The Most Important Use Case for Verifiable Cities: AI Agents That Citizens Can Actually Audit

A deep dive into the Ethereum Foundation's call for implementations—and the one proposal that could transform how cities work

The Ethereum Foundation's Use Case Lab just issued an intriguing call for implementations: "Verifiable Cities." The brief asks where coordination and verifiability constraints are most acute in urban systems, and where public blockchains can meaningfully expand civic capacity.

After spending considerable time with the 20+ sources in their reading list—from Vitalik's essays on crypto cities to the comprehensive "Agentic State" paper on AI-transformed government—I believe there's one use case that stands above the rest. Not because it's the flashiest, but because it directly addresses the core problems cities face and builds on technology that's already deployed.

That use case: Verifiable Automated Policy Execution—AI agents executing rules-as-code on blockchain infrastructure.

The Problem No One Is Solving

Here's a scenario that plays out millions of times daily across every city: A citizen interacts with a government system that makes a decision affecting their life. A permit is denied. Benefits are calculated. A compliance violation is flagged.

The citizen has no idea how the decision was made.

Even when humans are nominally "in the loop," the reasoning is often opaque. Which rules applied? What data was used? Was the process consistent with how similar cases were handled? These questions are almost impossible for citizens to answer, and increasingly difficult even for the government employees processing the requests.

Now multiply this problem by the rise of AI. Ukraine's Diia.AI already handles government services through conversational AI. Abu Dhabi's TAMM 3.0 personalizes access to 800+ services. The State of Goiás in Brazil reduced project analysis time from one year to one week using AI agents.

This is happening. AI agents will handle an increasing share of government interactions. The question is whether citizens will be able to verify what these agents are doing.

The Convergence Thesis

Two technological developments are converging to make verification possible:

AI agents are becoming capable of real government work. Not just chatbots that answer FAQs, but systems that can "pursue outcomes, adapt through feedback, and coordinate across organizational boundaries," as the Agentic State paper puts it. The distinction matters: traditional automation acts like a "diligent clerk" following rigid scripts. Agentic AI acts like a "seasoned caseworker" navigating messy reality—handwritten notes, conflicting guidelines, judgment calls.

Blockchain infrastructure has matured for civic applications. California's DMV tokenized 42 million car titles on Avalanche, reducing transfer time from two weeks to minutes. Bhutan runs its national identity on Ethereum. Buenos Aires serves 2.5 million citizens with zkSync-based credentials. Quincy, Massachusetts issued blockchain-based municipal bonds with same-day settlement.

Neither development alone solves the verification problem. AI agents need something to verify their actions against. Blockchain records need something meaningful to record.

The Missing Piece: Rules as Code

The bridge between AI decision-making and citizen verification is rules-as-code—the practice of encoding legal rules as executable logic alongside their natural language versions.

MIT's Computational Law Report demonstrates this with the Blawx platform. When you try to express legislation as code, something interesting happens: ambiguities that humans navigate through "judgment" become blocking errors. The encoding process reveals unresolved conflicts in the original rules.

This is valuable in itself—clearer laws benefit everyone. But for our purposes, the key insight is: machine-interpretable rules are blockchain-verifiable rules.

When a policy exists as executable code, every decision can be checked: Did the system apply the logic correctly? Was the input data accurate? Which version of the rules was in effect? These become answerable questions, not matters of trust.

How Verifiable Automated Policy Execution Works

Here's the system in practice:

Citizen submits a request through any interface—mobile app, web portal, voice assistant, in-person kiosk. The AI agent normalizes the request regardless of entry point.
The agent gathers required information. It identifies what data is needed, retrieves what's available from government databases (with appropriate permissions), and prompts the citizen for anything missing.
The agent executes rules-as-code. The policy logic is encoded in a declarative format that maps directly to legislation. The agent runs the citizen's data against this logic to produce a decision.
Blockchain records the decision. The result, the rule version applied, and a hash of the input data are recorded on-chain. Zero-knowledge proofs enable verification without exposing sensitive citizen data.
Citizen receives the result with an audit trail. The response includes not just the decision but the reasoning chain: which rules applied, what evidence satisfied each condition, and how to appeal.
The system learns. Aggregate outcomes feed back into policy evaluation. When rules consistently produce appeals or unintended consequences, policymakers have real data for refinement.

Why This Is the Highest-Impact Use Case

Among all the possibilities—voting systems, municipal bonds, identity credentials, tokenized assets—why focus here?

It addresses coordination constraints. Cross-agency government processes currently require manual handoffs. When a building permit touches planning, zoning, environmental, and fire safety departments, each transition introduces delay and error. AI agents coordinating through shared blockchain state can orchestrate these workflows without the overhead.

It addresses verifiability constraints. This is the Ethereum Foundation's explicit focus. Citizens cannot currently audit automated decisions. Rules-as-code on blockchain changes that fundamentally.

It builds on proven components. Identity infrastructure is deployed (Bhutan, Buenos Aires, Estonia). Rules-as-code is demonstrated (MIT, New Zealand). Agentic AI is in production (Ukraine, Abu Dhabi). This proposal integrates existing elements rather than requiring new breakthroughs.

It enables incremental adoption. Cities don't need wholesale transformation. Start with building permits. Encode existing policy. Deploy agents for routine requests. Expand based on outcomes. The path is clear even if the destination is ambitious.

The Identity Foundation

A crucial point: this system requires identity infrastructure for both citizens AND AI agents.

The a16z crypto team identifies a striking statistic: "non-human identities now outnumber human employees 96-to-1" in organizations. These AI agents cannot transact autonomously because they lack verifiable credentials.

The solution is "Know Your Agent" (KYA)—blockchain-based identity that links agents to their principals, their constraints, and their liability. For government AI:

Principal linkage: Which department does this agent represent?
Constraint specification: What actions is it authorized to take?
Liability assignment: Who is responsible for its mistakes?

When a citizen's permit application is processed by an AI agent, they should be able to verify: Is this agent actually authorized by the planning department? Is it operating within its defined scope? What's its track record on similar cases?

Blockchain provides this infrastructure. The agent's identity, authority, and history become verifiable public records.

The Privacy Question

A reasonable objection: "Public audit trails expose citizen data."

The answer is zero-knowledge proofs. A ZK proof can demonstrate "this citizen met the eligibility criteria" without revealing specific income, health records, or other sensitive data. The decision is verifiable; the underlying data remains private.

Buenos Aires's QuarkID already operates on this principle—enabling citizens to prove eligibility without revealing more than necessary. The privacy-preserving verification layer exists.

What Cities Should Do

For municipalities considering this approach:

Start with identity. Every implementation in the research starts here. Invest in citizen credential infrastructure that supports selective disclosure and portability.

Pilot narrowly. Pick a single domain—building permits, business licenses, simple benefit calculations. Encode existing policy as rules-as-code. Deploy AI agents for routine requests only. Keep human oversight throughout.

Measure ruthlessly. Processing time. Citizen satisfaction. Audit completion rate. Appeal rate. Outcomes vs. intentions. If the system isn't improving these metrics, adjust.

Expand based on demonstrated trust. Citizen acceptance of automation correlates with transparency. As people see that they can verify decisions—and that verification reveals consistent, fair treatment—acceptance grows.

The First Mover Opportunity

The Agentic State paper makes a compelling case for urgency:

"The choice is not whether agentic AI will transform government, but whether governments will shape that transformation proactively or adapt to changes designed by others."

Jurisdictions that build this infrastructure first attract talent, investment, and citizen satisfaction. Estonia's five-minute tax filing didn't happen by accident—it resulted from decades of deliberate infrastructure investment that now pays compound dividends.

The Ethereum Foundation is offering technical support, exploratory funding, and ecosystem connections for implementations. The technology exists. The evidence supports feasibility. The question is which cities will act first.

The Ethereum Foundation's Verifiable Cities call for implementations is open now. If you're building at the intersection of AI, blockchain, and civic technology, this is the moment.

Sources: