Telstra is laying the foundations to use multi-agent systems in its processes, with an AI joint venture with Accenture and modernisation work on its data infrastructure contributing to the ambition.

Executive director of Data and AI, and joint venture co-CEO Dayle Stevens told iTnews that the telco has set a goal to “reinvent all of our processes with agentic AI.”

It will initially “go after two processes” but “hasn’t decided on exactly which ones” yet.

But the ambition is to source AI agents from multiple places: from its existing technology vendors, as well from Accenture via the consultancy’s ‘AI refinery’ offering.

“The very first thing to do though is to get the agentic architecture and platform in so that we can have any agents plugging into it in future,” Stevens said.

“We have a lot of partners, a lot of software providers and our own engineering teams as well. We have Workday, SAP, ServiceNow [and] Salesforce, and all of them are creating their own agents for their own software packages.

“Our goal is to be able to orchestrate all of that.”

The joint venture with Accenture has multiple potential roles to play here.

First, its ‘AI refinery’ – the umbrella branding Accenture gives to a suite of industry and domain-specific agents – will act “as an accelerator” for Telstra’s ambitions.

Second, a ‘Silicon Valley hub’ at Accenture’s offices in Mountain View is to be used as a space to pool Accenture’s expertise and to collaborate with technology partners based in the area.

“We’re creating a space to bring in engineering experts from some of our partners like Microsoft, Databricks and AWS and others into a space where we can collaborate, co-design and do different things together to be able to bring it into Telstra as well,” Stevens said.

While Telstra staff may occasionally visit, it’s not intended to be a Telstra “outpost” or a place the telco’s engineers physically spend time at, unlike similar models from other large Australian firms.

Most of Telstra’s data and AI team is based in either Australia or India, and interactions with the Silicon Valley hub would primarily be virtual.

“We wanted to … make it part of our team as well, not just go and do a trip once a year and get a whole lot of ideas and bring them back and try and make them work,” she said.

Data and AI strategy

One of the intentions of the joint venture with Accenture, which was announced earlier this year, is to help Telstra deliver its data and AI strategy.

Stevens outlined several pillars of the strategy.

First, the telco is simplifying and modernising its data landscape and data foundations in preparation.

“We have a lot of legacy technology systems, so we have a lot of legacy data that sits around that,” she said.

“A couple of years ago, we had nearly 80 different data platforms. We’re down to 30 but we want to get to three.”

This appears to comprise data in Azure, AWS and on-premises. The on-premises infrastructure hosts Telstra’s network data, which Stevens characterised as the telco’s “gold”.

The current thinking is to take advantage of a data mesh architecture to enable access to data sources, with some work with Microsoft Fabric underway.

Outside of data foundations and accessibility, the modernisation effort is also targeting improvements to data quality and “protection”.

The second pillar of the strategy is responsible AI. This is also listed as an “initial” focus area for the joint venture.

While the telco already has some real-time model monitoring in place, Stevens said there is a need to improve what it has to keep pace with its planned adoption of agentic AI.

“Part of my job is always to be balancing the risk and the opportunity,” she said.

“What I’m concerned about is that if we go as fast and scale as quickly as we want to, then we have to scale our responsible AI capabilities ahead of that curve.

“If we are tapping into that global experience on agentic or whatever else we might go after, the way of governing, monitoring and assuring AI has to stay ahead of that scaling.”

The third pillar is to build AI foundations, including the agentic architecture and platform.

“Before we even get to AI use cases and the value and opportunity for AI, [we need to think about] building reusable, repeatable AI foundations like an agentic platform to orchestrate agents,” she said.

“I really want to move away from AI being scaled use case by use case. 

“We’ve sorted out the data foundations, we’ve got AI foundations in place, we’ve got really strong responsible AI practices, and then it unleashes how much we can do on the AI front putting AI into the hands of people.”

AI reuse is already occurring, Stevens said, citing Ask Telstra, which is used by employees to search the telco’s knowledge base for answers to questions.

“We were able to move relatively fast on Ask Telstra because of the work that we’d done in migrating to the public cloud and having [a] machine learning ‘factory’ setup back then.

“We’ve just released Ask InfraCo, and it took us 88 percent less time to release Ask InfraCo than it did for Ask Telstra. 

“So, we have reuse in the gen AI space, [and] the same will go for agentic AI. We want to get to that reuse capability.”