From Singapore: Australia's largest telecommunications provider is edging toward a significant operational test of autonomous artificial intelligence, with a live production pilot of Salesforce's Agentforce platform now set to launch within roughly six weeks. But inside Telstra's boardrooms, the enthusiasm is tempered by a hard-nosed commercial question: does the technology actually pay for itself?
Marcella Wells, chief of Telstra's customer sales and commerce engineering group, disclosed the timeline at Salesforce's Agentforce World Tour in Sydney, as reported by iTnews.
"We've been exploring Agentforce for about nine months now and we're probably about six weeks away from actually piloting something in production, which is quite exciting," Wells said.Telstra has kept the specific business use case under wraps for now, but confirmed it is working closely with Salesforce on its broader AI strategy.
The pilot sits within a broader architecture play. Through Telstra's joint venture with Accenture, which runs a hub in Mountain View in California's Silicon Valley, the company has been working to make its existing Agentforce capability interoperable with Salesforce Foundations, a free add-on module that unlocks Agentforce across the Salesforce cloud platform. Wells said the joint venture had allowed the company to accelerate that architectural groundwork so it could move quickly once the integration comes online.
For now, the goals for the pilot are deliberately conservative. Wells said the initial focus would be on building the operational framework around running AI agents rather than deploying them at scale. That means working out how to monitor agents, run them securely, and take them offline when needed. "Our early pilot is going to be about figuring out the operating model," she said, pointing to the non-functional, governance-centred processes that must be in place before any agent touches a real business workflow.
That caution reflects a broader internal warning. At Telstra's recent half-year results, group technology executive Kim Krogh Andersen was pointed about the risks of getting the AI foundation wrong. If the underlying architecture is not right, he argued, the running cost of AI would outstrip its benefits entirely. Chief financial officer Michael Ackland echoed that sentiment, saying Telstra is closely scrutinising each AI investment to ensure cost is justified by commercial return. Telstra has catalogued 380 potential AI use cases, but having a long list is very different from having a profitable one.
The trade implications for Australia are direct: Telstra is the country's dominant carrier and its choices on AI investment and workforce structure ripple across the domestic technology sector. Its Accenture joint venture, valued at around $700 million, is specifically designed to accelerate the company's data and AI roadmap, and it signals that large Australian enterprises are increasingly offshoring specialist AI capability rather than building it domestically.
That tension is sharpest when viewed alongside Telstra's workforce decisions. As reported by iTnews, approximately 650 roles face elimination in the current restructuring round: around 442 positions are being moved offshore to Infosys, drawn mostly from enterprise software divisions, with a further 209 roles to go from the Accenture joint venture itself. Telstra has already provided 75 percent of its staff with AI tools and trained close to 9,000 employees on the technology as part of its Connected Future 30 strategy, yet it acknowledges its workforce will be smaller by 2030.
Critics, including the Communications Workers Union, have argued that workforce reductions and AI investment are difficult to separate in practice, whatever the official explanation. The union has previously accused Telstra of using AI as a cover for cuts that erode service quality. Telstra has maintained that the current round of job changes is driven by structural resets in its enterprise division rather than AI displacement, though the Fair Work Commission will inevitably be watching how those transitions are managed.
The broader context is instructive. Salesforce launched its telco-specific Agentforce for Communications product just days ago, targeting the revenue plateau that has hit carriers globally after years of 5G infrastructure spending. According to PricewaterhouseCoopers, telco revenue growth is forecast to slow to 2.9 percent by 2029, and customer churn in some markets has reached 40 percent. For Australian carriers, the pressure to extract more value from existing infrastructure through automation is real and growing.
Telstra's approach, starting with architecture and governance before scaling to customer-facing agents, is the kind of methodical framework that separates durable AI programmes from expensive proof-of-concept exercises. Whether the commercial returns justify the investment in a company of Telstra's scale, complexity, and regulatory obligations is a question that only the pilot data will begin to answer. Reasonable observers can disagree about how fast to move; what Telstra cannot afford is to rush the foundation and inherit a cost structure that defeats the purpose of the exercise.