AI-Native Networks & the Orchestration Problem
Legacy OSS and BSS platforms were built to process transactions and execute predefined workflows. They were never designed to coordinate multi-agent systems operating across RAN, core, transport, and IT domains simultaneously.
As agentic AI moves from experiment to operational necessity, operators are colliding with an architectural wall. It is that existing systems cannot ingest real-time telemetry, coordinate agent reasoning, enforce policy in real time, or route decisions back to the network at the speed autonomy demands.
The result is a dangerous choice: bolt AI agents onto legacy systems and accept operational friction and vendor lock-in, or redesign the orchestration layer to make AI-native operations a first-class design principle.
The Insight
AI-native networks are not software upgrades. They are architectural transformations that shift control of the network decision-making layer.
In a legacy architecture, vendors provide the OSS, vendors provide the RAN, vendors provide the core. Each provides hooks for integration.
In an AI-native architecture, the orchestration layer becomes the control plane. Whoever owns the logic that coordinates agent reasoning, decomposes operator intent, enforces guardrails, and routes decisions to the network owns the network.
Why This Matters
Three things are accelerating this shift simultaneously.
First, 3GPP Release 21 (finalized June 10, 2026) locked AI-native design into the 6G baseline, meaning specifications written between now and 2028 assume reasoning agents at the edge and center of the network. Second, agentic operators are discovering that intent-based management (operators state what they want, AI decides how) requires a reasoning layer that legacy OSS cannot provide. Third, GSMA CAMARA API monetization is moving from concept to commercial reality, with 300 instances of 20 different APIs live in 65 markets. That means network capabilities are being exposed to third-party developers and agents. If operators do not control the orchestration layer, vendors will.
What Operators Should Do Differently
Here is what telecom operators should do now:
Stop treating AI integration as an OSS extension. Start architecting a separate agentic orchestration layer that sits between intent (what the operator wants) and execution (RAN, core, transport, IT decisions).
Invest in agent fabric standards. Multi-agent systems require secure agent registries, standard interoperability protocols (Model Context Protocol, Agent-to-Agent frameworks), and TM Forum aligned governance. This is not a feature. This is the foundation.
Design for telemetry-to-decision velocity. AI-native networks run a continuous sense-learn-act loop. Legacy OSS logs data for dashboards. Agentic systems feed data streams into reasoning loops with sub-second latencies for RAN decisions and longer cycles for orchestration and assurance.
Map the control question: who decides how agents coordinate when there is contention? (e.g., cost optimization vs. QoS guarantees, energy efficiency vs. coverage). The operator must own this policy layer, not the vendor.
TRY THIS TOOL
Conduct an "agentic readiness audit" of your OSS and BSS stack against these three dimensions:
Real-time telemetry ingestion. Can your systems ingest and process network data at sub-second intervals across RAN, core, transport, and service domains, or are you still batch-processing logs into dashboards?
Agent discovery and orchestration. Do your systems have a secure agent registry and standard protocols for multi-vendor AI agents to register, discover capabilities, and coordinate reasoning? Or does each vendor integration require custom plumbing?
Policy enforcement and guardrails. Can your systems validate agent reasoning in real time and enforce operator intent (cost boundaries, SLA guardrails, security rules) before decisions reach the network? Or do agents operate without guardrails until a human catches the error?
If the answer to any of these is "we have custom scripts and integrations," you are not agentic-ready. That is not a judgment. It is a starting point. The question is whether you modernize now or inherit architectural debt during the 6G standardization freeze (June 2028).
STRATEGIC SIGNAL
The GSMA Open Gateway reported that 86 operator groups, representing over 300 networks and 80 percent of global mobile connections, have aligned around a common CAMARA API framework. Over 300 API instances are now commercially live across 20 different capabilities (location verification, quality on demand, SIM swap detection, device identification) in 65 markets.
This is not an innovation initiative anymore. It is a commercialization inflection point.
The strategic implication: network capabilities are becoming programmable, exposable, and monetizable. Agents (both in-house and third-party) will increasingly consume these APIs. If operators do not govern how agents discover, authenticate against, and consume these APIs, developers and vendors will fill that gap.
The next 18 months are the decision window for embedding CAMARA governance into the agentic orchestration layer. After that, migration becomes exponentially harder.
Brian C. Newman is a telecom and AI strategy consultant, course creator, and former Verizon technology leader with more than 30 years of experience across wireless networks, 5G, network operations, infrastructure modernization, and emerging technologies. He helps organizations understand how AI, connectivity, edge computing, and digital infrastructure are reshaping business operations, real estate, public safety, and customer experience.
