Akshat Mohindra, Leader, Software and Services – APJI, Ciena
Artificial intelligence (AI) has been around for many years. What has changed is that we have found new ways to explain it. I believe this is also how a community progresses and learns. To understand where we are today, it helps to trace the eras that brought us here.
AI eras
The first was the rule-based era, which was deterministic and governed by strict logic. The second era was predictive AI, which is centralised but informed by historical evidence, where guidance already exists, and all you need to do is leverage that historical learning to make decisions. Then came generative AI, using AI to generate content and ideas on your behalf. When you take it up a notch, you arrive at agentic AI, where you have further outsourced tasks, with certain rules and activities, to agents that run the show for you. We are now on the cusp of the growth of generative and inference AI, and eventually, agentic AI will encompass the final era, moving toward what I would call superintelligence, the rebirth.
Why networks must become autonomous
The communication industry has gone through massive transformation across eras. One of the most significant parts of the transformation was automation. Legacy systems were the first to be automated, but that automation happened in silos. That is human nature – automate what is easiest first, then move forward. The challenge is that the proliferation of AI has simultaneously increased the market demand for connectivity. The same infrastructure that once serviced a certain volume of demand now has far more tasks coming its way. How do you manage that? That is precisely where autonomous networks come in.
The case for autonomy is not driven by end users demanding it. It is driven by risk. To address the risk of challenged operations and missed opportunities, the market turned to automation. There are many challenges including multi-vendor integrations and lack of feature functionality making standards alignment essential for CSPs. Legacy systems are being challenged by newer, more intangible architectures. Further, data-driven alignment is of prime concern, which is the bedrock of why you need an AI-based system – because you can’t AI what you can’t see™. Petabytes of data flowing through the network are not unified and not consistent.
Moving up the maturity curve
The TM Forum, a globally recognised standards body, has published an autonomy maturity scale from L0 to L5. Most operators today sit between L0 and L2, not because they lack intent, but because they are bound by business systems, legacy software and budget constraints.
At L1, organisations start putting processes and scripts in place, making certain tasks repetitive and software-driven. At L2, they put open loops and policies in place, and the system reads through documents to deliver feedback (still open loop), while moving towards autonomy. At L3, intelligence begins to show. Systems start interpreting intent, maintaining a service level agreement during a traffic surge, and dynamically coordinating across domains. A common example is automated capacity optimisation during a major event, such as a cricket final or Super Bowl, where traffic spikes put a tremendous load on streaming partners, and the network dynamically understands and manages that surge. At L4, operators define outcomes. The conversation shifts from features to results. When we get to L5 is not the point. Fortunately, the benefits along the path matter.
The agentic era is already here
For generative AI, graphics processing unit consumption is the primary resource driver. For agentic AI, you will see far more central processing unit consumption because you are giving agents a significant set of tasks to execute. At the core of an agentic architecture is the supervisor agent, coordinating specialised agents across functions from schematic design that translates intent into logical design, to service assurance, route analytics and automation. The magic in agentic AI lays in the data, context and shared decisioning across the network. Grounding your strategy in the right OSS-embedded agentic approach is what will change “automation” from an efficiency boost to business-model catalyst.
The fibre, the hardware, the routers – all of it – are depreciating assets over time. It is only the software, continuously upgraded and evolved, that adds value to that same hardware and enables you to deliver more to your end customers.
If you are not talking about agentic AI today, you need to start. The pace of adoption is accelerating across the globe. Autonomous networks may sound like a pipe dream, but benefits begin at L2 and L3. Organisations do not need to wait for full autonomy.
