For the past year, the legal tech community has been transfixed by what looks like a heavyweight title fight between Harvey and Legora. These two platforms have dominated the conversation in Big Law, commanding eye-watering valuations and firm-wide rollouts that have been announced with all the fanfare of a new practice group. But I think the industry is asking the wrong question entirely.
The real shift happening right now is not about which platform wins. It is about whether we should be bundling our AI at all.
The Wrapper Problem Nobody Wants to Talk About
The more honest practitioners and CTOs will tell you what they have been saying in forums for months: Harvey and Legora are, at their core, wrappers. They are polished interfaces built on top of general-purpose large language models, and that fact carries significant implications for anyone who has committed to a per-seat contract priced in the thousands of dollars annually.
As frontier models like Claude become more sophisticated and introduce native enterprise features, the value proposition of a premium third-party AI layer starts to look increasingly fragile. The firms that signed multi-year deals on the assumption that these platforms had some unique and defensible intelligence baked in are beginning to wonder whether they bought a capability or simply a brand. The answer, in most cases, is closer to the latter.
What I Think Replaces It
My view is that we are entering an era of deliberate unbundling, and the firms that recognise it early will be significantly better positioned by 2027.
The architecture that makes sense is what the technology world calls best-of-breed. Rather than leaning on one platform to handle everything from legal research to contract review to client-facing work, the smarter approach is to build a modular stack where specialised tools do specific jobs well. A firm’s case management system and document management system become the true centre of gravity, not a vendor’s proprietary AI environment. Targeted tools built for specific practice areas, whether that is intellectual property, corporate drafting, or litigation discovery, tend to deliver better returns precisely because they are designed around the actual workflow rather than trying to anticipate every possible one.
The piece of this picture that I find most underappreciated is middleware. There is growing infrastructure emerging that allows firms to connect their own secure data directly to frontier models, bypassing the need for a vendor intermediary altogether. This is where I would be placing my attention if I were advising a firm today.
The Data Moat Is Overstated
The big legal data houses will tell you, with some justification, that their decades of proprietary data give them an insurmountable advantage. I think that argument holds up in a narrow lane. Where it falls apart is in what practitioners actually spend most of their time doing: synthesising information against a client’s specific internal documents, drafting based on prior matter context, and reasoning across a body of work that lives inside the firm, not inside a licensed database. That synthesis layer is, at present, wide open. It is where I expect the most consequential legal tech innovation of the next two years to occur.
What Firms Should Actually Be Doing
The firms I respect most right now are not the ones who have committed to a single platform and are waiting to see results. They are the ones running multiple tools side by side, comparing real adoption rates and real output quality rather than taking a vendor’s word for it. They are insisting that any AI tool they adopt connects to their existing systems of record rather than creating a new data silo. And they are quietly investing in AI literacy at the associate and partner level, because a firm that cannot identify a hallucination or write a competent prompt is going to be at a disadvantage regardless of what tools they buy.
A Final Thought
The Harvey versus Legora debate will probably continue for another year or two because it is a convenient narrative and there is a lot of invested capital on both sides pushing it along. But I think it is a distraction. The firms that come out of this period strongest will not be the ones that picked the right platform in 2026. They will be the ones that decided, early enough, to own their own stack.
As I have heard it put more bluntly: the establishment will keep talking about security and governance frameworks while the savvy firms are already building the infrastructure to run these models on their own terms. The data incumbents hold the archives. The AI companies hold the headlines. But the firms that choose their own architecture will hold the advantage.
This piece reflects the author’s opinion on emerging trends in AI legal technology.
Malcolm Pearson
Editor: Tech4Law
