- AT&T's Ask AT&T AI platform is used by 100,000 employees and receives 10 million API calls each day
- Exec BJ Nab told Fierce the operator has created sandboxes to allow employees to tinker with agentic workflows to help boost adoption
- In the wake of OpenClaw's release, AT&T is also building an new "superagent"
Using AI is kind of like learning to drive a car – you need to figure out how to accelerate and steer before you decide where you want to go. So, it makes sense that one of the keys to AT&T’s strategy as LLMs give way to AI agents is a concept many of us are already familiar with: learner’s permits.
Transformation teams across the company now have access to domain-specific AI sandboxes that provide a drag and drop interface enabling them to build agentic workflows. The idea is to give employees a safe environment in which to learn how new agentic tools are built and operate – before asking them to pick a use case to implement.
“We’ve done a lot of work on the front end to give people hands-on experiences as opposed to this nebulous ‘I’ve got to create this killer use case but I don’t even know what I’m talking about yet because I don’t know how these systems work,’” AT&T’s AVP BJ Nab told Fierce. “We’ve solved that gap with that permit to learn.”
The learner's permit is a key tool for building trust and knowledge among employees and improving adoption of new AI capabilities in an industry that is slow to change. But it doesn’t mean AT&T is letting employees take their new AI pals out for an unsupervised spin in production. For that, they also need a permit to build outside the sandbox as well as a permit to operate, Nab said. You know, because safety first and all.
Nab, who is GenAI Product Lead for Ask AT&T and part of the operator’s Chief Data Office team, acknowledged that getting to this point has been a process. And the foundational building block to it all has been Ask AT&T, the GenAI interface now used by over 100,000 people across the company.
Getting the ball rolling
AT&T has been using what Nab characterized as “traditional AI” for years, but its first foray into generative AI came in June 2023 with the internal rollout of ChatGPT-based Ask AT&T. The tool is designed to be used across company domains, to help care representatives answer customer questions, to help engineers optimize network performance, to upgrade legacy code and to provide HR support.
Nab said Ask AT&T was trained on AT&T's internal knowledge base – a task made easier by the fact that it had launched a data center of excellence a decade earlier – and deployed in a secure Microsoft Azure environment. The operator also scored agreements with Microsoft and OpenAI specifying that the prompts and responses flowing through Ask AT&T could not be used to train ChatGPT, Nab said.
Nab said the rollout started slow; AT&T wanted to “ensure we knew the beast we were unleashing.” But it took less than six months to scale Ask AT&T to all of the 100,000 or so users it has today.
This staggering pace of adoption was accomplished by recruiting GenAI leads and ambassadors in each department.
“It was never a mandate,” Nab said. “We would enable them with the technology, but we basically lined up a set of ambassadors to help tell the story and make it relevant within their particular business units. And I think that helped a lot on the human productivity.”
The approach is different from that regional operator C Spire took with its AI NOC implementation, but there’s a similar throughline. AI implementation and adoption is smoother when employees are invested in the outcome rather than handed a tool and told to use it.
In 2024, AT&T rolled out internal course to help employees expand their AI knowledge. Nab said over 77% of management have taken at least one GenAI course and the average number of courses taken across the company is seven.
Today, Nab said most knowledge workers across the company are using Ask AT&T – whether they know it or not. He noted there are around 10 million API calls to the platform each day, but humans only account for around 2-3% of overall token demand. The rest of the calls come from internal and customer-facing systems. Put another way, there are a lot of machine-to-machine conversations already happening at AT&T.
Superagent
The focus now is preparing for the agentic AI wave that’s coming.
If 2022 and 2023 were the “big splash moment” for GenAI, the start of 2026 is shaping up to be the same for agentic with the release of OpenClaw, he said.
AT&T has already started work building what Nab described as a “superagent” inspired by the OpenClaw release but “safe for AT&T.” The key, he said, will be infusing it with AT&T context, policies, security, governance and quality control. And while the superagent is still undergoing internal testing and auditing, Nab noted the operator is already building new coursework to educate its employees.
“That’s our big push right now,” he concluded.