I have shared a number of posts on what Agentic AI can do in Telecoms. I think it is time to reflect on some of the interesting work taking place in standards bodies and research communities (e.g. MUSE, SafeAdapt, ENTRUST, 3GPP, ETSI and IETF) around adaptive agentic intelligence, continuous learning and governed autonomy.

The evolution of Agentic AI is worth considering.

The first generation was largely about execution. We gave AI agents tasks, defined objectives and expected them to get things done.

The next step was collaboration. Agents began working alongside people, other agents and enterprise systems, coordinating activities and handling increasingly complex workflows.

But I believe the real turning point comes with Agentic AI 3.0.

Integrity and trustworthiness.

This is particularly important in Telecoms, where decisions made by autonomous systems can have a direct impact on network availability, service quality, security and customer experience.

An agent that can reconfigure network resources, for example, must do more than identify an optimisation opportunity. It needs to understand the potential impact, respect operational policies, consider dependencies, explain its decisions and operate within clearly defined authority.

The same applies to security. Detecting and responding to threats is valuable, but doing so without creating service disruption or introducing new vulnerabilities requires trust, accountability and appropriate controls.

For CSPs, integrity and trustworthiness cannot be afterthoughts. They need to be fundamental to how agents operate.

So, where does this take us next?

I would argue that Agentic AI 4.0 is about adaptation, continuous learning and governed autonomy.

Consider some of the practical challenges facing telecom operators today:

  1. Network configuration and optimisation: Moving beyond static rules and predefined automation towards agents that learn from network behaviour, changing traffic patterns and service demands, and adapt configurations while respecting change controls and service assurance constraints.
  2. Security operations: Moving beyond alert detection and response towards agents that learn from emerging attack patterns, correlate threats across domains and adapt defensive measures without compromising network availability or security policies.
  3. Network assurance and field operations: Addressing the persistent challenge of fragmented fault information, lengthy root-cause analysis and inefficient truck rolls. Agents could correlate alarms, topology, telemetry and historical incidents to identify likely causes, recommend corrective action and coordinate field engineers. Over time, feedback from resolved incidents could improve future diagnosis.
  4. Customer services: Moving beyond scripted interactions towards agents that understand customer context, identify the underlying cause of recurring problems, coordinate actions across billing, CRM and network systems, and learn from service outcomes. The objective should be fewer repeat contacts, faster resolution and a better customer experience.

The real opportunity is not simply to make agents more autonomous. It is to make them more context-aware, adaptive and capable of improving through experience, while remaining within clearly governed boundaries.

I see the evolution as:

  1. Agentic AI 1.0 – Execution
  2. Agentic AI 2.0 – Collaboration
  3. Agentic AI 3.0 – Integrity and Trustworthiness
  4. Agentic AI 4.0 – Adaptation, Continuous Learning and Governed Autonomy

Each builds on the previous generation.

One important distinction: continuous learning does not necessarily mean agents changing their own underlying models or rewriting their own rules. It can also mean learning from operational feedback, refining decisions and adapting strategies through controlled, validated processes.

Perhaps the next big challenge for Telecom CSPs is not just deploying agentic AI, but establishing the foundations that allow it to learn and evolve safely at scale.

We know how to build agents that execute and collaborate. The industry, standards bodies and research committees are now working towards agents that can be trusted. The next frontier is building agents that can learn, adapt and evolve without compromising that trust.

Is Agentic AI 4.0 the next logical step, or are we overlooking another dimension of this evolution?

About the Author

Chief AI Officer at Taskgentic.ai. Leading the development of enterprise agent architectures that combine collective intelligence, modular workflows and rigorous security guardrails for complex operational environments.

About Taskgentic.ai

Taskgentic.ai is developing the next generation of AI-driven task intelligence, helping Communications Service Providers move from isolated AI agents to secure, reusable and governed ensembles and assemblies that deliver measurable business outcomes.