Aug. 20, 2026
Every city is now building two at once: the physical one that planners have spent generations learning to shape, and a computational one that is quietly reshaping it from within. The impact of that digital infrastructure — increasingly powered by artificial intelligence (AI) systems and tools — has on life is much less visible, but it is equally important.
These ubiquitous systems can determine whose application gets reviewed fairly, whose neighborhood receives a bond improvement, and even who is more likely to be profiled for a crime. It’s commonplace that cities adopt standards for decision-making about the physical environment — our roads, water, and buildings — but how much attention is being paid to ensuring proper safeguards are in place for consequential AI?
Not enough. But that doesn’t have to be the case. Public agencies can use various levers to ensure accountability, explainability, and effectiveness in adopting AI systems and tools. Further, planners can help make sure that happens. The trick is to make sure those levers are pushed and pulled before anyone signs on the dotted line.
Recent years have seen a Cambrian explosion of AI tools and products that support planning decision-making behind the scenes. AI can passively collect and analyze data about roadway usage, automate zoning compliance review, and assist in design charrettes.
The myriad use cases of AI in planning promise new ways to comprehend the urban environment and introduce real efficiencies. They also come with important tradeoffs and risks, many of which are under-articulated by professionals in the field.
AI as invisible urban infrastructure
When I served as the lead of emerging technology in San Antonio, we routinely asked questions like, what if AI-generated bond recommendations inadvertently perpetuated inequitable biases? Or what if an automated permitting review system was prone to error 20 percent of the time? Our team helped set the city’s appetite for such risks, understand failure modes, and consider who is most likely to be affected by them.
Yet, I quickly discovered that standards for comparing vendor performance in these domains, or even to understand what an acceptable risk threshold is, are not readily available for municipal use of AI.
It’s commonplace that cities adopt standards for decision-making about the physical environment — our roads, water, and buildings — but how much attention is being paid to ensuring proper safeguards are in place for consequential AI? Not enough.
The Levels of Service (LOS) model offers a valuable parallel. LOS are contractual requirements, typically tied to physical performance of roadway assets. For example, when measuring the LOS for a road, you might consider a performance indicator such as the Pavement Condition Index.
Similarly, AI performance can be defined by measurable, use-case–specific, community-validated performance thresholds that define what acceptable deployment looks like before a contract is signed.
These performance thresholds are referred to among the AI community as benchmarks. Without such benchmarks for government use of AI, governments are left to “buy blind,” writes Jessica Tillipman, an associate dean for government procurement law studies at George Washington University.
Procurement as a lever of public oversight
Most American cities do not build AI systems — they buy them. The market for AI applications in government and public services is expected to grow 17 percent annually to more than $50 billion by 2030.
That makes public procurement a powerful tool for public oversight of emerging technologies because it is a vehicle for standards to become enforceable through contract terms, unlike policies or guidelines. Planners are already at the table where those contracts are shaped and can advocate for the performance standards and transparency needed to ensure public accountability.
Since procurement law was built around physical things, such as construction and commodity goods, it rarely surfaces or evaluates the AI tool usage that is embedded in professional services contracts. Using undisclosed AI in service delivery creates a “shadow AI” risk that planners need to watch out for, especially when the stakes are high.
AI performance can be defined by measurable, use-case–specific, community-validated performance thresholds that define what acceptable deployment looks like before a contract is signed.
Consider this scenario: A public works department hires a planning firm, which in turn partners with a third-party vendor to conduct a street quality assessment to inform bond-investment decisions. The vendor uses AI to assign a condition grade to every street in the metropolitan statistical area. When results reach the city council, a council member notices that the wealthiest neighborhoods are recommended for investment first. Why?
Could public works explain that outcome? Could the planning agency? Neither one built the model; a third-party vendor did. The public works department is one layer removed from the contract that the planning agency has with the vendor.
These kinds of challenges illustrate an opportunity to transform government contracts from an acquisition tool into a governance tool.
Who is at the table?
No single city has the market leverage to set these standards alone. Cities must pool their purchasing power to establish shared performance standards for AI procurement and establish market clarity that vendors need to build responsible products. The GovAI Coalition is one example of a community working to crowdsource AI policies, procurement workflows, and successful use cases.
Planners already understand cross-jurisdictional standard-setting. They also understand, sometimes all too well, how critical it is to develop appropriate standards for tools that affect the communities where they are deployed. Seattle recognized that when it established its Community Technology Advisory Board, which it consulted directly to help shape the city's AI plan and responsible AI policy, before any vendors were under contract.
Planning for the invisible
The planning profession is fundamentally about ensuring that the built environment supports thriving communities. This mandate has always led planners to expand their scope into new territories when the built environment demanded it, ranging from public health to highways to climate resilience. Digital infrastructure, including AI, is the next expansion.
However, to do this successfully, planners will need to consider that the bulk of AI for government use is either contracted directly or nestled deep within professional service contracts: two areas planners can influence with thoughtful advocacy of performance standards, benchmarks, service-level agreements, and inclusion of community voice in shaping goals for outcomes and transparency.
If we wouldn't contract for a bridge without a performance standard, we shouldn’t contract for AI without one, either. The safety and well-being of our communities may depend on it.
RECOMMENDED ARTICLES
