Any firm can be AI-native; every firm must remain professed.

I have spent eleven posts arguing about a line. This is the one where I say where it runs, and then admit what the whole exercise is worth.

The declaration has two halves and the first is a genuine invitation rather than a grudging concession. Any firm can be AI-native. Not only the venture funded new entrants designed around the machine from their first day of trading; the two partner high street practice, the legal aid firm running on fumes, the sole practitioner with a filing cabinet and a good reputation in one town. If what stands between an ordinary regulated practice and AI-native operation is a defensible way of supervising the machine, then that way of working can be written down, certified, taught and shared, and the crossing becomes available to everybody rather than to whoever raised the largest round. That is the ambition behind NativeLaw and it is the reason I would rather publish this argument than keep it.

The second half is the boundary. Every firm must remain professed. The machine may do everything up to the line, and everything is a word I mean generously: intake, assembly, drafting, checking, comparison, research triage, the entire body of work that consumed my early years and never deserved the rates it was billed at. Take all of it. Go faster than I am comfortable with. The line itself belongs to the professed, and it is the one thing we will not let the machine touch.

That boundary is not my invention, and this is the declaration where the mapping stops being a lawyer's argument and becomes the statute itself. The Legal Services Act 2007 provides that reserved legal activities may be carried on only by authorised persons. That is this declaration in long form: a statutory scheme whose entire architecture is a distinction between who may and who may not, drawn around authorisation rather than capability. The BSB does the same job through the Scope of Practice rules in Part 3 of its Handbook, which set out what each category of authorised person may do, and by implication what nobody else may. Neither instrument says anything about how good you are at the work. Both are about standing, because standing is the thing that can be withdrawn.

Now the admission, which I put at the end of the thesis too and would rather not bury. Almost nothing in these twelve declarations is original.

Philosophers argued the gap between lawful and just for centuries before I wrote Declaration VI. Legal scholars were writing about the indeterminacy of law long before anyone trained a model on a corpus of judgments. Plenty of people have observed that judgement becomes the bottleneck of the AI age, although nearly all of them treat the bottleneck as a defect to be engineered away rather than the point. Even the format is borrowed, and openly: Rome wrote the Twelve Tables because the plebeians refused to keep living under law that existed only in patrician memory, and commitments written down in public where they cannot be quietly rearranged are the oldest legal technology there is.

And the claim of non-novelty is checkable, which is why the codes are mapped provision by provision on the site. Set the declarations against the rules that already bind us and almost every one turns out to be an existing obligation restated. Accountability for work carried out through others. The duty not to mislead by act or omission. The properly arguable threshold. Confidentiality, and the privilege built on it. Compulsory insurance. The requirement to be able to justify your decisions. Both regulators have now addressed AI directly and neither has written new rules, because the old duties already reach the machine; the SRA said in its 2023 Risk Outlook that you remain responsible and accountable for the outputs from AI you are using, and the Bar Standards Board's guidance, in force from 18 May 2026, works by explaining how existing duties apply to new technology.

So this series asks for no rewiring of the profession. It asks for a refresher, read again at the moment the machine makes it urgent.

There is one thought in the series I would defend as more than a restatement, and it is the reason I keep coming back to imperfection. The law has always known it is flawed, and it says so out loud in its own machinery. Every right of appeal is the system conceding that its first answer may be wrong. Every dissenting judgment is the system preserving the argument it has just rejected, in case the future needs it. The law survives and evolves precisely because of that admission; the flaw is not a weakness in the mechanism, the flaw is the mechanism. Humanity is imperfect in the same way, because nature is, and imperfection is how nature moves. Nothing evolves that was already finished.

AI is a product of humanity and of nature. It is trained on our words, our judgments, our contradictions and our mistakes, built by imperfect makers out of imperfect material, and it inherits our imperfection at scale and at speed, wrapped in a fluency that hides it. There is no shame in that and no surprise. But it means the machine can never be the still point the system checks itself against. Only something that knows it is flawed, and has publicly promised to answer for it, can do that job. The law worked this out centuries ago and its answer was not a better instrument. Its answer was a person, who professed.

Which is why I am unmoved by the version of disruption that treats the roll, the vow, the duties outranking client and market, and the privilege shielding what a client says in confidence as so much friction. None of those arrived by accident and none survives on nostalgia; each was built and kept by people who had watched what happens without it. Strip them out and what remains may be fast, cheap and fluent, but it is not law. It is a product shaped like law, binding nobody, answering to nobody, with no one professed standing behind it when it goes wrong. Which is why I think the real revolution runs the other way. The prize is not law remade as another software market. It is AI built native to what law actually is; the machine given everything up to the line, and the vow kept on the other side of it. Harder to build, slower to sell, and the only version worth having.

People assume the two halves of my working life sit oddly together, a quarter of a century of immigration practice on one side and legal AI on the other. I think they only make sense together. The technologists can build the loop. Only practitioners know what the human in it has to do, because we are the ones who professed.

Every piece of work I have sent into the world since my admission has carried my name. That line of ink is not a signature in the ordinary sense; it is the vow, renewed one client at a time. Twelve declarations, and they all come down to that.