Nokia and Google Cloud launch six AI agents with Gemini to automate telecom networks

By DPL News · June 23, 2026.

Nokia and Google Cloud have announced a collaboration to integrate Google's Gemini models within the Nokia Assurance Center software suite. The central objective is the development and deployment of six AI agents specialized in telecommunications that will accelerate the path toward fully automated and autonomous networks. The partnership responds directly to the challenges facing telecom operators: managing increasingly complex network infrastructures, with continuously growing data traffic volumes, while maintaining security and operational continuity.

The starting context is that traditional and manual network management methods are losing effectiveness in the face of the scale and speed that modern networks demand. Google and Nokia expect these agents to help telecom providers distinguish critical infrastructure problems from mere operational 'background noise,' in order to speed up repairs, reduce downtime and protect operators' revenue.

Vivek Jaiswal, senior vice president of Autonomous Networks at Nokia, declared: 'The AI era demands a new kind of network: one that is programmable, AI-native and capable of operating at machine speed. With Gemini-powered agents integrated into Nokia's automation portfolio, we are helping telecom providers move beyond manual operations to maximize performance, ensure reliability and find new efficiencies in their data.'

**The six specialized agents in detail**

The collaboration is structured around six agents with well-differentiated functions:

1. **Router agent**: Acts as the central orchestration layer. It interprets user intent and manages communication among the rest of the agents, ensuring at all times compliance with the operational guardrails established by the operator.

2. **Event classification agent**: Analyzes ongoing alarms and compares them with historical patterns to identify root causes and assess the real operational impact of each incident.

3. **KPI selector agent**: Provides expert domain interpretation of network performance metrics, including complex definitions and units of measurement, to support the reasoning of the other agents.

4. **Anomaly reasoning agent**: Investigates unusual network behavior to determine whether a detected deviation is a genuine problem or a false alarm, thereby reducing unnecessary workload on technical teams.

5. **Action reasoning agent**: Compares active events with automation catalogs to recommend specific corrective steps, acting as an advisory layer before executing any action.

6. **Dashboard agent**: Allows technical teams to quickly generate visual analyses and monitoring screens through natural-language prompts, without the need for complex configurations.

**Technical architecture: Google Cloud's ADK and Gemini Enterprise Agent**

Nokia developed these capabilities using Google Cloud's Agent Development Kit (ADK) on top of the Gemini Enterprise Agent platform, with intelligent routing and conversation paths that leverage Gemini's multimodal reasoning. A relevant technical aspect is that the multi-agent framework does not require complex managed services: it runs on the standard storage and compute of each operator's cloud infrastructure, which eliminates the need to acquire new dedicated compute capacity.

This design decision is significant from an adoption standpoint: by being able to run on already existing cloud configurations, telecom operators avoid additional infrastructure costs, which lowers the barrier to entry and accelerates deployment times.

**The 'glass box autonomy' model: humans in the loop**

One of the most relevant elements from the perspective of responsible agentic AI is the concept Nokia calls 'glass box autonomy.' Instead of excluding human operators from the decision-making process, the framework introduces the action reasoning agent as an advisory layer. This agent presents its recommendations to human engineers, who retain final approval over critical checkpoints before any correction is executed and automatically logged.

For low-risk scenarios previously approved by the operator's internal policies, the same architecture can support fully closed-loop automation, that is, without human intervention. This places the solution on a spectrum of configurable autonomy: from operator assistance to full automation, depending on the level of trust and operational risk that the operator itself defines.

This selective 'human-in-the-loop' approach is especially relevant in the current regulatory context, since it allows operators to comply with human-oversight requirements for critical decisions, while maximizing efficiency in routine low-risk tasks.

**Quantified impact: reduction of times and false alarms**

Nokia provides concrete figures on the expected impact of the multi-agent architecture:

- Reduction of network problem resolution times by between **50% and 80%**. - Reduction of false alarms and unnecessary escalations. - Possibility of running on existing cloud configurations, avoiding new infrastructure costs for operators.

These figures, if confirmed in real deployments, would represent a very substantial operational improvement for telecom operators, whose network operations centers (NOC) manage tens or hundreds of thousands of daily alarms in large-scale networks.

**Availability and commercial roadmap**

The router agent and the event classification agent are already in operation at the time of the announcement. When the platform officially launches as a SaaS model on Google Cloud Marketplace, expected in **September 2026**, operators will be able to immediately implement this initial package of certified agents to work with Nokia Assurance Center. The remaining agents, described as more complex, will be delivered through continuous software updates after the launch.

The distribution model as SaaS on Google Cloud Marketplace is relevant: it facilitates acquisition, billing and integration for operators who already use Google Cloud as a cloud provider, and reduces the procurement friction common in the telecommunications market.

**Nokia's broader program toward autonomous networks**

The collaboration with Google is part of a broader Nokia program to accelerate full autonomy in the operation of telecommunications networks. Nokia argues that these networks will need to be prepared for what it calls an AI 'hypercycle,' which will exponentially increase token, session and data traffic in the coming years, as AI models become integrated into more applications and services.

In parallel with the Google Cloud announcement, Nokia also presented the latest version of its autonomous networks suite, with improvements in RAN (Radio Access Network) automation and new AI-based frameworks for IP, fixed and optical networks. The suite introduces on-premises deployment options and new use cases aimed at improving concrete business outcomes, such as better VoLTE service quality, greater network observability and optimized user experience in radio access networks.

In addition, Nokia's MantaRay SMO solution—aligned with Open RAN standards and with multi-vendor field capabilities for traditional RAN—now incorporates non-real-time RIC functionality, with AI-enabled rApps that manage complex radio networks, detect anomalies and support dynamic network slicing.

**Implications for agentic AI in telecommunications**

This announcement is a practical and concrete example of how agentic AI—that is, systems composed of multiple agents with specialized roles that collaborate to solve complex tasks—is finding industrial application in a sector as critical as telecommunications. The multi-agent architecture presented by Nokia and Google follows the orchestrator-specialized-agents pattern that has become established as a reference model in the agentic AI ecosystem: a central agent (the router) coordinates agents with specific competencies (classification, KPI analysis, anomaly detection, action recommendation, visualization).

The choice of Gemini as the base model and of Google Cloud's ADK as the development framework reflects Google's bet to position its agentic platform in high-value industrial verticals, where the complexity of the data and the criticality of the decisions justify the use of advanced multimodal reasoning models.

As sector context, the AI-driven telecom network automation market is one of the fastest-growing segments in the field of enterprise AI. Telecom operators manage networks with millions of nodes, sensors and alarms, which makes AI-based automation not only desirable but practically necessary to maintain operational competitiveness.

**Comparison with the approach of other sector players**

In general, Nokia competitors such as Ericsson and Huawei are also developing automation and AI capabilities for networks, although with different partnership models and architectures. The Nokia-Google alliance with Gemini competes directly with similar initiatives that use other large language models and cloud infrastructures such as AWS or Microsoft Azure. The fact that Nokia opts for Google Cloud and Gemini—instead of building proprietary AI capabilities—indicates a sector trend toward specialization: network manufacturers focus on the specific domain of telecommunications knowledge, while delegating AI reasoning power to the major foundation model providers.

**Regulatory perspective**

From the perspective of the European AI regulatory framework, the 'glass box' design with human oversight for critical decisions is aligned with the principles of the EU AI Act, which requires adequate levels of human oversight for AI systems used in critical infrastructures. Telecommunications networks are considered critical infrastructure in most jurisdictions, which potentially places these systems in high-risk categories under the EU AI Act. The fact that Nokia has explicitly designed a model where the human retains final approval over critical decisions appears to be a conscious response to this regulatory environment.

**Outlook**

The official launch on Google Cloud Marketplace in September 2026 will be the first real test of mass adoption. The next indicators to follow are: how many operators adopt the solution in its first months of availability, whether the resolution-time reduction metrics (50%-80%) are confirmed in real production environments, and how the delivery of the more complex agents evolves through the promised software updates.

In the longer term, if the multi-agent architecture proves its value in telecommunications networks, it could become a reference model exportable to other critical infrastructure sectors—energy, transport, utilities—where alert management, anomaly detection and response automation present structurally similar challenges.