The Law.com survey of 121 legal leaders confirms what we already knew: general counsels with access to advanced AI models are pulling work back in-house and using it to push back on law firm fees. That matters acutely to mid-market UK law firms, accountancy practices, and insurance counsel operations. But the story misses the critical distinction: in-house lawyers using Harvey or Claude or Microsoft Copilot to draft motions or review contracts are solving a different problem to the one most mid-market firms actually face. They are automating high-volume, low-complexity work. The real threat to traditional pricing isn't that AI is replicating complex judgment—it is that clients now have proof they can do the routine stuff themselves, and they will no longer pay law firms £300 per hour for it.

This is the inevitable endgame of the 'big model' approach that has dominated legal AI for three years. Products like Legora and Luminance built their value on giving lawyers and in-house counsel access to frontier language models, reasoning engines and retrieval systems. They work. The problem is that they work equally well for the client as for the service provider. Once you hand a client a sufficiently capable tool, you have given them a business case to do more of their own work. The firms that will survive this shift are not those trying to compete on access to the same AI models as their clients. They are the ones who build proprietary workflows, client-specific knowledge bases and governance frameworks that turn AI into something bespoke, defensible and locked into their practice. That requires a fundamentally different architecture than fine-tuning a large language model.

Trovix's approach—and this is where we differ sharply from the market consensus—is to build AI systems that are firm-specific, client-specific, and designed for regulatory constraint from the ground up. When you implement Trovix Sift for document intelligence or Trovix Aria for fee-earner knowledge work, you are not getting better access to OpenAI or Anthropic. You are building a closed, auditable system that connects your firm's own data, your own precedents, your own risk appetite, and your own compliance obligations—whether that is SRA ABS rules, FCA Consumer Duty, ICO UK GDPR requirements or PRA operational resilience. The in-house counsel at the other end of the engagement cannot replicate that system with a ChatGPT subscription. They do not have your data. They do not have your institutional knowledge. And critically, they cannot prove to their board, their audit committee or their regulators that they have done it safely. That is the moat. That is what you should be building right now.

For any mid-market law firm, insurer, accountancy practice or financial services counsel still deciding on AI implementation: stop waiting for the 'best' general-purpose model to appear. It will not save you. Instead, audit which work actually leaves your firm—which matters most to margin, which is most at risk of client insourcing—and build proprietary AI workflows around that work. Use Trovix Audit to document and prove your governance to regulators and clients. Make your AI implementation so integrated into your client relationships that pulling it apart would damage the client's business, not strengthen it. This is not about having AI. It is about having AI that is harder to replicate than your competitors' and more trustworthy than your clients' homegrown alternatives. The firms panicking about the Law.com survey are the ones still thinking about AI as a cost-per-task problem. The ones building moats are thinking about it as a client lock-in problem. You should too.

Source: Law.com