The $2.1 billion stampede into legal-tech startups this year tells a partial truth dressed up as progress. Yes, AI is making routine work — contracts, incorporations, document review — faster and cheaper. But the Forbes story glosses over the hard question UK regulated firms are now facing: which products are built for firms that answer to the SRA, FCA or PRA, and which are built for Silicon Valley's version of law? Most venture capital in this space has flowed to companies optimising for speed and unit economics, not for the governance, audit trail and liability controls that a mid-market UK practice actually needs. That gap is not shrinking. It is widening.

This is part of a larger pattern we are watching closely. The legal AI market is bifurcating. On one side sit the consumer-facing platforms — Harvey, Legora, and the cottage industry of GPT wrappers — which treat legal work as a commodity to be automated down to zero human touch. On the other side sit the enterprise systems — Luminance, LexisNexis, Thomson Reuters — which assume you have a budget measured in seven figures and a compliance team to match. Neither camp is built for the reality of a 50-person law firm, a mid-sized insurance underwriting function, or an accountancy practice that needs AI to work inside existing workflows without creating new regulatory debt. The $2 billion is mostly flowing to the first camp. The second camp is profitable but conservative. There is almost no one building for the third.

Here is what we believe actually matters: AI in regulated firms must be explainable, auditable, and integrated with real human responsibility. A contract-generation tool that works 95 per cent of the time is worthless to a solicitor if she cannot explain to the SRA what the AI did, why it failed in that 5 per cent, or who is liable when it does. Most of the platforms attracting venture capital have not solved this. They have optimised for throughput. The ones that do solve it — Luminance deserves credit here — tend to be vertical and expensive. We have built Trovix Aria and Trovix Sift differently: as knowledge assistants and data extraction tools that surface their reasoning, leave the human lawyer in control, and integrate with your existing practice management and case management systems. That means slower adoption, lower growth curves, and less venture capital. It also means firms actually use them without getting sued.

If you run a mid-market legal, insurance, accounting or financial services firm, the practical advice is blunt: do not buy hype-stage products just because $2 billion landed on legal AI this year. Instead, ask three questions. First: can I explain to my regulator exactly what this AI does and why? Second: does it integrate with my actual workflow, or does it require me to rebuild it? Third: if it gets something wrong, who is liable and can I prove it? Products built for scale and speed fail all three tests. You need something built for control and governance. That is a smaller market, which is why it attracts less capital, which is why most UK practices are not yet seeing genuine options. The democratisation of legal AI is real. But most of what is being democratised is risk wrapped in convenience.