AI News

AT&T Puts Nearly 500 Agentic Workflows into Production, Plans to Triple That Number Next Year

21ic Electronic NetworkUpdated Nov 16 21:07

AI summary

AT&T already has nearly 500 agentic workflows in production and plans to triple that number next year. Its agentic AI platform processes about 45 billion AI tokens per day and over 1 million transactions per month, and many workflows can autonomously complete tasks ranging from customer onboarding to sales prospecting. AT&T adopts an architecture of a streamlined set of Agents combined with centralized business logic; workflows that affect customers must pass security, privacy, and compliance reviews, and are allowed to run autonomously only after accuracy reaches above 90% and the risk is deemed acceptable.

Why read this

AI score 61

AT&T disclosed the architectural trade-offs and governance thresholds after putting nearly 500 agentic workflows into production, which can serve as a reference for enterprise-level Agent deployment.

Topics

Tags

Body · AI translation

AT&T is reportedly pushing agentic AI from an auxiliary tool for boosting employee efficiency to an underlying reconstruction of business processes. Currently, the company has nearly 500 agentic workflows in production and plans to triple that number next year.

Luke Corbin, Vice President of Business Digital Transformation and AI Assistants at AT&T Business, revealed at an Amazon Web Services event in Boston that the company processes about 45 billion AI tokens per day, its agentic AI platform handles more than 1 million transactions per month, and many workflows can already autonomously complete tasks ranging from customer onboarding to sales prospecting.

From "Automating Tasks" to "Redesigning Processes"

The report says AT&T's strategy is not simply to automate individual tasks, but to rethink entire business processes. The "Business Agent Foundry," established in early 2025, works with various business units, using industry frameworks to map processes and identify opportunities to eliminate redundant steps, organizational handoffs, and legacy application dependencies.

One early case was automating interactions with AT&T's mainframes. In the past, when a customer's fiber installation order changed, employees had to retrieve the work order from a queue, manually update the record, and then return it to the workflow; agentic automation reduced this manual step, lowering delays and accelerating service delivery. Corbin said it is meaningful to make advanced technology "talk" to the company's oldest technology. In another case, AT&T halved the time required to activate dedicated internet connections for buildings where fiber infrastructure is already in place.

Streamlining Agent Architecture and Centralized Governance

In terms of architecture, AT&T did not take the route of "a large number of independent specialized Agents," but instead maintains a relatively streamlined portfolio of Agents and executes workflows through centrally managed business logic. Corbin explained that business logic is stored in context or memory, a retrieval Agent is responsible for identifying applicable knowledge, and a planning Agent determines the actions, tools, and data needed to complete a task.

Governance runs throughout the process. Employees can create personal productivity Agents on their own, but workflows that affect customers must go through the "Business Agent Foundry" and undergo security, privacy, and compliance reviews. Every production workflow initially has human oversight, and only after accuracy reaches a medium-to-high level of above 90% and risk is acceptable are Agents allowed to run autonomously. Continuous monitoring evaluates performance and detects drift. Success metrics include labor savings, revenue generation, operating costs, and customer satisfaction. Corbin said productive AI investments achieved a fivefold return within one year.

Legacy Systems Become the Business Case, with a Future Toward Decentralization

Corbin said AT&T is a 151-year-old company built on a large number of legacy processes and applications, and maintaining these applications and training employees is costly. When simpler generative workflows can offset part of those costs, the business case stands on its own. Sales prospecting is also one of the benefiting scenarios: Agents identify potential customers by verifying customer location and checking fiber availability, and representatives then invest time in follow-up.

Scaling these capabilities also requires integrating fragmented data. Corbin acknowledged that the company has gone through many mergers and acquisitions, its data infrastructure is complex, and it needs to provide reliable connectivity among data centers, cloud infrastructure, and mobile workers. Encouraging employees to proactively propose ideas for Agent applications helps reduce resistance to AI replacing jobs. Corbin believes that in the long run, responsibility for AI-driven business transformation should be distributed across the entire organization rather than concentrated in a dedicated team: "My team should not exist forever; what we do should be fully decentralized."

Source

21ic Electronic Network