- Nokia and Google Cloud are turning Gemini-powered AI agents loose on telco network troubleshooting
- The first agents target alarms, KPIs, anomaly detection and remediation, with more on the way
- Nokia says the agents could speed up network fixes and lower costs, but that doesn’t mean opex is disappearing entirely
DTW IGNITE 2026, COPENHAGEN, DENMARK – Nokia teamed up with Google Cloud to infuse AI into telecom operations, leveraging the latter’s Gemini technology to build six specialized agents capable of tackling complex network problems.
“What they essentially do is they bring in a certain amount of autonomy in the way networks are managed today,” Vivek Jaiswal, SVP of autonomous networks at Nokia, told Fierce. “It allows us to move telecom operators past the manual troubleshooting, and by embedding these agents into our software, networks can automatically triage the issues and probably fix those issues very, very rapidly.”
The initial cohort of agents includes a router agent, which acts as a central orchestration layer; an event triage agent, which analyzes ongoing alarms; a KPI selector agent, to aid reasoning by acting as a network performance domain expert; an anomaly reasoner agent, which sorts real issues from false alarms; an action reasoner agent, to recommend remediation steps; and a dashboard agent.
Nokia plans to launch its agentic platform in Google Cloud Marketplace in September, at which point operators will be able to deploy the the router and event triage agents via Nokia Assurance Center. The remaining four agents will be made available on a rolling basis via software updates.
For those interested in a first look, Nokia and Google are showcasing a demo of the initial agents here in Copenhagen this week.
But Rodrigo Brito, Nokia’s VP of secure and autonomous networks, told Fierce that it has “plenty of other” agents in its pipeline beyond those six. These include a topology expert, a services design agent and security agents, he said.
While they are being packaged together in Nokia’s software platform, Brito added each agent is basically its own small software asset with the lifecycle and investment of a product. Basically, a lot of time, attention and care went into each.
Building network intelligence
Work on the project sprang out of the launch of Google Cloud’s Autonomous Network Operations Framework at DTW last year. “This is an excellent proof point of where we are moving from talking and pilots to outcomes,” Sridhar Gollapudi, Google Cloud’s head of telco industry solutions, said.
After spending a few months learning how to work together, the pair got cracking. Jaiswal said it took about six or seven months to get to where things stand today, with work picking up steam around February of this year.
The companies took a “glass box” approach, which combines autonomous capabilities with observability and human oversight. This ensures that while AI is doing all the heavy lifting in terms of data analysis and figuring out the problem, human engineers retain control over the decisions, Jaiswal said.
There is an option to fully automate certain actions based on historical patterns and the AI’s confidence levels in its assessment, but in most cases the AI will just present its analysis and recommendation to a human for approval rather than taking action on its own, he added.
Renata Silva, head of Nokia’s Autonomous Networks Business, added there are a number of differences between traditional telco machine learning capabilities and the new agents. She noted that while constant learning happens with both, machine learning is better for deterministic tasks as well as analysis and pattern recognition across large data sets. The agents, in contrast, are more suited for reasoning and complex actions. And because they can explain their conclusions, the agents also elicit more trust from users, she said.
“If you ask me my opinion, that’s the reason why autonomy has not evolved so much while we only had machine learning in the picture,” Silva said. “Now there’s a huge new wave of investment in this area because the agents are that giving extra explainability and trust that was not there before. It’s not a black box.”
Asked how the agents were trained, Jaiswal said Nokia relied both on its own experience and partnerships with customers to compile the basic data needed to get the agents functioning properly. He noted, though, that because every operator environment is different, some fine tuning may be required from operators that adopt its technology.
The cost question
By deploying these agents, Nokia said operators will be able to slash network problem-solving times by 50% to 80%. The idea, in theory, is to reduce costly downtime and slash operational costs.
But with AI token costs creeping up on enterprises across the board, Fierce wanted to know whether adopting these agents would actually result in a net cost savings.
Jaiswal said as the industry becomes more and more autonomous, the expectation is that operating expenses (opex) will come down. He noted, though, that that doesn’t mean opex will be zero.
“They will have to continue to spend some money. And if today all the money is being spent on manually fixing and finding issues and doing all of that, some of that manual cost will come down and will get replaced, probably by some AI-based cost,” he said. “But overall, the expectation is that the cost will come down for them.”
Catch all of our coverage from DTW Ignite 2026 right here.