Why we build specialized AI, not bigger models
One size fits none. Why a civic platform needs purpose-built architecture rather than a general model with a clever prompt.
There is a default answer in AI right now, and the default answer is: use a bigger model. Whatever the problem, a larger general-purpose system with a better prompt will eventually cover it.
For a great many problems that is true, and it is why the default exists. For the problems we work on, it is not.
Civic problems are local by definition
A general model knows what a pothole is. It does not know that this particular county splits road maintenance across two departments by pavement type, or that a specific municipality has a separate escalation path for anything within fifty feet of a school.
That knowledge is not obscure — it is simply local. It lives in a city's own processes and it is different in the next city over. No amount of scale produces it, because it was never in the training data and it changes when a department reorganises.
Auditable beats impressive
When a system routes a citizen's report to the wrong department, somebody has to be able to explain why. Not in the abstract — for that report, on that day.
That requirement rules out a great deal. A model whose reasoning cannot be inspected is difficult to deploy in public infrastructure, not because the technology is unsound but because the accountability is missing. Every specialized model we build is auditable and explainable by design, which is a constraint we accept before we start rather than a feature we add later.
Purpose-built has a cost, and we pay it
Building for a specific problem is slower than prompting a general one. It means understanding a domain before writing anything, and it means the result does not transfer to an unrelated field for free.
We think that trade is correct for civic infrastructure. The alternative is a system that demonstrates well and fails quietly in the specific conditions it was deployed into — which, for a resident waiting on a repair, is indistinguishable from not working at all.
One size fits none. We engineer tailored architectures for specific community and civic challenges.
What this means in practice
It means we say no to problems we do not understand yet. It means our models are smaller and narrower than the headlines. And it means that when a city asks how a decision was made, there is an answer.