According to a report by Paul Gillin on SiliconANGLE, AT&T Corp. is expanding its use of artificial intelligence beyond improving employee productivity, increasingly turning to autonomous agents to redesign the way its business operations run. As of the time of the report, nearly 500 agentic workflows are already running in the company's production environment, and the telecommunications corporation plans to triple that number over the course of next year. In addition, the company now processes about 45 billion AI tokens daily across the enterprise, and its agentic AI platform handles more than 1 million transactions each month, with many processes already operating autonomously across a variety of tasks.
The Expansion of Autonomous Workflows at AT&T
According to Luke Corbin, assistant vice president of digital transformation and AI at AT&T Business, the company now processes about 45 billion AI tokens daily across the company. Speaking at an Amazon Web Services Inc. (AWS) event held in Boston earlier this month, Corbin noted that many of the nearly 500 agentic workflows already deployed in production are now operating entirely autonomously. These autonomous workflows handle a wide range of business activities, extending from customer provisioning to sales prospecting.
Corbin explained at the event that AT&T’s approach emphasizes rethinking entire business processes, rather than settling for isolated automation of single tasks. According to him, the company transitioned rapidly from the stage of experimenting with generative AI toward systematically identifying business processes that can be redesigned comprehensively around the capabilities offered by autonomous agents.
Operations of the Business Agent Foundry and Navigating Legacy Systems
At the core of this transformation stands a group named the Business Agent Foundry, established in early 2025 to work directly with the company's individual business units. Corbin noted that instead of reaching out to employees to solicit isolated automation ideas, the group maps business processes using industry frameworks. Through this mapping, the team identifies opportunities to eliminate redundant process steps, do away with organizational handoffs between different departments, and reduce dependencies on legacy application systems.
One of the group’s early initiatives involved automating interactions with AT&T’s mainframe computers. In the past, making changes to customer fiber installation orders required company employees to retrieve tickets from queues, manually update records in the system, and place the tickets back into the workflow. Agentic automation eliminated portions of that manual process, thereby reducing delays and accelerating service delivery times for customers. Addressing this milestone, Corbin remarked that it was kind of cool to see some of the company’s most advanced technologies interacting with some of the oldest technologies in its possession.
In another example presented by Corbin, the company succeeded in halving the time required to provision dedicated internet connections in buildings where fiber infrastructure had already been pre-installed.
A Centrally Managed Logic Architectural Approach
The technological architecture developed by AT&T departs from standard industry approaches, which tend to build large numbers of independent, specialized agents for each task. Instead, the company opted to maintain a relatively small set of agents that execute business workflows using centrally managed business logic. Corbin noted in this context that the company stores the business logic in context or in memory.
Within this framework, retrieval agents are responsible for identifying and retrieving the relevant knowledge needed to perform a task, while planning agents determine the required actions, which tools to use, and what data is necessary to successfully complete the given task.
Another operational area benefiting from this architecture is sales prospecting. Autonomous agents validate customer locations and verify fiber infrastructure availability in advance within that area, before sales representatives invest time and effort in pursuing and advancing deals with those prospective accounts.
Governance, Human Oversight, and Performance Metrics
Governance occupies a central role in AT&T's implementation model. Individual employees across the organization are permitted to independently build agents to boost their personal productivity; however, any workflow that could potentially affect customers must pass through the Business Agent Foundry. Within this framework, workflows must undergo security, privacy, and compliance reviews. Corbin explained that the company democratized agents for personal productivity, but restricted them so employees cannot independently deploy workflows that could touch customers or materially impact the business.
Furthermore, every workflow designated for production initially launches with integrated human oversight. Only after an agent's accuracy reaches the mid- to high 90% range, under an acceptable risk profile, are agents granted permission to operate autonomously. A continuous monitoring framework routinely assesses the performance of deployed agents and actively works to detect drift in their operational behavior over time.
Success metrics include labor savings, revenue generation, operating costs, and customer satisfaction. Corbin emphasized that the company's investments in production AI applications are achieving a fivefold return on investment within a single year.
Infrastructure Challenges, Legacy Heritage, and Cultural Shift
Corbin addressed the reality that AT&T is a 151-year-old enterprise that has utilized computers since they were first invented, meaning its operational foundation rests on numerous legacy processes and applications. According to him, the expenses associated with maintaining these legacy applications, training staff to operate them, and conducting routine maintenance are extraordinarily high. When a portion of these expenditures can be offset by simpler generative AI workflows, it creates a business case that readily justifies itself.
Nevertheless, scaling these capabilities enterprise-wide demands infrastructure that extends far beyond AI models alone. The company was required to consolidate fragmented data pools and provide dependable network connectivity across data centers, cloud infrastructure, and mobile field personnel. Corbin noted that the company has undergone numerous mergers and acquisitions throughout its operational history, a dynamic that directly impacts the complexity of the organization's data infrastructure.
To manage resistance from employees concerned that artificial intelligence might replace them, the company encouraged personnel to actively propose their own ideas for deploying agents. Another key factor that helped diminish workforce friction was the opportunity to eliminate inefficient business processes that staff disliked using. Corbin pointed out that there is a tremendous number of people throughout the enterprise who possess knowledge and passion regarding historical business processes, but who were fundamentally dissatisfied with them.
AT&T is currently shifting its operational focus from refining existing workflows toward leveraging agents to introduce new services, expand into new markets, and support emerging sales channels. Ultimately, Corbin believes that responsibility for AI-driven business transformation should be distributed across the entire organization, rather than remaining concentrated in a specialized team like the Business Agent Foundry. In his view, his team should not exist over the long term, as these transformational activities should eventually become entirely distributed across the enterprise.