Two of the world's largest telecommunications carriers announced significant steps this week in their race to deploy autonomous AI agents, one embedding voice AI directly inside phone calls across Germany, the other weeks away from a production pilot in Australia. The announcements, made at opposite ends of the globe, reveal how rapidly the industry is moving from AI experimentation to operational deployment — and how unresolved the questions of cost, governance, and accountability remain.
At Mobile World Congress in Barcelona, Deutsche Telekom unveiled what it describes as a world first: the Magenta AI Call Assistant, built in partnership with voice AI company ElevenLabs. The assistant is a network-embedded solution bringing real-time AI capabilities, including live translation and administrative support, to everyday phone calls on any device, without requiring additional apps. The technical integration at the network level was carried out by Radisys. The assistant will roll out to Deutsche Telekom customers in Germany later this year, with support for up to 50 languages planned over the next 12 months.
The mechanics are straightforward by design. The assistant activates during a call with the words "Hey Magenta"; without activation, no conversation content is stored or analysed. Customers must opt in before use, and all call participants are notified when the assistant is active. That consent architecture matters, given the sensitivity of live call data and Europe's strict data protection standards. Deutsche Telekom says voice recordings are not saved and fully comply with EU data protection laws.
The capabilities extend well beyond simple transcription. The assistant delivers real-time capabilities such as live translation, automatic call summaries, and contextual question answering, with future enhancements expected to allow the AI to book restaurants, schedule medical appointments, arrange travel, and complete forms during a conversation. Deutsche Telekom board member Abdu Mudesir framed the ambition plainly: "We are breaking new ground with strong partners like ElevenLabs. With our Magenta AI Call Assistant, we are the first in the world to offer such AI functions directly over the network. We are removing barriers. No apps, no special devices, no technical complexity."
The partnership has deeper roots than a single product launch. Earlier this year, ElevenLabs deployed AI voice agents to enhance Deutsche Telekom's customer service operations, and Deutsche Telekom also participated in ElevenLabs' Series C funding round in January 2025. That financial relationship gives Deutsche Telekom a stake in ElevenLabs' success and signals a longer strategic alignment, not simply a vendor arrangement.
Meanwhile, in Australia, Telstra is moving on a parallel, if less conspicuous, track. As reported by iTnews, the carrier is only weeks away from launching an agentic AI pilot to be led out of its customer sales and commerce engineering group, as it continues to use the technology to chase cost out of its operation. The pilot runs on Salesforce Agentforce, the enterprise AI agent platform, and stems from nine months of internal exploration. Commerce engineering group chief Marcella Wells was candid about the conservative scope: "Our early pilot is going to be about figuring out the operating model, working at how we observe these agents, run them securely, take them offline — all of those non-functional processes around running Agentforce."
That measured approach reflects a tension running through Telstra's AI strategy more broadly. At its recent half-year results presentation, Telstra said it had identified 380 ways it could use AI, but its chief financial officer Michael Ackland warned that the carrier was closely examining the technology to ensure its cost justified its commercial return. Group technology executive Kim Krogh Andersen put the risk plainly: if Telstra fails to get its foundational AI architecture right, "we will actually see the run cost of AI outperform the benefits of AI." Telstra has given AI a central role in its Connected Future 30 strategy, providing 75 per cent of staff with AI tools and training nearly 9,000 of them on the technology.
The workforce dimension of this transformation demands honest scrutiny. At a time when AI is creating uncertainty about job security globally, Telstra has moved to cut back its workforce, with around 650 roles facing elimination. Some 442 roles are going offshore to Infosys, the majority from its enterprise software divisions, and a further 209 roles face cuts from its joint venture with Accenture. The timing is not incidental: the same AI platform being piloted to improve efficiency is generating the business case for workforce reduction. Unions and workers have legitimate cause to ask what "efficiency" means for those whose jobs disappear in its wake.
Independent researchers have also raised questions about the quality of AI voice systems. Academics studying synthetic voice technology have pointed to accent bias in models from companies including ElevenLabs, with concerns that voice AI can struggle to represent or understand regional accents, particularly for non-native English speakers. "I'm a big proponent of purpose-built AI systems," one researcher told Wired. "This seems too general to just unleash on a population without safeguards." Deutsche Telekom's opt-in model addresses some of these concerns, though it does not resolve the underlying bias questions.
The broader competitive picture for telcos is challenging regardless of AI. Industry data suggests the sector faces a structural revenue problem: years of heavy investment in 5G infrastructure have barely preserved existing revenue bases, with average revenue per user continuing to decline. AI agents are being positioned as a path to cost discipline and new service differentiation simultaneously. Whether that promise survives contact with operational reality at scale remains, for now, an open question.
What is clear is that the two models on display this week represent different bets on where AI value accrues in telecommunications. Deutsche Telekom is embedding AI as a consumer-facing product, creating a service layer inside the network itself. Telstra, for now, is treating AI as an internal productivity tool, focused on getting governance and architecture right before any public-facing deployment. Both approaches carry risk; both have genuine merit. The telco that masters the governance challenge without sacrificing the commercial opportunity will have found something genuinely valuable.