The Lawyer as Builder

A hands-on seminar where students take a real statutory scheme and ship a working, open, public-interest deliverable for it — by directing AI agents the way a lawyer directs a junior associate — and verify it, in the open, against the real thing. The law is the curriculum. The build is the proof.

Why now

The law is good. It rots in the administrative layer — the forms, the logins, the instructions. A state child-support calculator that asks an ordinary parent for a bar number. A department of revenue that tells parents, in writing, to search Google. The doctrine is sound; the delivery is broken. And we know the gap is a choice, not a constraint, because one state closed it: Washington ships a free calculator that runs in your browser, needs no account, and prints the very worksheet you file. Students should learn to see that gap — and that it can be closed.

The course runs on two worked builds. First, free, open-source child-support calculators for seven states, built by directing AI agents — one read the statutes into a precise spec, one wrote the engine, one deployed, one checked every screen — then validated against the states' own tools, where I found two of my own errors and fixed them in the open. Second, the Red Binder Project: free, bilingual emergency-preparedness documents for immigrant families — powers of attorney, healthcare directives, guardianship nominations — assembled the same way. One computes a number; the other assembles the instruments a family needs. Same method, two shapes of the same locked box.

Format

Experiential, not lecture. Not a coding class — the rigor is legal. Students bring the judgment; the machine does the math, in public.

The spine — the legal rigor

• Statutory interpretation under pressure. Reducing a messy scheme to a precise, testable spec. The part where most of the work hides.
• The advice / information line. UPL, disclaimers, where a tool stops and a lawyer starts.
• Verification discipline. Check against ground truth — Washington as the benchmark for what good looks like. Be wrong in the open. Fix where everyone can see.
• Building for vulnerable users. Privacy and data minimization, plain language, bilingual access, the realities of a family preparing for a parent's detention.
• Open source, licensing, stewardship. Build to own, not rent. Who maintains a public good after the semester ends.
• Judgment vs. mechanics. We're hired for judgment, not math skills. Let the machine do the work, and show the rule behind every line.

Capstone — choose one build

• A calculator. A state's guideline tool that doesn't exist yet, or a better one than the state ships — open, auditable, citing the rule per line.
• A document-assembly tool. For a population that needs it: detention and emergency preparedness, expungement eligibility, eviction answers, simple-estate kits. The Red Binder model.
• A model legislative or administrative directive. Sometimes the highest-leverage build is the rule that makes the state ship the tool. Washington proves it's possible; a student can draft the mandate. (This is the full subject of the companion course, offered here as a first-class option for the student who wants to build at the policy layer.)

Each is a real, open public good — MIT-licensed where it's code — with the school's name on it.

What the school gets

• Experiential credits — the accreditation pressure point — from work students actually want to do.
• A current AI offering that's hard to staff otherwise, taught by a practitioner who shipped.
• Free public tools tied to the clinics already doing this work by hand.
• A clean press story: students publishing public goods under the school's name.

What I'd need

A faculty sponsor, a room, and a slot in the next cycle. I bring the curriculum, the toolchain, and two worked builds as templates — the seven-state calculators and the Red Binder Project. Glad to start with a single guest session to test the fit before anyone commits to a semester.