How do optical networks move beyond cruise control to agentic operations?

AI is reshaping expectations for automation as organizations adopt natural language prompting, inferencing and AI/ML-assisted workflows to evolve their applications and operations. 

In my last post, I described how AI-assisted automation created the cruise control needed to transition to intelligent network operations and reduce service delivery time to minutes. This transition will increase optical connectivity demands for GPU compute resources hosted in data centers and enable network to meet these demands much more rapidly. 

As AI inferencing matures, optical transport networks will become larger and more complex. This calls for a new agentic approach to automation. By combining intent-driven closed-loop operations with optical-domain intelligence, agentic automation can help network operators convert KPI data into insights, recommendations and actions that support proactive issue resolution, SLA protection and better task prioritization.

What is an optical-domain agent?

TM Forum classifies autonomous network taxonomy from 0, fully manual, to 5, fully autonomous. Many optical network operators have achieved level 2 or 3 automated operations for selected use cases and are now setting their sights on level 4.

An optical-domain agent is a system that accesses transport-layer data, planning models, historical insights, inventory and alarms to break down an intent or problem into multiple structured steps and perform complex tasks autonomously. Guardrails such as policies, safety mechanisms, rules, algorithms and human oversight help ensure that AI systems operate responsibly and address problematic behavior when needed.

Agents also host the data required for functional operations. They are context-aware and can relate current conditions to past network events. With processing capabilities, they can compile algorithms, develop functions, extract patterns from network data and access LLMs or broader content to support deeper reasoning. Agents can also communicate with other agents or tools to fulfill their intent.

How are reasoning and context adopted in optical network automation?

Optical-domain agents can develop cognition through embedded memory that stores documented operating procedures, current and historical network insights, and vendor-based documentation. They can detect and measure anomalies, screen out-of-scope context and provide recommendations for humans to review before changes are made to the network. Higher-risk recommendations can be verified by humans in the loop or humans on the loop through a digital twin.

AI/ML-enabled agents can access optical information and tools before inferencing and formulating a response. Their knowledge base can span product resources, industry standards and vendor-specific method-of-procedure playbooks. 

Retrieval-augmented generation (RAG) combines inferenced outputs with domain-specific sources owned by network operators to produce meaningful, relevant responses. Operators can rely on RAG to:

• Reduce hallucinations through improved documentation.
• Improve responses by updating documentation libraries.
• Provide responses that can be traced to a source.

RAG has two stages: fast search and re-ranking. Re-ranking stops the model from reasoning over irrelevant information. Together, the agent and RAG are like an open-book exam where the outcome is based on what type of information is used and how the student solved the problem with it. Using irrelevant material produces false answers.

Optical agents have pre-built closed-loop capabilities, including access to tools and APIs that can push configuration information to the network. Today, the industry is taking a cautious approach and driving a mandate for vendors to make sure their agentic systems keep humans in the loop and adhere to internal operational policies when they write to the network. Agents must get human approval and produce an auditable report before executing consequential write actions. Notification mechanisms, response summaries and action reporting are minimum requirements to build trust.

What role can agents take in optical network automation? 

According to Appledore Research, the top telecommunication provider use cases for AI agents are network optimization and service assurance. Nokia WaveSuite and Transcend address these use cases with capabilities that use AI-assisted operations or agents to:

• Detect anomalies, capacity utilization and silent degradation.
• Reason across alarms, KPIs, network topology and logs.
• Suggest corrective actions for humans to review .
• Continuously monitor the impact of implemented actions.

With agent-enabled automation, operators can reduce OPEX and improve service quality. WaveSuite and Transcend support an explainability framework that provides auditable real-time logs for each task across dashboards, to-do queues, AI reasoning traces, historical context and performance metrics. These platforms will evolve to adopt Model Context Protocol (MCP) interfaces to connect domain agents with OSS/BSS to report their findings.

How does agentic automation change infrastructure planning?  

Agentic automation changes the way compute infrastructure is dimensioned. Agents may need to access GPU resources or other tools multiple times to formulate a response. When they access GPU resources, they consume units of text known as tokens, which have OPEX implications and vary for different use cases. 

WaveSuite and Transcend support Bring Your Own Cloud across on-premises GPU resources, private clouds, hybrid configurations and the operator’s own model hosting platform. They also provide consumption dashboards that let operators track tokens when agents access external LLMs and reset token capping requirements based on their OPEX and business objectives. 

How can agent automation address industry priorities for adoption?

Operators have three key priorities for the shift toward agent-based automation.

1. Interoperability and agent-to-agent communication

Optical-domain agents need a common interface that lets them tell each other what they can do and authenticate their identity. They also need a policy layer that governs what they can share, along with an explainable audit trail for their interactions. TM Forum’s Open Digital Architecture documents these characteristics and will be essential for connecting service operations agents to domain agents from OSS/BSS to ensure interoperability with other vendors’ agents. 

2. Silent failures and predictive insight

Optical-domain agents can expose hidden issues to operators. By tracking KPI metrics such as flapping alarms, span degradation, KPI drift and statistical outliers, they can address silent network degradation before services fail. This helps operators move from reactive to proactive and predictive operations.

3. From insights to action

Optical-domain agents are bounded by the data and tools they are allowed to access and the output they generate. Today, operator policies ensure humans play an essential role in validating output from agents. Consequently, digital twins are gaining traction as an important and effective way for operators to validate agents’ recommendations for high-risk decisions before going live. 

Nokia optical engineering services can create digital twins that validate and summarize actions executed by agent-based automation. This shifts the operator’s role toward responsible decision-making and auditing of agent-produced outcomes.

Accelerate your automation journey with Nokia optical agents

Nokia can help your business stand out in a digital economy fueled by AI. With automation powered by our optical-domain agents, you can accelerate oversight of continuously scaling AI networks, reduce TCO and focus on strategic initiatives to improve long-term business objectives.

Contact us to schedule a demo or trial of agentic use cases, or to discuss ideas for co-creating use cases that meet your business needs.