AI for Public Agencies. Bounded, auditable, human-reviewed.
Every agency has been pitched a chatbot. Very few have been offered something they can defend in a public meeting. We work on narrow, well-scoped AI use cases with source citations, audit logs and a human in the loop — starting with whether you should be doing it at all.
We grow your business with
The position
We do not lead with chatbots, and you should be suspicious of anyone who does
Public agencies have spent two years being pitched general-purpose assistants bolted onto their websites. The pitch is easy and the demo is impressive. The problem arrives later, in public: a system that answers a resident's question about eligibility, fees or a deadline, confidently and wrongly, with no citation, no log of what it said, and nobody who reviewed it.
An agency cannot defend that. Not to a council, not to a reporter, not to the resident who relied on it. Which is why the useful conversation is not "do you want AI" but "which specific, bounded task is worth automating, what does it cite, who reviews it, and what record does it leave?"
California has been building policy scaffolding around this since Executive Order N-12-23 directed state agencies to examine generative AI — its risks as well as its uses. Guidance and pilots exist at the state level. What most local agencies are missing is not permission; it is a way to scope a project small enough to succeed and documented enough to survive scrutiny. That is the same discipline we bring to municipal and special-district web work.
Where it works
Use cases narrow enough to be defensible
Each of these is bounded, grounded in documents you control, and improved rather than replaced by human review.
Knowledge-base assistants
Answering from your own published material and nothing else, with a link to the source page on every answer, and a clear handoff to a person when it does not know.
Records request triage
Classifying and routing incoming requests, identifying likely responsive record types, and drafting acknowledgements — with the determination itself left to staff.
Permitting and code enforcement support
Helping applicants find the right form and understand what a submittal needs, and helping staff summarize case history. Not deciding anything.
Case-management workflows
Housing and homelessness services where intake volume is high, notes are unstructured, and summarization saves hours — with human review before anything is acted on.
Document and meeting summarization
Draft summaries of long packets and recordings, clearly labelled as drafts, reviewed before publication.
AI readiness and governance advisory
Policy, disclosure language, procurement questions, vendor evaluation and a written record of decisions — often the most valuable engagement and sometimes the only one needed.
Non-negotiables
What we build into every engagement
These are not premium options. If a project cannot carry them, we will tell you it is not ready.
How we start
Readiness before implementation
Most agencies should buy the first two steps and then decide. We are comfortable with that outcome.
Problem framing
What is actually expensive — volume, wait times, staff hours, error rates? If nobody can name the cost, there is no project worth doing yet.
Readiness and governance review
Data, records retention, disclosure, procurement path, and who is accountable. Delivered as a document your counsel and your board can read.
One narrow pilot
A single use case, a measurable baseline, a fixed evaluation period and a defined stopping condition. Scoped so that failing is cheap and informative.
Evaluate honestly
Against the baseline, including what it got wrong. We will recommend stopping if the results say so.
Operationalize or stop
If it works, it gets logging, review workflow, documentation and training. If it does not, you have a defensible written record of why you did not proceed.
Questions
The questions agencies actually ask
Can you just build us a chatbot?
What about hallucinations?
Is any of this a public record?
Does an AI interface have to be accessible?
We have no AI policy. Where do we start?
Are you reselling somebody's platform?
Related
Related services and reading
ADA Title II Compliance Deadline: April 2027 and 2028
View page →BlogGEO for Government Websites
Read article →BlogGovernment Software UX Design
Read article →BlogPlain-Language Web Content for Government
Read article →BlogWebsite Content Governance With Multiple Editors
Read article →ServiceCalifornia City & County Website Design
View page →Project Managers who will work with you on your project!
Wondering what this costs? Every service has published pricing — no discovery call required to see it.
View pricingConsidering an AI project?
Tell us the problem rather than the technology. We will tell you honestly whether AI is the right instrument, and what it would take to run it defensibly.
Book a Strategy Call →