1926 meets 2026

Last Tuesday Anthropic launched Claude for Legal — a legal-AI platform that plugs directly into the document management systems, research stacks and Microsoft 365 deployments of most large law firms. Twenty integrations, twelve practice-area plug-ins, an embedded agent that carries context across Word, Outlook, Excel and PowerPoint. Several global firms are now using it on live matters.

The same week, I posted a paper property information form to my solicitor in a stamped envelope. The form is the TA6, the standard seller's questionnaire in a UK house sale. I filled it in by hand. I signed it in ink. The Royal Mail will deliver it.

Both of those things are the real 2026. The gap between them is the most useful structural fact in the legal-services market right now, and it is not the cliché about old technology meeting new.

The vendors are not lying

The legal-AI category is doing the work it says it is doing. I want to be clear about that before anything else.

Productivity inside a top-tier law firm has moved faster in the last twelve months than in the previous decade. Drafting, research, due diligence, comparison, contract review, citation checking, summarisation — all real, all materially faster, all now plumbed directly into the document management system and the firm's preferred research platform. One global firm has confirmed roughly five hundred per cent usage growth across five thousand seven hundred lawyers in six weeks. That is not a marketing line. It is a press release from the firm itself, with the numbers attached.

I have friends who are lawyers. They are not over-impressed by easy things. They are quietly impressed by what they are now able to do in an afternoon. The leverage ratios inside the firms are changing. The capacity is real.

If you are inside the room — partner, associate, knowledge manager — you are already living in the future.

Two layers of legal work

Legal work, simplified almost to the point of caricature, sits in two layers.

The upper layer is professional judgement: research, drafting, comparison, redlining, due diligence, advice. It is high value, high variance, and almost entirely unregulated in form. Nothing in statute says a contract has to be drafted by hand, or that a litigation strategy has to be written in a particular template, or that a memo to a partner needs a witness. This is the layer the frontier models are eating. It is the right layer for them to eat. The work is textual, the inputs are dense, and the value created compounds with better tooling.

The lower layer is the regulated transaction itself. Conveyancing, deeds, wills, probate, marriage, divorce, certain notarial acts. Where the upper layer is governed by professional standards, the lower layer is governed by statute. The Law of Property Act 1925. The Wills Act 1837. The Electronic Communications Act 2000 and its long, slow patchwork of amendments. The TA6 form. The TR1 transfer deed. ID1 identity verification. HM Land Registry's Digital Identity Standard, which the regulator currently describes as optional with a "Safe Harbour" for early adopters.

That distinction is the whole game. AI productivity gains compound where the law allows them to. The base of the legal stack is statute. Statute is shaped by Parliament and the Law Commission, not by GPU supply or intellect.

What I am living through

My UK house sale this month runs entirely through the lower layer.

I am filling in the TA6 by hand because the buyer's side of the chain is not yet using the qualified electronic signature pilot. I am verifying my identity in person because the digital standard is genuinely optional. HM Land Registry now accepts qualified electronic signatures, which is real progress, but counterpart-signing across a chain of three or four solicitors remains painful enough in practice that the default in May 2026 is still wet ink on paper. The average time to exchange in the UK has just slipped past one hundred days. A hundred days. For a private transaction with a willing buyer and a willing seller, in a country with one of the most digitally connected populations in Europe.

I hunt for a Biro to sign the printed documents. Docusign? In this transaction, in this chain, it is a distant dream.

HM Land Registry's own published roadmap commits to "almost all simple registrations" being automated by 2035. That is nine years from now. On the current pace of frontier model releases — roughly one major version every six to nine months across three labs — that is approximately fifty model generations away.

My solicitor is faster than she was last year. I am exactly as slow as I was in 1996. The system underneath us both is as slow as 1926.

Where the productivity gain actually accrues

Once you draw the two-layer line, the commercial question becomes obvious. If AI productivity gains compound in the unregulated upper layer, and the regulated lower layer moves at the speed of statute, then the gain accrues to the middle layer that straddles them — the law firm, the conveyancing platform, the broker — and not to the consumer who is stuck on the floor below.

This is not a moral failure. It is what the structure produces. The firm's economics improve because the AI replaces hours that used to be billable to a junior. The firm keeps the spread. The buyer's experience improves only when statute moves. The buyer pays roughly the same conveyancing fee, waits roughly the same one hundred days, and posts roughly the same paper form.

The reform package currently in early consultation projects fee reductions "of over twenty per cent" if it lands. That is the size of the consumer-side prize on offer when statute eventually moves. Compare it with the upper-layer gain — five hundred per cent usage growth in six weeks, an order-of-magnitude shift in seat economics over the next two years — and the asymmetry tells you exactly where capital should be deployed and where it should not.

And in a different bit of the same 2026

A widening observation that is making the rounds in my own household this week.

My eighty-seven-year-old mother received a SIM backup modem from BT last weekend. The accompanying letter described the kit as providing her with "network resilience". She uses the internet to watch Strictly Come Dancing.

We can ship enterprise-grade failover infrastructure to the over-eighties for the sake of Saturday night light entertainment. We cannot ship the Land Registry a working digital deed before 2035.

This is not a complaint about BT. The modem is, in its own context, sensible — there are legitimate continuity reasons to dual-home critical telecoms to mobile fallback, and the over-eighties demographic has the highest household-internet dependency curve. The point is the contrast. The country that deploys mobile-network failover to retirees for ITV streaming is the same country whose statutory transaction layer cannot accept counterparty digital signatures inside a typical conveyancing chain.

The capability is not the problem. The capability is everywhere. The problem is the gap between where capability lands and where statute permits.

The steel-manned case for "this will normalise"

The honest counter-argument is that the regulated floor will catch up faster than I am suggesting, because the political constituency for digital transactions is now strong enough — buyers, sellers, lenders, fintech, the Treasury, even the regulator itself — that the Law Commission's pipeline will move. The Buying and Selling Homes consultation, the digital remortgage proof of concept, the planned removal of the Land Registry paywall, the new Code of Practice for estate agents: each is real and pointed in the right direction. Five years from now the consumer-side experience could look meaningfully better.

I take that case seriously. I think it is roughly correct on direction and significantly wrong on speed. Statute compounds in committee years, not in model years. The 2035 date HM Land Registry has put on its own roadmap is the regulator's honest number. Treat it as the floor of what is plausible, not the ceiling.

The second-order read

For anyone allocating capital into legal services — funds with stakes in conveyancing platforms, ALSP buyouts, document-automation businesses, RegTech — the right unit of analysis is not "how much AI does this business use". It is "which side of the statute line does this business sit on, and which side of the wire captures the productivity it generates".

A business that sits on the upper layer and bills a regulated counterparty is on the right side. A business that sits on the lower layer and has to wait for statute to move before its margin can expand is on the wrong side. Most legal-services investment theses do not draw this line yet. They will, because the gap is too large to keep ignoring.

The cheerful version of all of this is that the upper layer is genuinely getting better, and the people who work in it are genuinely getting more leverage. The honest version is that the gain is captured by the practitioners and not yet by the public — and that asymmetry is the business model of UK legal services for at least the next decade.

I am cheerful about the legal-AI category. I am also signing in ink this fortnight. Both can be true.

1926 meets 2026. All in a day's life in the UK.