Nokia Integrates Google Cloud's Gemini Models and Deploys Six AI Agents for Telecoms
Nokia and Google Cloud have announced an expansion of their partnership to incorporate Google's Gemini models into Nokia’s network management software suite. This integration will enable the deployment of six specialized AI agents designed to automate problem-solving for telecommunications operators.
Six AI Agents to Accelerate Network Automation
The partnership focuses on integrating Gemini models into the Nokia Assurance Center, Nokia’s network management software solution. The two companies will develop six specialized AI agents, each designed to handle distinct operational tasks or to collaborate in solving complex network issues. These agents operate based on Google Cloud’s Agent Development Kit (ADK) and run on Google Cloud’s standard computing and storage infrastructures. Nokia emphasizes that this multi-agent framework operates seamlessly on standard tools such as Kubernetes and Google Cloud Storage, ensuring full compatibility with existing client environments.
SaaS Launch Scheduled for September 2026
Two of the six agents — the routing agent and the event sorting agent — are already fully operational. The official launch of the platform as a SaaS service on the Google Cloud Marketplace is scheduled for September 2026, at which time operators will be able to immediately deploy this first set of certified agents. The remaining four, more complex agents will be delivered progressively through successive software updates. These continuous deployments will extend to broader applications of Nokia’s network portfolio (Unified Inventory, Data Suite, and Orchestration) starting from late 2026 and throughout 2027.
A Recommendation Model Preserving Human Intervention
The architectural framework introduced by Nokia is based on what the company describes as 'glass box autonomy.' The action reasoning agent functions as a consultative layer, presenting human operators with recommendations accompanied by a confidence score. Engineers retain final approval before automatic execution and logging of corrections. For low-risk scenarios approved by operational policies, the same architecture can also support fully closed-loop automation.
