Login
Register
Telstra's composable architecture has helped reshape not only how the company develops network products, but, crucially, how customers use them. And it has also laid solid foundations for agentic AI as the company examines how to model the context in which agents act.
Joanne Taaffe, TM Forum
Telstra creates new network value and lays agentic AI foundations with ODA
When Telstra released its Adaptive Networks Product Experience (AN PEX) for enterprise customers in May, it represented much more than a simple product launch. It was the fruit of a profound and ongoing change in the company’s structure, products, working practices and culture underpinned by a composable architecture based on TM Forum’s Open Digital Architecture (ODA), as Mark Sanders, Chief Architect, Telstra, explains in an interview with TM Forum.
“Adaptive networks product experience represents a strong inflection point. It meant a lot to our people, and meant a lot to our organization,” says Sanders. And he expects the architectural and modelling discipline behind it will serve Telstra as the telecoms industry looks towards AI agents.
Through a set of canvas-like tools, AN PEX allows enterprise customers to design, configure, implement and change their network solutions through real-time visualization and intuitive drag-and-drop interfaces.
Telstra’s teams drew on the example of hyperscaler service offerings to create AN PEX, describing it as “an inversion of control” because it places network product decisions firmly in the hands of enterprise customers and the operator’s partners.
Hyperscalers “bring various pieces of infrastructure, and they connect together and deploy software and place workloads. The customer is very much solutioning that and ... working out the costs … and getting quotes,” points out Sanders. “We’ve adopted that for networks.”
This customer-driven approach to developing and delivering the network as a product was made possible by the Telstra Reference Architecture Model, or TRAM. Based on TM Forum’s ODA, TRAM uses an API-first approach to delivering composable network and technology services.
The composable way of working inherent to ODA, and by extension TRAM, allowed Telstra’s product owners “to show high degrees of innovation and the ability to think fundamentally differently and experiment on different business models, different customer experiences, and how the customer is going to interact with that product and interact with our organization,” explains Sanders.
Composability also gave teams across the business the flexibility they needed to collaborate on developing, finessing and delivering PEX, which won a TM Forum Excellence Award.
“The composed product wasn’t from one part of the organization. It was a cross-company piece of work with re-usable components,” points out Sanders. Indeed, increased cross-function collaboration means that “with awards we’re often sitting there going, where does the trophy go? Because it touches a huge amount of teams.”
Composable ways of working have increasingly shaped the operator’s strategy since it started work on ODA as part of its T25 three-year plan, in 2022 and is now a key part of the Connected Future 30 Five-year strategy.
TRAM has “really changed how we work inside the organization,” says Sanders, informing not only investment and planning, but also “a lot of our business processes, a lot of our method of making decisions and acceptance of technical debt.”
He adds: “We’re very disciplined now, from how we are modelling our business to modelling the technology that needs to update, evolve and change.”
The network-as-a-product vision, of which PEX is a part, is a core element of Telstra’s latest five-year strategy, Connected Future 30, which launched in May.
“When we think of Connected Future 30, we’ve coined a phrase ‘network as a product’ … to signify how we might find different ways to drive value from our network,” explains Sanders.
ODA and Telstra’s TRAM have also laid solid foundations for future development of AI agents.
“What's going to stand us in good stead [with AI agents] is the discipline of the TMF and the modeling discipline that comes behind it and the principle … of abstraction and composability,” says Sanders.
AI and AI agents, for example, require rich contextual information to perform complex network automation tasks. But it has to be the right information at the right point in a process. For this reason “the importance of modeling context is critically important.”
Defining boundaries is key to context modelling and, as Sanders points out “TMF, ODA, TRAM – it is all about bounded contexts of technology … and using APIs to abstract components. We want to make sure we adhere to those principles. So, I think the first starting point is to leverage that foundation … and that framework.”
Sanders emphasizes the importance of telcos collaborating to ensure they continue to benefit from the flexibility that composable IT has ushered in.
“If you look at a lot of the vendor proposals … they’re reverting back to: ‘I’ll give you an agent, I’ll give you a data structure, I’ll give you a data solution, and I’ll give you the tech that I’m actually here to sell you,” warns Sanders. “The first … action for the industry is making sure … the agent and AI era continues to adhere to decoupled abstraction.”
As a result, Sanders believes “there is a whole piece of work around a loosely coupled architecture and reasoning.” He adds: “What we don’t want to do is couple ourselves to … [an] LLM [large language model] that we have anchored on our whole business with point-to-point integrations. “
For Telstra, being anchored in context means being anchored in knowledge.
“Anchoring in knowledge not only benefits agents and LLMs, but it benefits any form of structured reasoning where we want more than a probabilistic answer, particularly when we get into parts of the network and at the right level of accuracy for the right job.”
Not every area of the business will expect the same performance from agents, points out Sanders. Whereas autonomous networks that are becoming intelligent will place high demands on the acceptable probabilistic outcome, other areas of the business may rely more heavily on the judgement of human co-pilots.
“I’d like to see the industry move forward with a shared language, an ontology of how we can model knowledge for a telco business … so we can improve on the reasoning that we’re doing,” within a given context.
This would see telcos move from simply passing data to “figuring how we’re passing technology knowledge at this point in time.”
When it comes to the network, “the knowledge needs to be semantically rich” according to Sanders. “It needs to not just be knowledge around the topology of the network. It needs to have business processes coded in, our engineering design limits, our policies and even regulatory frameworks,” he explains.
“Those things need to be coded into the knowledge behind it,” says Sanders, based on a shared understanding of the attributes that can be linked to knowledge.
“There’s going to be some things that are unique to our business, but there’s going to be things that are common to this industry,” says Sanders, “and this should happen within TM Forum.”
