AT&T now runs nearly 500 agentic AI workflows in production

[Scott98]

A major telecoms company is using AI agents at a big scale. SiliconANGLE reports that AT&T runs nearly 500 agentic workflows in production, processing about 45 billion tokens a day and more than one million agent transactions a month

AT&T plans to triple the number of workflows next year. It says agents halved provisioning time for dedicated internet in buildings that already have fibre, and that its production AI investments are reportedly returning fivefold within a year

Agents only run on their own once their accuracy reaches the mid to high 90 percent range. Every workflow starts with human oversight

Luke Corbin of AT&T Business said the company stores the business logic in context or in memory. He described linking agents to mainframes as some of the most advanced technology talking to some of the oldest the company has, and the piece is a case study based on a talk at an AWS event

Requiring high accuracy before letting agents work alone is a sensible rule that other companies could copy. Halving provisioning time is the kind of practical result that justifies the investment. Connecting modern AI to old mainframes shows how much legacy technology big firms still rely on

Should companies require human oversight before letting AI agents work alone? And which business tasks suit agents best?

AT&T uses agentic AI to reinvent business processes from the ground up - SiliconANGLEsiliconangle.com