PPUVA
Search Join Us
← All Issues
Active Civic Technology Policy

Hampton Should Own Its AI Workflows

AI compute costs are collapsing. Cities that lock into vendor contracts today will be overpaying by an order of magnitude within two years — and paying a vendor to own institutional knowledge that should belong to Hampton.

Section 01: The Opportunity

AI is changing fast — faster than most government procurement cycles can track. Inference costs (what it costs to run an AI query) have dropped roughly 100x over the past two years alone. Within the next 12–18 months, that trajectory points toward another dramatic reduction.

That creates an unusual window: cities that act now can build AI-powered workflows at low cost, own the design, and reap savings as costs continue falling. Cities that sign multi-year SaaS contracts instead will be locked into yesterday's pricing on tomorrow's commodity.

What "AI Workflows" Means

A workflow is the structured process by which a city task gets done — how a planning inquiry gets answered, how a development application gets reviewed, how a council packet gets summarized for staff. AI can accelerate every one of these. The question is: who designs and owns that process?

What Vendors Are Selling

Companies like the Australian AI planning firms recently pitching US municipalities are selling workflow design plus a thin AI layer on top. As AI models become commodities, the ongoing value — and the vendor lock-in — comes from owning the workflow, not the model. Once a city depends on a vendor's process design, switching costs are high.

Section 02: The Cost Trajectory

The economics are not speculative. They are already underway:

  1. 2023

    GPT-4 Class AI: $60 per million tokens

    Running a sophisticated AI query cost organizations roughly $60 per million tokens — limiting practical use to high-value, low-volume tasks.

  2. 2025

    Same capability: under $1 per million tokens

    A 60x+ cost reduction in two years. Open-weight models (Llama, Mistral, Qwen) now match GPT-4 class performance at near-zero cost for self-hosted deployments.

  3. 2026–2027

    Projected: another 10–100x reduction

    Hardware efficiency gains, model compression, and competition among providers continue to compress costs. Cities signing 3–5 year vendor contracts now will be locked into 2025 economics in 2028.

Section 03: What Hampton Should Do Instead

PPUVA Recommendation

Build Once, Own Forever

Rather than purchasing a vendor's AI planning platform, Hampton should invest in designing its own workflows — with outside help if needed, but retaining full ownership of the process design, the data pipeline, and the output format.

The underlying AI model can be swapped as better, cheaper options emerge. The workflow — how Hampton handles a zoning inquiry, reviews a permit, summarizes a staff report — is institutional knowledge. That should belong to Hampton.

A practical path forward:

  1. 01.

    Identify one high-volume workflow — planning inquiries, permit status lookups, or council packet summarization — as a pilot. Start small, prove value, own the design.

  2. 02.

    Retain workflow ownership in any vendor engagement. If Hampton works with outside partners, contracts should specify that process designs, prompt libraries, and data pipelines are city property.

  3. 03.

    Avoid long-term per-seat or per-query SaaS commitments until the cost curve stabilizes — likely 2028 at the earliest. Short-term pilots with exit clauses protect the city's flexibility.

  4. 04.

    Designate an AI workflow coordinator — a city staff role or contracted position — to own this strategy. Every vendor the city talks to has someone in this role working for them. Hampton should have someone working for Hampton.

Section 04: The Risk of Inaction

The risk isn't that Hampton adopts AI too slowly. The risk is that Hampton adopts it on someone else's terms.

Cities that outsource their AI workflows today are making a bet that vendors will share cost savings as compute drops. History with enterprise software suggests otherwise. SaaS vendors price on value delivered, not on what the underlying infrastructure costs them.

Hampton has a narrow window to establish its own capability before vendor dependency sets in. The city's residents deserve public infrastructure — including AI infrastructure — that serves them, not a vendor's recurring revenue model.

Section 05: How You Can Help

This issue is moving fast. If Hampton is currently evaluating AI vendor contracts:

  • Contact your City Council member and ask whether any AI vendor contracts are under consideration in the FY27 budget process.
  • Attend City Council meetings and ask during public comment what the city's AI strategy is and who owns it.
  • Contact PPUVA if you work in city government and have insight into current vendor discussions.

This position brief was assembled by the Peninsula People's Union of Virginia. Published March 2026. We welcome corrections and additional sourcing.