The Lessons of the Digital Parcel Map
In the 1980s and 90s, counties across the country took their first serious steps into digital GIS. Parcel maps that had lived comfortably in filing cabinets for decades were transformed into tidy digital layers. It was progress in the purest sense. Clerks could find things. Planners could analyze things. Assessors could sleep slightly better at night.
Most counties did not have the staff, hardware, or spare time to digitize their own land records. Vendors stepped in to fill that gap, and they provided real value. The capability gap was genuine. The gratitude was also genuine.
The dependency, less so.
In many cases, the vendor controlled the software environment where the new digital parcel layer lived. In some cases, they controlled the data itself. Recovering parcel data that had been structurally captured inside a proprietary system took years of effort, substantial expense, and occasionally a conversation with legal counsel that no one enjoyed. A few counties never fully extricated themselves. These are not dusty parables from a simpler time; some of those arrangements still echo through current budgets.
Most counties avoided the worst outcomes, though not because vendors experienced a sudden philosophical awakening about public stewardship. Technology moved slowly enough that the GIS community could recognize the pattern and respond. GIS managers compared notes at conferences. Quiet alarm bells became slightly louder alarm bells. Counties built internal capacity. They began to treat spatial data as public infrastructure rather than a feature of someone else’s application.
The Shapefile: A Clunky Tool for Sovereignty
And then there is Esri. Anyone in local government knows the relationship. Esri dominates the GIS software market. Licensing is convoluted and expensive. Upgrade cycles are determined elsewhere. Switching platforms requires stamina and a tolerance for disruption that few elected officials find endearing.
Yet GIS has not become a permanent system of data captivity. Why?
Because you can always export the shapefile.
The shapefile is not glamorous and far from perfect. It carries field length constraints. It does not handle topology elegantly. It arrives as a small parade of files that must remain together or risk existential crisis. But it is portable. You can move your parcel layer. You can open it in another system. You can migrate to PostGIS or QGIS if circumstances demand it. The data remains separable from the application.
That separation matters more than most procurement checklists acknowledge. Esri may dominate the software ecosystem, but it does not own the institutional representation of the land itself. Counties retain sovereignty over the substrate. The cage, if one insists on calling it that, includes a door.
The New Substrate: Institutional Memory
Now consider what is happening with AI in local government. When a vendor offers to ingest your ordinances, process your meeting minutes, and build an AI assistant that answers staff questions about policy, they are doing more than deploying a chatbot. They are constructing a machine-readable representation of your institution’s memory.
Documents are chunked. Embeddings are generated. Vector indexes are built. Retrieval pipelines are tuned so the system can reason over your ordinances, policies, workflows, and decisions. That processed layer becomes the substrate on which your AI capabilities depend.
In many current govtech AI stacks, that substrate does not travel. There is no shapefile equivalent.
Embeddings reside in proprietary vector stores. Indexing schemas are undocumented. Chunking logic exists as an internal implementation detail. If you leave the platform, you do not export the knowledge layer. You rebuild it. From scratch.
Reconstructing a knowledge layer is not a weekend project. It means reprocessing years of ordinances, staff reports, resolutions, tax data, GIS layers, personnel policies, and the quiet footnotes that explain why something is done the way it is. Not to mention fine-tuning and corrections over time by staff who know better. When the core structure is inseparable from the vendor’s application, the cost of switching providers shifts from a mere inconvenience to a potentially catastrophic threat to your budget and sanity.
From Software Dependency to Knowledge Dependency
GIS lock-in revolved around tools. AI lock-in increasingly revolves around how your institution’s memory is represented in machine-readable form. In GIS, even counties deeply invested in Esri licensing retain movable data. The ecosystem remains viable because the substrate remains portable, however inelegant that portability may be.
With AI, the substrate is often fused to the vendor’s system. The way your government “knows” what it knows in computational terms is being built inside someone else’s architecture, governed by someone else’s design decisions.
GIS matured over decades. AI procurement cycles move at the pace of pilot projects and budget amendments. Knowledge layers are being assembled now through chatbots and workflow automations, “quick wins” that feel harmless in isolation. Accumulated over time, they form structural dependency faster than most county governments can diagram it on a whiteboard.
Five Questions for the Knowledge Layer
The window to ask hard questions remains open. It will not remain so indefinitely. Before signing the next AI contract, the useful questions are rarely about interface features or per-seat pricing. They are more structural:
- Who owns the processed knowledge corpus?
- Can embeddings be exported in a usable, documented format?
- Is the chunking schema transparent and reproducible?
- Can another system re-index the data without reconstructing the entire pipeline?
- What precisely happens to the knowledge layer at contract termination?
If the answers are unclear, the county is not simply procuring a service. It is allowing a third party to define how the institution’s memory is computationally structured, without a guaranteed path outward.
Conclusion: Securing the Digital Commons
Local government has seen this movie before. The GIS community did not avoid the worst outcomes through luck. Practitioners recognized that parcel data represented public infrastructure. They added capacity and acted before dependency hardened into inevitability.
The same discipline is required now, only faster and with a clearer understanding of what is being built. The shapefile is not perfect. It merely exists.
If AI in local government is to mature into a healthy ecosystem where innovation and public sovereignty can coexist, the knowledge layer must be portable. Counties must own the structured representation of their institutional knowledge. Not necessarily the model. Not necessarily the orchestration engine. But the computational form of their policies, decisions, and memory should not evaporate at renewal time.
Without that separation, modernization takes on a different character. The rhetoric promises efficiency; the architecture quietly centralizes cognition behind a contract clause. We have encountered this dynamic before in quieter forms. The difference now is that the substrate in question is not just maps or parcel lines.
It is how your government thinks.