← All tech signals
Infrastructure

The Supervised Network

Verizon and Nokia say AI agents are being built into mobile network operations; in Nokia’s design, people sign off high-stakes changes. GSMA Intelligence puts most operators at partial automation.

What is it?

Verizon and Nokia say AI agents are being built to detect faults, isolate the cause and fix routine problems. GSMA Intelligence puts most operators at partial automation, so working agents are early. In Nokia’s design, people keep sign-off on high-stakes changes.

Why now?

Verizon says it is piloting AI agents on production infrastructure, and Nokia has two of its six agents running in its own environment. NGMN’s August report, summarised by ABI Research, asks what agents still lack to coordinate across multivendor networks.

Who’s building it?

Verizon, Nokia with Google Cloud, and the groups reporting on network autonomy: GSMA Intelligence and NGMN.

VerizonNokiaGoogle CloudNGMNGSMA Intelligence
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — suppliers packaging network agents, and operators piloting their own.
Buyers — mobile operators wanting agents to handle faults across the networks they run.
Gatekeepers — standards bodies, and engineers who sign off changes under Nokia’s design.
What’s at stake
Whether human sign-off on high-stakes changes stays a real decision as agents prepare more of the work behind each change.
Which supplier’s agents sit inside the operations software.
How agents from different vendors share context and act together.
What’s unsettled
How far Verizon’s agent pilot reaches across its production infrastructure.
When Level 4 arrives; most operators expect it around 2030.
Which of NGMN’s requirements, from agent identity to auditability, become standards shared across vendors.
02 · How the demand matureda progression, not a timeline
Novelty
In Verizon’s pilot, a master agent sends sub-agents to isolate a fault’s source.
Retention
Two of Nokia’s six agents already run inside its own environment.
Willingness to pay
Nokia and Google Cloud plan a six-agent software package on Google Cloud Marketplace.
Category arc
Nokia’s design has agents handle routine network work while people approve large changes.
03 · Where the market goesthe forks, not a prediction
→The supervised network: agents handle routine work alone while people approve high-stakes changes, as in Nokia’s glass box design.
→Autonomy deepens toward Level 4 in more domains, and sign-off covers a thinner band of changes.
→Agent standards stay spread across many bodies, and each supplier’s agents work best inside its own stack.
→Most operators stay at partial automation, and agents remain assistants to engineers.
How far agents run alone is set by two tests: whether operators can check what agents did, and whether agents from different suppliers can share context.
04 · The Agentic Delegation deskadjacent markets, moving now
The Co-Piloted Airspace
Software plans, staff accept or pass.
FAA’s new scheduling engine has every recommendation reviewed by staff.
FAA Newsroom
Wiring the Agent Web
Shared protocols for agents working across companies.
A2A moves under the Agentic AI Foundation alongside MCP
Techzine
The Agent Divide
Large firms scale agents ahead of smaller ones.
40% of large firms say they scale AI agents in a function
FM Magazine
05 · The Wire
01
Most operators have not moved past partial automation.
TelecomLead ↗
02
Verizon says it is piloting agents across production infrastructure.
CMOtech ↗
03
Nokia’s design lets agents act alone on low-risk network work.
RCR Wireless ↗
04
Agents lack a common ecosystem to coordinate across vendors.
ABI Research ↗
The panel on this signal
They agreed on
–In Nokia’s design, agents act alone on low-risk routine work, while high-stakes changes to core networks need human sign-off.
–GSMA Intelligence: most operators have not moved past partial automation.
–Operators still lack a common ecosystem for agents to coordinate across multivendor networks.
The security researcher disagreed on
–Where low-risk ends and high-stakes begins is a setting, and automatic low-risk changes can add up to a high-stakes one.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.