AT&T runs nearly 500 agentic AI workflows and plans to triple them in 2027
AT&T has nearly 500 agentic AI workflows in production and plans to triple that number in 2027, a company executive said at an AWS event in Boston on November 4, in a disclosure reported November 14. The platform handles more than a million transactions a month, and AT&T processes about 45 billion AI tokens a day.
📊 Scorecard: our Bullish call on $T is scored from $22.18, the last close before it ran. Its first result comes after the next close.
The News
AT&T has nearly 500 agentic AI workflows in production and plans to triple that number in 2027, Luke Corbin, AT&T Business's AVP of digital transformation and AI, said at an AWS event in Boston on November 4. The disclosure became public through a report published November 14.
The scale of the operation is large by any measure. The platform handles more than 1 million transactions a month, and AT&T processes about 45 billion AI tokens daily. A token is the small chunk of text an AI model reads or writes, so 45 billion tokens a day means an enormous volume of AI work running through the company's systems around the clock.
The agents work on tasks from customer provisioning — setting up a new customer's service — to sales prospecting, finding and qualifying potential customers. They act autonomously, but only after earning that trust: an agent is allowed to work alone only once its accuracy reaches the mid-to-high 90% range within a risk profile. In plain terms, it has to be right roughly 95 times out of 100 before a human stops checking its work.
The work is run through a Business Agent Foundry that AT&T set up in early 2025. A foundry is a workshop where things are made — here, the agents themselves and the platform they run on. The disclosure was made at an AWS event, and the platform runs on Amazon's cloud. AT&T is separately a Microsoft Azure customer — its Ask AT&T agent platform is built on Azure — which is why Microsoft also appears as a read-through ticker on this story.
AT&T closed Friday at $22.18, flat on the day. Amazon closed at $266.66, down 0.29%, and Microsoft closed at $523.88, up 0.17%. All of the figures in this story are company-reported — AT&T's own account of its programme, not independently verified numbers.
Nearly 500 AI agents in production, 45 billion tokens a day — AT&T is running one of the largest agentic deployments in corporate America.
Why It Matters
AT&T's numbers matter because they show agentic AI moving from experiment to infrastructure. A workflow in production is one that real customers and real revenue depend on, and 500 of them — handling more than a million transactions a month — means agents are doing work that used to need people, at a scale where the savings and the risks are both material.
The accuracy gate is the detail that makes the story credible. Letting an agent act alone only once it is right roughly 95 times out of 100 is a risk-managed rollout, not a rush. It also explains why the programme can scale: each workflow graduates to autonomy on evidence, so tripling the count in 2027 is a matter of running the same process more times, not of taking a bigger leap.
For the cloud giants, the story is a reminder of where agentic AI actually runs. AT&T's agents sit on AWS and Azure, and 45 billion tokens a day is compute that somebody bills for. But the direct beneficiary is AT&T itself: a telecom operator using agents to provision customers faster and prospect for sales more cheaply is cutting its own costs, which is the oldest and most reliable AI payoff.
The date flag is worth stating plainly. The event was November 4; the disclosure only became public through the November 14 report. That does not make it stale — it makes it newly public — but readers should know the numbers describe the programme as of early November, not as of this weekend.
Analyst's Take
Our call is Bullish. Nearly 500 agentic workflows in production, more than a million transactions a month and 45 billion tokens a day is not a pilot programme — it is a company that has rebuilt core business processes around AI agents and is now scaling them threefold. The accuracy gate is the most encouraging detail: agents earn autonomy only at mid-to-high 90% accuracy, which is the discipline that separates a real deployment from a demo.
The read-through for the cloud providers is steady rather than spectacular. AT&T's platform runs on AWS and Azure, so every additional workflow is more cloud consumption for Amazon and Microsoft — but at their scale, one customer's agent programme is a small increment. That is why both are tagged as mixed here: the story is genuinely bullish for AT&T's operations, and only neutral context for its cloud suppliers.
The honest caveat is verification. Every number here is company-reported, from a single executive speaking at a vendor's event, and AT&T has not disclosed what the agents save or earn. Tripling to roughly 1,500 workflows in 2027 is a plan, not a result. But the scale already in production — 500 workflows, a million transactions a month — is hard to fake, and the direction is clear.
Reader poll: are you bullish or bearish on AT&T? Vote: Bullish, Bearish, or Too early.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is an agentic AI workflow?
An agentic workflow is a task handed to an AI agent — software that can plan, use tools and take actions on its own, rather than just answering a question. At AT&T these agents work on jobs from customer provisioning to sales prospecting, acting autonomously once they are trusted to do so.
How big is AT&T's deployment?
Nearly 500 agentic workflows in production, handling more than 1 million transactions a month, with AT&T processing about 45 billion AI tokens daily. A token is the small chunk of text an AI model reads or writes — 45 billion a day is an enormous volume of AI work.
When did AT&T say this?
At an AWS event in Boston on November 4, where Luke Corbin, AT&T Business's AVP of digital transformation and AI, described the programme. The disclosure became public through a report published November 14, which is why it is news now.
How does AT&T decide when an agent can act alone?
Only after its accuracy reaches the mid-to-high 90% range within a risk profile, according to Corbin. In plain terms: the agent has to be right roughly 95 times out of 100 on a task before it is allowed to do that task without a human checking it.
What is the Business Agent Foundry?
It is the internal team AT&T set up in early 2025 to build and run these agents. A foundry is a workshop where things are made — here, the agents and the platform they run on.
Why are Amazon and Microsoft listed as tickers?
AT&T is the story, but the disclosure came at an Amazon Web Services event, and AT&T is separately a Microsoft Azure customer for its Ask AT&T agent platform. Amazon and Microsoft are read-through tickers: they supply the cloud infrastructure the agents run on. AT&T closed Friday at $22.18, flat on the day; Amazon closed at $266.66, down 0.29%; Microsoft closed at $523.88, up 0.17%.
👁
Source: SiliconANGLE