Thomsons and Faculti chose to build general purpose legal AI in house. The rest of the Australian legal market is converging on a handful of vendors, with consequences for client data, fees, and competitive moat.
Most Australian law firms say they are racing to adopt AI. The race, on closer inspection, has barely started, and most competitors are running on the same off-the-shelf engine.
A handful of firms have built their own general-purpose AI platforms. Thomsons and Faculti are the only two Australian practices known to have done so, according to the Australian Financial Review's Law Partnership Survey. Nearly every other respondent in the survey subscribed to one of the major off-the-shelf legal AI providers. Harvey, the vendor whose Australian customer list includes Ashurst, is the default reference point.
This is the structural story buried under the marketing noise. The Australian legal market has not "adopted AI." It has bifurcated along a build-versus-buy fault line, and almost the entire industry has chosen the same side of it. The two holdouts look eccentric. The structural consequence is that the rest of the market now shares a common dependency: its feature roadmap, its vendor pricing, and its client-facing AI capabilities, with whichever few vendors have won the deal.
Off-the-shelf legal AI is dominant because it is fast. Harvey and its competitors offer a bundled path: hosted models, document-system integration, and an internal chatbot that firms can deploy without building it themselves. For most firms, the alternative is a multi-year capital commitment that only a few balance sheets can support.
The two Australian firms that chose to build argue differently. Thomsons and Faculti have framed their proprietary platforms as a way to keep sensitive matter data inside their own infrastructure rather than feeding it into a third-party model. In an industry where client confidentiality, legal professional privilege, and cross-border data flows are board-level issues, the build choice is also a sovereignty choice. It is harder to quantify, harder to market, and harder to copy, which is part of why it is rarer.
Buying does not mean ignoring governance. Clayton Utz, which has publicly disclosed its ISO 42001 certification for AI management, is an example of a large Australian firm choosing a buy-side strategy while layering it with a formal AI management system under ISO 42001, the international standard for organizations that develop or deploy AI. ISO 42001 is not a build choice. It is the paperwork that says a firm has decided what it is buying, on what terms, and with what guardrails.
Smaller and specialist firms are elaborating the same split. Wotton Kearney, the insurance and disputes boutique, announced in October 2025 an AI-powered platform built on Appian for its legal matter management and client reporting operations. Barry Nilsson, the national insurance practice, has published a three-year AI plan that runs through capability, governance, and client delivery in that order. Neither firm looks like Thomsons or Faculti in scale, but both are using the buy-or-build question as a forcing function for how they want their practice to look in three years.
The market consequence is that the rhetoric of "every firm is doing AI" misses the underlying pattern. Australian legal AI in mid-2026 is a market with two giant off-the-shelf platforms at one end and roughly two sovereign builders at the other, with an emerging tier of specialist and boutique entrants in between. A client reviewing outside counsel will increasingly see the same AI vendor on multiple pitches without that vendor being a point of differentiation. The firms that win differentiation are the ones that have either built their own moat or set up the governance to claim it.
What to watch: whether any of the off-the-shelf buyers announce an in-house build (cost, sovereign client demands, or vendor price increases are the usual triggers), whether the ISO 42001 certification wave reaches mid-tier firms, and how the next edition of the AFR Law Partnership Survey breaks the builder count. Two is still a small enough number for it to double.