Most agent-governance work assumes a general enterprise setting: an LLM agent calling tools, a policy engine checking scopes, a log somewhere. A paper published this week takes a narrower and more concrete target: agentic AI running inside cloud-edge-radio infrastructure, the layered environment that modern telecom networks actually run on. The proposed Agent Governance Framework (AGF) starts from a specific problem. Agentic AI systems are increasingly deployed across distributed cloud–edge–radio infrastructures, where autonomous agents make decisions with direct operational and economic impact — provisioning spectrum, reconfiguring edge resources, triggering network changes that cost money or affect service. The paper's move is to enforce those decisions through smart contracts: SCs enforce business rules and policies, constraining autonomous actions according to predefined operational intent. That's a departure from how telecom infrastructure has historically been governed. Distributed radio, edge, and cloud infrastructures have traditionally been managed through rule-based, centrally controlled management and orchestration systems — a single authority setting policy top-down. AGF instead treats accountability as something that has to be built into the architecture as agent autonomy expands across heterogeneous networks, requiring transparency, verifiability, and compliance with SC-defined requirements rather than trust in a central operator. The framework's design requirements read like a checklist for what telecom operators will need before they let agents touch live network infrastructure: identity management and access control robust enough to uniquely identify and cryptographically authenticate every participating entity, including network providers, consumers, and the autonomous AI agents themselves acting between them. What makes this notable isn't the smart-contract mechanism itself — plenty of agent-governance proposals reach for blockchain-style enforcement. It's the domain. Telecom infrastructure is one of the few places where agentic AI failures have immediate, measurable, physical consequences: dropped calls, misallocated spectrum, radio interference. A governance framework built for that environment has to account for real-time constraints and multi-party accountability (network providers and consumers as distinct, cryptographically distinguishable stakeholders) in a way that generic enterprise agent frameworks don't need to. The paper frames its contribution as answering specific research gaps and questions about how to guarantee trustworthy autonomous decision-making across heterogeneous cloud, edge and radio infrastructures — a signal that the authors see this as infrastructure-specific work, not a repackaging of general agent-governance ideas for a telecom audience. Whether SC-based enforcement can actually keep pace with the latency demands of radio-layer decisions is the open question the paper doesn't fully resolve, but the framing itself — accountability as an architectural requirement for physical-layer autonomy — is worth tracking as telecom operators move from testing agentic AI to deploying it.