'Legal AI is dead'.
So said Max Junestrand, CEO of Legora, at the company's 'Precedent' event in London. Specifically, Junestrand unveiled the new Legora Agentic Operating System ('aOS'), which purportedly represents a 'generational [shift] from assistive AI to agentic execution, redefining what a lawyer can achieve'. Thus, Legora is posed as the trendsetter, rewriting the narrative on what a lawyer is capable.
Yet, equally, Harvey's new launch of 500 ready-to-use agents was declared the end of the 'legal assistant era'. In another domain, Anthropic reportedly held back release of their new 'Mythos' model on safety grounds, citing capabilities so advanced that they gave pause to the people who built it.
Each of these three announcements was marked by a doomsday framing, where subtle or overt, making the previous thing sound inadequate because the new thing is scarily good.
It's worth identifying what is marketing 'noise' and what is legitimate cause for celebration/concern.
Declaring a category as 'dead' is a specific marketing move. When Legora says legal AI is 'dead', what it means is that the version of legal AI that isn't Legora is dead. The version that is Legora - an agentic operating system, with agents that handle work from matter intake through to client delivery - is what comes next.
Harvey's framing is more architectural. It calls itself 'the operating system for legal and professional services.' Operating systems are infrastructure, and infrastructure is unavoidable. The implication is that choosing Harvey is less a software decision than an inevitability, being the kind of thing you adopt rather than evaluate. These framings work because they reframe the competitive question away from 'is this tool better than the alternative?' and toward 'am I on the right side of history?'
Harvey's headline figure of 500 agents is also a number designed to feel comprehensive: a library so large that whatever your workflow, there is an agent for it. The agents were built and tested by lawyers, the announcement says, alongside design partners including Allen & Overy, Gowling WLG, and Clayton Utz.
But there is something that does not change across any of these announcements, and it is a crucial question for a practitioner in an Australian firm to answer before trusting the output: what legal corpus, case law, and regulatory environment is the model trained on?
Clayton Utz is an Australian firm, so Harvey has Australian design partners, and that is worth acknowledging. Yet, design partnership is not the same as jurisdictional calibration. Having an Australian firm help test agents built on a predominantly US and UK legal corpus is not the same as building a product that reflects how Australian law actually works: its statutes, its regulatory texture, its body of case law, the way its courts reason and document their decisions.
A model that has not been trained on Australian legal material will produce outputs that are structurally plausible and jurisdictionally unreliable: instead of being dramatically wrong, they might be confidently imprecise in ways that clear first review.
The announcements keep coming because the market is moving fast and attention is scarce. Doomsday framing cuts through. 'Legal AI is dead' performs better than 'we shipped a feature', and an operating system metaphor sounds more essential than 'we added 400 more agents.' However, we at Habeas have a commitment to measured and authoritative marketing, because we believe that product capability speaks louder than headlines.
Tools work best when adopted without friction into an intended user's workflow. Habeas was built for Australian law: as the starting point, rather than being a supplementary layer on top of a US or UK foundation. The training data reflects Australian legislation and case law. The workflows reflect how Australian practitioners actually work. When the tool produces an output, an Australian lawyer can evaluate it against the law they practice.
The operating system metaphor that Harvey and Legora have both converged on is compelling for a reason: an operating system is infrastructure, sitting beneath the work rather than alongside it. Still, an operating system only runs the software its environment actually supports. Harvey and Legora are building operating systems for legal work broadly. Habeas is building one for Australian legal work specifically.
