Field notes

What we're learning, written down — and out loud.

Engineering judgment, governance ideas, plain-language explainers, and arguments about what technology owes the public. Part of how we keep the work transparent and the conversation honest.

Practice

A human who can actually see.

"A human reviews it" only means something if the human can see what they're approving — the reasoning, the close calls, not just the verdict.

August 2026 · 5 min read
Governance

The contract is the control.

Public-interest AI is won or lost in procurement. By the time the code is written, most of the decisions that matter have already been made.

July 2026 · 5 min read
Practice

Maintenance is the mission.

The riskiest moment for a public system isn't the build — it's year three, when attention moves on and no one is left to keep it true.

June 2026 · 5 min read
Perspective

Technology that works for the public: a builder's case.

Why software that runs public life should be held to the standard of public works — built to last, maintained, and governed in the open.

May 2026 · 5 min read
Approach

Beyond AI: the four things that actually decide whether a system serves people.

The deciding factor is almost never the model. It's people, processes, policy, and public understanding.

April 2026 · 6 min read
Governance

Build with, not about: what community governance really requires.

Real governance means authority, data ownership, rules enforced in code, and the right to say no.

April 2026 · 5 min read
Governance

Why staying out of a data regime is the strongest protection you can offer.

The strongest protection isn't a stronger lock — it's not collecting the thing at all. Scope is a safety feature.

March 2026 · 4 min read
Explainer

How to read an algorithm that's deciding your benefits — a plain-language guide.

You have the right to understand a system that decides your benefits. Five questions to start with.

March 2026 · 6 min read
Transparency

One truth for everyone: why the public should see exactly what officials see.

When officials see one version of the data and the public sees another, the gap is where trust dies.

February 2026 · 4 min read