AT&T’s John Miller on Transforming a Legacy Telco to Deepen Customer Connection

In this episode, Markus sits down with John Miller, VP of Consumer and Retail Solutions at AT&T, to explore how one of the world’s largest telecom companies is using AI agents and LLMs to enhance service delivery and modernize operations. John explains how his team deployed the very first GPT-style interface implemented across AT&T and how it is empowering employees throughout the organization.He also shares AT&T’s strategy for evolving from traditional workflow logic to asynchronous, agent-driven interactions that better mirror customer behavior. From managing petabytes of unstructured data to wrapping legacy mainframes with AI interfaces, John provides actionable insights for any enterprise looking to embrace generative AI.Timestamps00:00 Episode Start03:00 A broad overview of how AT&T is leveraging AI06:15 The best solution for scale 09:30 Reshaping workflows to better serve customers15:20 How to design your org's AI agents20:55 Different models for different tasks22:50 Empathizing with your customers 29:35 Tackling legacy systems one step at a time35:55 The data problem44:15 Advice for others beginning their journeyEpisode Key TakeawaysLegacy systems don’t have to be a blocker: AT&T is proving that even decades-old mainframes can be modernized by “wrapping” them with AI agents. This approach enables gradual transformation without disrupting end users or retraining staff.Perfect data isn’t a prerequisite: Instead of waiting to clean every dataset, AT&T uses AI to surface and correct data quality issues in real-time. This lets them move fast while improving data fidelity through use.Empathy matters in AI adoption: User trust and satisfaction rise when AI interactions mimic natural human communication. AT&T saw measurable improvements by designing agents that pause, clarify, and respond with warmth.AI success is iterative, not linear: AT&T's most effective agents didn’t launch fully formed. They evolved through constant prompt tuning, testing, and feedback. Treat agents like teammates who learn and improve over time.Top Quotes“For companies trying to wait until they have perfect data, you’ll never have perfect data. So it’s better to just start. Even with imperfect data, you can begin to see meaningful trends and value, especially when your AI tools are designed to adapt and refine based on real-time feedback.”“We’re wrapping legacy systems like a python [by] slowly squeezing out functionality until they’re obsolete. This approach lets us modernize without disruption, giving employees a seamless experience while we replace backend systems incrementally. Over time, those old systems just quietly phase out.”“AI lets you break the traditional workflow logic. Now we meet customers wherever they are in their journey, instead of forcing them through a rigid sequence. It’s a shift from thinking about linear steps to enabling outcomes through flexible, asynchronous interactions.” Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.