Most CPSs are looking for modernized, cloud native automation platforms, says Blue Planet’s Kailem Anderson
RCR Wireless‘ Kelly Hill spoke with Blue Planet VP and GM Kailem Anderson about what AI is and what it is not in the realm of telecom, and how high-quality data is the heart of real-time, context-aware use cases that will require AI intent.
Editor’s note: This interview has been lightly edited for clarity and brevity.
You’ve had a number of contract wins as of late, like Telefonica Deutschland, Swisscom, Lumen, and others. What are they asking for?
Anderson: Most communications service providers (CSPs) are looking for modernized, cloud native automation platforms. In many cases, the OSS has been around for 20 years or more — built for static networks as opposed to the dynamic, constantly changing world of 5G, Edge, and IoT use cases.
CSPs want modern systems that help them deliver services on demand, with the flexibility to change how the services are packaged. They want a lifecycle set of automation products that meet their needs for wireline, wireless and cloud, and they want a path toward next-gen technologies that can bring an enhanced customer experience (CX) and monetization opportunities.
Speaking of next-gen technologies, how do telcos want to leverage their data and AI, and what’s the biggest hurdle?
Anderson: We must demystify the idea that you can dump all your data into a data lake and get amazing outcomes with AI. This isn’t going to happen.
There’s no context awareness in a data lake, which means you can’t link planning, orchestration and assurance. As soon as the data’s in the lake, it’s old. It might be fine for long-term capacity planning, but not for real-time or context-aware use cases. When you want to drive real-time change in the network, and when you want to leverage agents and agentic frameworks to solve real-time problems, you need context awareness.
“Context awareness” is a term that’s thrown around a lot, so can you better define what it means?
Anderson: The data must have “state” and relationship built into it, so when you make a change to a system related to an event, it plays back to the change and to your planning. In AI, data has to be context aware and it has to have “intent” so you can use it to drive real-time change.
Note from editor: “Context-aware AI intent” is a term that refers to an intelligent system capable of analyzing both historical and real-time data to before acting. It considers many factors, such as network conditions, user history, device types, and business priorities so it can act on high-level business goals (as opposed to just simple, task-based automation).
Intent means that rather than following static rules, AI agents pursue high-level goals, with AI determining the best course of action for a particular outcome. For example, the intent might be to “provide seamless 5G service to all enterprise customers.”
Is there a problem of AI retrofitted to legacy OSS?
Anderson: Yes, there’s a lot of sugarcoating in communications about of what AI is and what it’s not. A lot of vendors are claiming to have “AI capabilities” when, in fact, just leveraging existing rules-based automation systems. And if it’s embedded in the product, the telco really doesn’t have a way to know.
Telcos want AI integrated across planning, inventory and orchestration, and service assurance.
When AI is implemented as a development environment that focuses on democratization of AI, the telco no longer wants to be beholden to one vendor, especially when introducing AI. The telco wants to use its own AI and apply it on top of the data set that the vendor provides.
To that end, we’ve created a development environment through which we introduce AI use cases across the portfolio. We expose that SDK so the telco — its AI teams, and its partners — can leverage our environment to bring its own LLM, and apply it against our data sets.
Then, we take use cases into our agentic framework, so that in addition to the agents we expose, the telco can create its own agents across its own data sets. Again, the telco doesn’t want to be locked into a single vendor that develops AI for them, nor do they want to be locked into a particular AI model, such as Gemini or ChatGPT. They want flexibility to apply LLMs they choose against our data set.
Can agentic AI operate cross domain?
Anderson: Yes, you can train AI agents to function in one area, but when you ask agentic AI to pull network data, or to pull from disparate data sources, it breaks down in its ability to use the data like a human would.
What is the state of cross-domain of agentic AI right now?
Anderson: This is rapidly changing. Today is going to be very different than what you see in six months. A year ago, we weren’t talking about Model Context Protocol (MCP), or agent to agent as part of an agentic framework. Now, they are common.
To have agents work together in agent orchestration is nascent, but protocols like agent-to-agent (A2A) in Linux and MCP empower autonomous AI agents in the automation and management of complex tasks in telecommunication networks. By using MCP to connect AI agents to external data sources and tools, such as legacy systems and network components, we are getting closer to agent-to-agent comms and cross-domain and cross-layer AI and automation.
Are CSPs using agentic AI with MCP now?
Anderson: Yes, for advanced, domain-specific automation across networks, as with inventory management, service orchestration, service assurance, and 5G network slicing.
I believe in six months you’ll see more agent-to-agent orchestration as libraries become richer. We are working on agent-to-agent orchestration, like multilayer networking, where agents discover an optical network, or an IP network, then cross-layer stitching of separate agents, who identify and model out congestion.
We are getting to a point where agents can ingest customer experience or billing data, overlaying that information across agents so that they can handle orchestration or planning for value-add use cases. It’s all evolving, and there’s a need for a robust agent library and for foundation agents. As we learn how to package them for specific use cases, cross-domain, cross-layer agent orchestration will be more sophisticated. It’s a matter of “when,” not “if.”