Photo by Kindel Media on Pexels

The telecom industry has spent years talking about autonomous networks. Self-healing infrastructure, zero-touch provisioning, AI-driven traffic optimization — the vocabulary of automation has become fluent across boardrooms and engineering teams alike. But as the industry edges closer to actually deploying agentic AI systems capable of making real-time decisions without human approval, a critical question has emerged: how do you get operators to trust a machine they can’t fully see inside?

At DTW Ignite in Copenhagen — one of the industry’s premier gatherings for digital transformation in telecommunications — Cisco stepped forward with a framework that may offer the most pragmatic answer yet. Rather than pitching a leap of faith into full autonomy, Cisco is advocating for a graduated trust model that begins with transparency, builds through demonstrated reliability, and only then unlocks the door to closed-loop operations.

The Agentic AI Moment in Telecom

Agentic AI represents a significant evolution beyond traditional machine learning models. Where conventional AI might flag an anomaly or generate a report, agentic systems are designed to take sequential, goal-directed actions — negotiating across tools, APIs, and data sources to accomplish complex tasks with minimal human prompting. In a telecom context, that could mean an AI agent autonomously rerouting traffic during a fiber cut, dynamically adjusting spectrum allocation in a dense urban 5G deployment, or proactively resolving core network faults before customers experience degradation.

The potential is enormous. Analysts at McKinsey have estimated that AI-driven automation could reduce network operations costs by 20 to 30 percent while simultaneously improving service quality metrics. For operators already battling margin compression and surging data demands, those numbers are hard to ignore.

But the risks are equally real. A misconfigured autonomous action in a live network isn’t a software bug to be patched quietly — it can cascade into outages affecting millions of subscribers, regulatory scrutiny, and reputational damage that takes years to repair.

Open-Loop First: The Foundation of Trust

Cisco’s core argument at DTW Ignite centers on what the company calls an open-loop first philosophy. Before any AI agent is permitted to execute changes autonomously, it must first operate in a recommendation mode — surfacing proposed actions to human operators alongside confidence scores, reasoning chains, and the underlying data that drove the decision.

This approach directly addresses one of the most persistent objections to AI in network operations: the black box problem. Operators have historically been reluctant to cede control to systems they cannot interrogate. By mandating explainability as a precondition for autonomy, Cisco is essentially proposing a probationary period for AI agents — one in which the system proves its logic before it earns its independence.

Confidence scoring is particularly significant here. Rather than binary outputs, Cisco’s framework envisions agents that communicate degrees of certainty — acknowledging, for instance, that a recommended configuration change carries high confidence in normal traffic conditions but reduced confidence during anomalous load patterns. This kind of calibrated uncertainty gives human operators actionable context rather than opaque directives.

Human-Centered Workflow Design

Beyond explainability, Cisco is emphasizing the importance of designing agentic workflows around human cognition rather than simply bolting human approval onto AI-native processes. This distinction matters enormously in practice. An AI system that bombards a network operations center with hundreds of micro-decisions per hour hasn’t empowered human oversight — it has effectively eliminated it through cognitive overload.

Effective human-centered agentic design means intelligent escalation: the system handles routine, well-understood decisions autonomously while surfacing only genuinely ambiguous or high-stakes scenarios for human review. It also means audit trails that are legible to engineers, not just data scientists — timestamped action logs with plain-language summaries that support both real-time monitoring and post-incident analysis.

The Road to Closed-Loop: Earned, Not Granted

The ultimate destination — closed-loop autonomy, where agents act and adapt without human checkpoints — remains firmly on the roadmap. But Cisco’s framework treats it as an achievement to be unlocked progressively, calibrated to specific domains, network segments, and risk profiles rather than applied as a blanket operational mode.

A mature deployment might see closed-loop autonomy operating confidently in well-understood scenarios like routine firmware updates or predictable traffic load balancing, while maintaining open-loop advisory roles in more complex domains like cross-domain service assurance or security response. This tiered model aligns closely with the TM Forum’s Autonomous Networks framework, which defines six levels of network autonomy from fully manual to fully autonomous — a reference architecture that is gaining significant traction among major carriers globally.

Industry Momentum and Competitive Landscape

Cisco isn’t alone in this conversation. Ericsson, Nokia, and a growing roster of cloud-native startups are all advancing their own agentic AI narratives for telecom. What differentiates the trust-first framing is its acknowledgment that technical capability and operational readiness are not the same thing. Building an AI agent that can autonomously manage a network segment is a different engineering challenge than building one that operators will actually allow to do so.

For carriers evaluating agentic AI investments, the Cisco framework offers a practical procurement lens: prioritize vendors who can demonstrate not just model performance but explainability infrastructure, confidence calibration, and workflow integration that genuinely supports rather than bypasses human judgment.

Outlook: Trust as the New Technical Requirement

As the telecom industry moves deeper into 5G Advanced and begins laying conceptual groundwork for 6G — where network complexity will dwarf anything operators manage today — the question of autonomous operations will only intensify. The networks of the next decade will likely be too dynamic and too intricate for purely human-managed operations at scale.

But the path to that future runs directly through the trust deficit that exists today. Cisco’s message from Copenhagen may be the industry’s most important reminder that in the race toward agentic autonomy, the fastest route is not the most aggressive one — it’s the most transparent.