White mannequin robot riding a black children's bicycle with training wheels and glowing green tires through a dark hall, a blue-lit roller door and a bright studio light behind it (AI image)

Illustration

Companies3 min read

AT&T makes its AI agents earn their autonomy

AT&T has around 500 agentic AI workflows in production and plans to triple that next year. The scale is impressive, but the rule behind it matters more: no agent works alone on day one, and autonomy only comes once accuracy is proven.

Key facts

  1. According to a company talk, AT&T runs about 500 agentic AI workflows in production and plans to triple that number in 2027.
  2. Every production workflow starts under human oversight and only runs autonomously once accuracy reaches the mid-to-high 90 percent range.
  3. AT&T processes about 45 billion AI tokens a day and says production AI investments return fivefold within a year (company claim).

Manfred „Luigi“ LugmayrBy Manfred „Luigi“ Lugmayr

Researched and written with AI, to Luigi's brief, positions and editorial rules. Automatically checked against the sources. How I work →

You know the pattern. The new agent tool is set up, the demo looked great, and by Monday the agent is already emailing customers. Nobody ever decided when it was allowed to do that. AT&T does the exact opposite, at a scale most teams will never touch.

Luke Corbin, who runs digital transformation and AI at AT&T Business, walked through the program at an AWS event in Boston earlier this month. SiliconANGLE turned it into a case study.

What AT&T is actually running

  • About 500 agentic workflows are in production, and the plan is to triple that next year.
  • The company processes roughly 45 billion AI tokens a day; the agent platform handles more than a million transactions a month.
  • Corbin says production AI investments return fivefold within a year. That is a company claim, not an audited number.
  • In sales, agents check a customer's location and fiber availability before a rep spends time on the lead. Dedicated internet connections in buildings that already have fiber now get provisioned in half the time.

Autonomy is a promotion, not a feature

The line that stuck with me: every production workflow starts under human oversight. It only runs autonomously once its accuracy reaches the mid-to-high 90s and the risk profile is acceptable. After that, monitoring keeps checking for drift.

In other words, autonomy at AT&T is not a toggle in a settings menu. It is a promotion after probation. That is how I would treat a new hire: prove you can do the job, then you get signing authority. Human in the loop is not a permanent state here. It is a phase with a defined exit.

One more detail: employees can build their own personal productivity agents. Anything that materially touches customers or the business has to go through the internal Business Agent Foundry and pass security, privacy and compliance reviews. Two lanes, one clear line between them.

Fix the process first, then build the agent

The part I like most: the foundry does not just collect automation ideas. It maps the process first and cuts unnecessary steps, handoffs between departments and dependencies on legacy systems. Only then does an agent come in. AT&T runs a fairly small set of agents with centrally managed business logic instead of hundreds of specialist bots.

My rule of thumb fits right in: AI does not fix broken ways of working, it widens the cracks. Automate a chaotic approval process and you get a fast chaotic approval process.

The catch

These numbers come from a company talk. The case study gives no costs, no models and no real error rates. And a telco with its own agent team is not your five-person marketing department. The rule still transfers one to one, precisely because it costs nothing. For why rubber-stamping alone does not work, see my piece on the Wharton study.

What I would do now

  • Set a probation period: For every agent that touches customers (emails, ads, lead scoring), write down a threshold: how many reviewed cases, what hit rate, and who signs off on promotion.
  • Split the two lanes: Anyone can build personal helpers. Anything customer-facing needs an owner and a sign-off.
  • Cut before you automate: Map the process once and ask of every step whether it needs to exist. The agent only gets what survives.

Bottom lineAutonomy is not a setting. It is a promotion, and every agent has to earn it.

  • Share

Got AI news? Submit news →

FAQ

Do AT&T's AI agents work entirely without humans?

Not at first. According to AT&T, every production workflow starts under human oversight and only runs autonomously once accuracy reaches the mid-to-high 90 percent range. Monitoring continues after that.

Can any AT&T employee build their own agents?

For personal productivity, yes. Workflows that materially affect customers or the business go through the internal Business Agent Foundry and security, privacy and compliance reviews.

Source

SiliconANGLE: AT&T uses agentic AI to reinvent business processes from the ground up

Black Friday 2026

Black Friday deals on AI tools, straight to your inbox.

I collect the AI tool deals worth having and send them in the AI Briefing, along with the AI news that matters, every Wednesday and Friday. Five minutes to read.

  • Black Friday deals on AI tools, collected for you
  • The key AI news, briefly explained
  • Wednesday and Friday, unsubscribe in one click

In the AI Briefing right now

Image for the story: Perplexity and Runway: Black Friday brings credits, not discountsDealPerplexity and Runway: Black Friday brings credits, not discounts
Image for the story: AT&T makes its AI agents earn their autonomyNewsAT&T makes its AI agents earn their autonomy
Image for the story: North Korea's deepfake meetings: a face on the call proves nothing nowNewsNorth Korea's deepfake meetings: a face on the call proves nothing now
Blonde woman with glasses smiling as she scrolls on her phone in an office

Manfred “Luigi” LugmayrFrom Manfred “Luigi” Lugmayr: 30 years in marketing, seven of them as CMO at burgerme.