The lawyer who builds: governing AI from the inside

Last month, a bar association sat me on a panel between a futurist and an ethicist and called me the technologist. Um, who, me? Lawyers advise on technology, they don't claim to be it. But I guess it's a realization that the word fits, and that it fits because of a specific change in how I work. Somewhere in the last two years I stopped looking from the outside about systems and started building the systems I was being asked to govern.

The conception I want to describe is one I've started calling, only half-lightly, the lawyer who builds. Not a coder, not a legal engineer. A lawyer whose raw material is judgment, who builds enough to make that judgment real, and who can help shape governance because they understand it from the inside.

I don’t mean applications, or solutions for a lawyer being a lawyer, not really. It mapping a portion of legal judgement into a system.

Privacy taught me to design, not advise

The seed was planted back in my mind when an "agent" was something from a Bugs Bunny cartoon. I spent years as a privacy lawyer, and privacy is where I learned that the interesting work happens upstream of the advice. At Atlassian I helped shape a global program. At Box I sat with the engineers moving the company's infrastructure to the cloud, where the architecture decisions and the legal commitments turned out to be the same decisions in different clothes. At Grindr we were responsible for the data of ten million people across a hundred and fifty countries, where a privacy failure wasn't a fine — it was a danger to someone's life.

Privacy by design taught me something: you cannot protect data you cannot see moving. A policy that promises protection without a mechanism that produces it is a sentence, not a safeguard. When I moved into AI work, that instinct was even more tangible. The question isn't whether information should be protected. It's how we'll know it was protected, when the thing handling it is a non-deterministic system passing information to other systems that make their own decisions. Protection you cannot verify is a hope.

What NYU made me prove

The conception got precise when I had to write it down for people who wouldn't let me be vague. This spring I wrote a discussion draft for a workshop on fiduciary duties and AI, convened by the GliaNet Alliance and the NYU Information Law Institute. The paper is called "The Fiduciary's Footprint: How Legal AI Agents Can Prove They Were Loyal."

We have centuries of law on what it means for a fiduciary to be loyal, careful, honest, and discreet. None of it was written for a software agent that takes an action at three in the morning with no human in the room. Writing for that audience, I couldn't hand-wave. I had to take four abstract duties and turn each into something an agent could produce as evidence. Loyalty became authority binding — proof the agent acted only within its mandate. Care became consensus verification — proof the decision was checked rather than guessed. Disclosure became a human review record. Confidentiality became hash-only enrollment — proof the system could show its work without exposing the secret. Once you specify a duty tightly enough that a machine could satisfy it and produce a record, you're no longer describing the rule. You're authoring it.

Why I had to build it to believe it

The argument wasn't real to me until it existed as a working thing. I took the four-layer idea and built it, a prototype I called a certificate of action, a tamper-evident record of what an agent did, the policy it ran under, the evidence it relied on, and the consensus that approved it. I wrapped it in an interface so a non-engineer could look at a queue of agent decisions and, for each one, see the reasoning chain, the tools it touched, and where a human stepped in.
I needed to build it because building is a form of testing that writing cannot do. Written down, "the agent proves it was loyal" is more tangible. Prototyped, within an hour you have to decide what counts as proof, who gets to see it, and what happens when two agents disagree.

I came out with a small thesis: the spec is the governance. The document that tells the agent what it may do is no longer a description of the control. It is the control. This is the precise sense in which I had become a technologist — not because I write production code, but because I build enough to understand, to prototype, to pressure-test an argument against a running system before I put my name on it. My material is still judgment. I've stopped pretending judgment can be exercised at a safe distance from the machine.

What a lawyer-technologist actually is

The word "technologist" is older and wider than "engineer." In most fields, a technologist applies technology inside a domain rather than building it as a profession. Law has been slow to claim this figure, and where it has tried, the role collapses into 2 ideas: the sophisticated buyer who procures tools and calls it a technical identity, or the legal engineer who builds workflows but leaves the duty at the door. What I mean sits between them, and it belongs to a lawyer specifically.
The scarce input in an agentic system isn't the code; capable people can write that. The scarce input is the judgment about what the system is allowed to do, what counts as it having done the right thing, and who owns the consequence when it doesn't. That judgment is native to a lawyer. A lawyer-technologist is the person who can hold that judgment, express it in a form the system will obey, and read the record the system returns.

The last few weeks have reinforced, models are temporary. The durable layer is the other one — reasoning, provenance, the ability to make a duty verifiable, the willingness to own the output the moment it lands in front of a court or a customer. That layer survives the churn.

On that panel I closed by saying the role for lawyers now is to be a curious steward — to shape how this technology lands on our profession instead of waiting to be told. I'd add one thing. You don't get to govern from the inside if you were never willing to go inside. The lawyers who build are the ones who will still be holding the duty when the systems are making the decisions.