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

AI is real, useful, and easy to over-credit. We use it where it earns its place. But when a public system actually serves people — or quietly fails them — the deciding factor is almost never the model.

Four things decide it instead.

People

Who is in the room when the thing is designed, and who is on the receiving end when it ships. A system designed without the people most exposed to its harms will reliably find new ways to harm them.

Processes

The intake form, the queue, the appeal, the workflow. This is where most public technology succeeds or fails a person — long before any algorithm is involved.

Policy

The rules, consent regimes, and retention schedules that decide what a system is allowed to do. Good policy, enforced in code, is more protective than any model is clever.

Public understanding

Whether the people a system affects can see how it works, question it, and push back. A system no one can scrutinize cannot be held accountable.

Get those right and a plain database can be a public triumph. Get them wrong and the most advanced model is just a faster way to fail people.

This is why we describe our work as public-interest technology, not AI. The model is a tool in the kit. The discipline is everything around it.