly important for everyone, you know, in organizations where AI has increasingly become embedded, is to be aware that they themselves, Ben, you talked about being flexible and nimble, they themselves have to be flexible and nimble in their own habits and in their own tech use as well, because it's going to change really, really rapidly over the course of the next few months. Right, well, let's wrap it up there. That's actually the third podcast we've done in a row on legal AI and legal tech, which I just think sort of suggests how big a deal is at the moment, and we know from seeing how many of you listen to the episodes, you think it's really important as well. So thanks very much for tuning in. We may or may not have our next podcast on AI, who knows, something huge may also happen in the course of the next two weeks, but in the meantime, you can find out about everything that we've been talking about and more on the lawyer.com, and we will be back again in a couple of weeks. Goodbye. Bye-bye.