Google's legal AI play is a connector list, and Harvey is on it

The mechanism worth reading twice is the audit log. According to TNW, what has held agentic tools out of large firms is supervision rather than accuracy: a practice cannot deploy something it is unable to oversee when privilege and client confidentiality are at stake, and centralised IT controls with logging are aimed at that objection ^[9]. So the thing on offer is custody of the record of who asked what, of which document, under whose permissions.

The structure says more than the naming does. Of the four components in the product, one is the legal skills and the other three are connectors, a partner ecosystem and governance ^[10]. Five connector targets are named, and Legora is wiring in directly, which puts six third-party legal systems within reach of the same console, two of them rival assistants ^[11].

TNW's reading is that firms have already bought specialist tools, will not rip them out, and that a hyperscaler's realistic move is therefore to sit above the stack ^[18]. Legora reaching $100mn in 18 months is why displacement was never the plan ^[2]. The direct integration, where a response links back to the relevant Legora thread with citations, source grounding and permissions intact ^[6], is what annexation looks like when everyone involved is being polite about it.

Accenture, Deloitte and KPMG in the partner ecosystem ^[7] carry the other half of it. Nobody at a magic circle practice deploys an agentic platform without a systems integrator, and consultancies bring the change management the software does not ^[12].

The skill likeliest to pay for itself is DSAR fulfilment. Responding under GDPR is expensive, deadline-bound and largely mechanical, which per TNW makes it the clearest case where automation returns money instead of demonstrating a capability ^[16]. Citation verification appearing in the same list is a concession of a different sort, since fabricated case law has already reached real filings ^[13].

Against that, TNW argues the deciding factor is procurement: a firm already on Workspace and Google Cloud has a path a standalone vendor has to argue for, and in enterprise software that tends to beat product quality ^[14]. Anthropic has put Claude into Microsoft Word with contract review leading its use cases, OpenAI is selling into the same accounts, and Lexroom is building for civil-law jurisdictions that US-centric products handle poorly ^[17]. All of them are pitching a buyer whose partner compensation does not reward efficiency and whose billable hour sits awkwardly with faster associates ^[15].