Weekly signal

Between June 22 and June 30, 2026 the dialogue around agentic AI in infrastructure and city planning shifted from proof‑of‑concept examples to concrete infrastructure enablers and governance patterns. Four developments stood out: Forrester’s operational framing for agentic modernization, TM Forum’s AI‑native ODA work at DTW‑Ignite, NVIDIA’s Omniverse runtime updates, and a governance‑first product pattern from HelloTwin (their June 24 event). Together these items show the pieces cities must align to safely and effectively deploy agents: modern APIs, AI‑native runtime/registry patterns, edge/connectivity policy, and a decision authority that ties agents to auditable outcomes.

What changed

Forrester (Jun 22, 2026) — modernization is not optional. Forrester published a focused advisory that reframes agentic deployments as an architecture problem: brittle, undocumented interfaces and siloed data are the primary reasons agent pilots stall. The report calls out an "inside‑out" modernization path — identifying which system boundaries must be hardened (APIs, identity, telemetry) so agents can be reliable and auditable when they interact with critical urban services (permitting, asset maintenance, traffic control). For planners and CIOs this turns modernization from a long‑term backlog item into the gating factor for agentic automation.

TM Forum / DTW‑Ignite (June 23–25, 2026) — an AI‑native control plane for connectivity and operational domains. At DTW‑Ignite the community introduced AI‑native extensions to the Open Digital Architecture (ODA): a governed execution layer for agent workflows, MODaaS patterns to manage model selection and cost, and security/trust agents that enforce policy across hybrid networks. For city planners and utilities this matters because smart‑city autonomy depends on predictable, low‑latency connectivity and enforceable cross‑domain policies (network slices, edge compute placement, SLA‑bound actuation). The TM Forum artifacts and demos provide a reusable blueprint for how carriers, cloud providers, and cities can align on an execution model that supports agentic loops at scale.

NVIDIA Omniverse updates (release notes last updated Jun 26, 2026) — cloud and runtime hardening that reduces integration friction. NVIDIA’s June updates to Omniverse on DGX Cloud and Kit releases include storage APIs, improved USD/CAD conversion tooling, and streaming/renderer optimizations. Those changes matter operationally: higher‑fidelity, cloud‑streamable digital twins with robust asset pipelines are now easier to host in enterprise or hybrid cloud environments, lowering the engineering barrier to spin up agent‑executable simulation loops that feed planning, permitting simulation, and real‑time operations. In short: fewer bespoke pipelines and more off‑the‑shelf runtime for city digital twins.

HelloTwin (June 24, 2026 event) — governance as product: the “Digital Authority” pattern. HelloTwin’s public materials and June 24 live event presented a semantic digital twin that acts as a single, governed decision owner — it defines metrics, resolves semantics across apps, and delegates to agents within tightly scoped sandboxes while logging decisions for audit. For municipal use this offers a concrete pattern: don’t hand agents unfettered execution — instead bind them to a digital authority that owns the decision, scope, and audit trail. This governance‑first approach aligns with both legal/regulatory needs and operational safety for city actions.

Why this matters for infrastructure & city planning

Cities and infrastructure systems are high‑stakes environments: mistakes cause safety, service continuity, and equity harms. Agentic AI promises faster plan synthesis, automated asset triage, predictive maintenance scheduling, and real‑time traffic/energy coordination. But those capabilities require more than good models — they require:

• hardened interface and telemetry stacks so agents see reliable, auditable inputs (Forrester’s point),
• an AI‑native execution and trust plane that maps models and agents to network/edge constraints (TM Forum),
• digital‑twin runtimes that handle real world fidelity at cloud scale (NVIDIA), and
• governance constructs that own the decision and prevent runaway agent behavior (HelloTwin).

Practical next steps — a tactical playbook (for CTOs, city planners, digital‑twin teams)

1. Rapid inventory & gap analysis (week 0–4):

   • Catalog the systems agents would call (permits, asset register, SCADA, GIS, mobility APIs). Tag whether each has: stable API, identity/auth, telemetry stream, and SLAs. Use Forrester’s modernization framing to prioritize the top 3 integrations that must be hardened first.

2. Map execution & connectivity requirements (week 2–6):

   • For each pilot, define latency, bandwidth, and edge compute needs. Align with carrier or municipal network partners and evaluate TM Forum ODA patterns — can you request a network slice or guaranteed edge execution window for an actuation workflow? Start small (one traffic corridor or one set of critical assets).

3. Prototype a twin + agent dry‑run (month 1–3):

   • Use Omniverse (or an equivalent USD/point‑cloud runtime) to create a simulation loop: ingest the asset geometry, attach simplified sensor telemetry, run an agent to generate a plan, and record the decision trace. Focus on deterministic I/O (storage API, USD pipelines) so the prototype is reproducible between cloud and on‑prem.

4. Build a governance scaffold (month 1–4):

   • Implement a “Digital Authority” in the pilot domain: one owner for the decision, scoped agent permissions, preconditions for actuation, and mandatory human confirmation for high‑impact decisions. Log inputs, reasoning, and outcomes for audits. Integrate policy agents for continuous compliance.

5. Measure & iterate (month 3+):

   • Treat each pilot as an operational experiment: measure time‑to‑decision, error rate, human override rate, and cost per decision. Use those metrics to justify further modernization sprints (APIs, telemetry fidelity, identity hardening).

Risks and guardrails

• Agentic sprawl: without an authority/registry, agents proliferate and conflict. Use a registry and governance policy agent (TM Forum patterns) to control agent surface area.
• Data fidelity: simulation drift will produce unsafe actuation — prioritize deterministic pipelines and test in simulation before any real actuation.
• Legal/compliance: municipal mandates will demand auditable traces; design the Digital Authority to record decisions and evidence with retention policies.

Bottom line

This week’s signals show the hard work required to move agentic AI from specialty pilots to safe, production infrastructure in cities. The technical and governance blueprints you need are now converging: plan modernization sprints, adopt AI‑native execution patterns where connectivity matters, prototype twin+agent loops with hardened runtimes, and require an auditable decision authority before any live actuation. If you’re building or funding city pilots, prioritize those four levers and treat them as the project’s critical path.

Sources

1. Forrester — "The Inside‑Out Approach To Agentic Software Modernization" (published Jun 22, 2026).
2. TM Forum — DTW‑Ignite 2026 sessions and AI‑Native ODA materials (DTW Ignite, June 23–25, 2026).
3. NVIDIA — Omniverse on DGX Cloud / Omniverse release notes (last updated Jun 26, 2026).
4. HelloTwin — "Digital Authority" product materials and June 24, 2026 event pages.