A knowledge manager at a mid-size firm gets a message from the managing partner: can we add AI to the research workflow before the end of the quarter? No budget line, no headcount, no discussion of what the research workflow currently is or who owns it. Add AI. She has been here before, with document management, with e-discovery, with practice management software, and she knows how it tends to go. The tool arrives, the training session happens, and the firm discovers six months later that what they built was a faster way to produce output nobody fully trusts.
Most firms announcing an AI officer are announcing a version of that dynamic with a title attached. The structural choice embedded in what Gilchrist Connell announced last week is worth more attention than the title itself.
On the surface, the news looked routine: seventeen promotions across the firm, a new senior appointment, a signal of confidence in the team. But the mandate Gilchrist Connell designed around that appointment repays examination. They folded AI, knowledge management, and legal learning under one brief, held by a principal with two decades of litigation experience. CEO Belinda Cohen described the firm's direction as practical, disciplined, and client-led, without compromising on quality or judgement.
That last phrase is doing real work. Plenty of firms have created AI titles in the past eighteen months. Most of those roles sit adjacent to practice leadership, a technology function responsible for vendor selection and internal training, reporting through IT or operations. The implicit theory is that AI is a tool, like a case management system, and the job of the AI officer is to deploy and maintain it. The tool and the institutional knowledge that makes the tool useful are treated as separate concerns.
Gilchrist Connell's model is built on a different theory. When you put AI, knowledge management, and legal learning inside the same mandate, you are making an argument: the value AI delivers in a law firm depends on what you feed it, how you organise what the firm already knows, and whether the people using it understand enough to trust it or override it. A principal with twenty years of litigation experience knows where legal reasoning goes wrong. She knows which questions a research tool can answer and which ones require a call to a senior partner who handled a similar matter in 2017. That judgement, applied to an AI function, is worth more than any amount of prompt engineering.
The dominant pitch in the market has run in the opposite direction. Speed: faster research, faster drafting, faster review. AI sits beside the lawyer and hands them output. The problem with that model runs deeper than speed. Speed is useful. The assumption underneath is that output is self-validating, that if the AI produces a plausible-looking answer, the lawyer's job is to check it before filing. That assumption shifts the burden of judgement from the tool to the practitioner, but strips away the context the practitioner needs to exercise it well. A litigator who knows the firm's matter history, who understands which sources can be trusted and why, who has watched AI get legal authority wrong, is in a different position from one who received an AI-assisted draft with no grounding.
Grounding is the central design problem, and it is where the Gilchrist Connell structure makes its argument most clearly. A general-purpose language model is trained to produce fluent, plausible text. Legal research requires something different: answers traceable to authoritative sources, that reflect how Australian courts treat a question, and that flag the limits of what can be known as clearly as they communicate what can be established. Those two things are in tension, and the tension cannot be resolved by adding a disclaimer. It requires a closed corpus of legitimate Australian legal sources, a search approach that retrieves before it generates, and citations that lead back to real documents rather than paraphrased summaries of documents that may or may not exist. The firms that grasp this build accordingly.
For the Gilchrist Connell model to work at scale, turning the AI and knowledge brief into an actual practice advantage, the tools underneath it have to be built for the grounding problem rather than around it. Habeas grounds every answer in verified Australian legal sources, with citations traceable to the source document. Foundational research that used to occupy a full morning can now be completed in minutes, which means the judgement layer, better informed and more quickly reached, is where the practitioner's time actually goes. That is precisely the benefit a unified knowledge and AI brief is designed to capture at scale.
Cohen's phrase, "without compromising on quality or judgement," functions as a design constraint, not a reassurance. Design constraints require someone to own them. Gilchrist Connell's structural choice encodes that distinction in an actual job description: AI is a knowledge problem before it is a technology problem, and the person responsible for knowledge had better understand what legal knowledge requires. The firms that treat those as the same question are working on the right one. The rest are still waiting for the training session to pay off.
If you want to try for yourself or get in contact, book a demo with us here. We also offer the capacity for self-serve individuals to sign up, and subscribe or register a free trial at app.habeas.ai.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
