How We Use AI
We use AI the way most offices now do: to research, to write code, to edit and to check our work. Here is exactly what it did, what people did, and who is responsible for what you read on this site.
What people did
- Decided what to ask. What a town would cost, what the roads need, what each ballot measure raises.
- Built the first models by hand. The town budget and the road costs started as Excel workbooks. The road workbook is published.
- Chose every assumption. The spending range, the reserve policy, the towns we compare against. AI did not pick any of them.
- Went to the primary source. Lyons and Louisville meeting packets, the Boulder County Assessor, the county engineer, contractor bids. We cite those documents, never an AI answer.
- Wrote what the site says. Content is dictated or drafted by a committee member. AI edits it.
What AI did
- Research. It helped us find sources. It was never the source.
- Code. AI wrote most of the code behind this website and the analysis models, to our specification. The budget’s risk simulation began as a basic model; AI helped rebuild it in Python so anyone can rerun it.
- Editing. It turns drafts into finished text, including some of our replies to residents’ questions.
- Auditing. We use it to look for our own mistakes. It has found real ones. The corrections are on How Our Numbers Have Changed.
- Consistency. Automated checks stop the website from building if a figure on any page disagrees with the model, or if a number we have retired reappears. People wrote the rule each check enforces.
The tools are Anthropic’s Claude, mostly, and OpenAI’s ChatGPT.
Who reviewed it
A person is responsible for every figure we publish. AI approves nothing.
- The models are published. Download the budget model, run one command, and it reproduces our published figures. Change an assumption and see what moves.
- Corrections are dated and public, including the ones that went against us.
- We held back what we could not review. We built an AI assistant that could answer questions about this site. We have not released it, because we could not yet check its answers to the standard we hold everything else to.
Noblestar and Cosmos
Nick Little, the committee’s president, founded Noblestar Technologies. Noblestar builds custom software. It does not build AI and has no AI models of its own.
- Our records of Niwot are a structured database, not AI. Parcels, roads, budgets, sources and figures, organised so that software — and AI tools — can read them accurately. Noblestar calls this kind of model a Cosmos. This website is built from it.
- The AI agents are in a different tool. Noblestar’s Command uses AI agents to build software. We use it to build and maintain this website. Nothing runs until a person schedules it, and a person approves every release.
- The committee’s internal database is used by Nick alone. Nothing in it is public and no AI acts on it unsupervised.
AI in a future Town of Niwot
That is not ours to decide. A town’s use of AI would be set by its charter commission and its elected council, in public meetings, where residents can weigh in.
Check us
If you want to see how a number was built, we will walk you through the model. Write to contact@niwot.town.