Who leads AI at AT&T?

AI at AT&T is led by Jeremy Legg (Chief Technology Officer, AT&T Services); Andy Markus (Senior Vice President and Chief Data and AI Officer); Mark Austin (Vice President, Data Science and AI). Ownership of AI on the org chart: Andy Markus, SVP and Chief Data and AI Officer; reporting line not published.

What AI platform does AT&T use?

Ask AT&T, the generative AI platform AT&T's Chief Data Office launched in Jun 2023, now bundles Ask Docs, Ask Data and the Ask AT&T Workflows agent builder, and carried about 45 billion tokens a day in Jul 2026.

Which AI models does AT&T use?

AT&T sends each request through its own AI Gateway to OpenAI or Anthropic models or to open-weight models such as Nvidia Nemotron, Meta Llama and Google Gemma, which handled about 40% of employee requests in Aug 2026.

How many AT&T employees use AI?

AT&T rolled Ask AT&T Workflows out to more than 100,000 employees, and more than half of them say they use it every day (Andy Markus, Feb 2026); AT&T had about 133,030 staff at the end of 2025.

What return has AT&T reported from AI?

AT&T sits at Piloting on an officially reported company-wide rollout: Chief Data and AI Officer Andy Markus said in Feb 2026 that the Ask AT&T Workflows agent builder had reached more than 100,000 employees, more than half of whom say they use it daily, and his Chief Data Office builds and owns Ask AT&T. AT&T has reported no business return it credits to AI in dollars, cost, headcount or revenue. The return multiples Markus has given are unbased executive estimates that conflict: about 2 times on generative AI investment to CIO.com in Jan 2026 and a 5-fold in-year free cash flow return at Mobile World Congress on Mar 3, 2026, neither with a dollar base or method and neither in a filing or earnings call, where AI is one unquantified lever in a $4 billion annual cost-savings target for the end of 2028. The roughly $150 million annual customer care saving appears only in a Google Cloud case study, not in AT&T's own releases. A dollar figure AT&T credits to AI, or reported returns in 2 functions, would move AT&T to Scaling.

Verified Nov 5, 2026

AT&T

Piloting

Telecom
133,030 staff
Ask AT&T, built in-house
100,000+ on Ask AT&T Workflows
~45B tokens a day
OpenAI, Anthropic, open-weight

AT&T deploys AI through Ask AT&T, an employee generative AI platform its Chief Data Office launched in 2023 in a dedicated Microsoft Azure tenant, which now routes about 45 billion tokens a day between OpenAI, Anthropic and open-weight models (Jul 2026). Chief Data and AI Officer Andy Markus said in Feb 2026 that the Ask AT&T Workflows agent builder had reached more than 100,000 employees, but AT&T has published no dollar return it credits to AI.

about.att.com

Their approach to AI

The platform: Ask AT&T, the generative AI platform AT&T's Chief Data Office launched in Jun 2023, now bundles Ask Docs, Ask Data and the Ask AT&T Workflows agent builder, and carried about 45 billion tokens a day in Jul 2026.
The models: AT&T sends each request through its own AI Gateway to OpenAI or Anthropic models or to open-weight models such as Nvidia Nemotron, Meta Llama and Google Gemma, which handled about 40% of employee requests in Aug 2026.
The reach: AT&T rolled Ask AT&T Workflows out to more than 100,000 employees, and more than half of them say they use it every day (Andy Markus, Feb 2026); AT&T had about 133,030 staff at the end of 2025.
The catch: AT&T's AI returns are ratios Andy Markus has given in interviews and at conferences, about 2 times in Jan 2026 and 5 times in free cash flow in Mar 2026, with no dollar base; AT&T's filings and earnings calls fold AI into a $4 billion cost program without giving AI's share.

Snapshot

Employees: 133,030; Dec 31, 2025, about 43% represented by unions (2025 Form 10-K)
Technology budget: Not disclosed; Capital investment of $22.0B in 2025, guided at $23B to $24B a year for 2026 to 2028, mostly network
AI/ML headcount: Not disclosed; AT&T publishes no count of its data scientists or AI engineers
Who owns AI: Andy Markus, Chief Data and AI Officer; Senior vice president; joined AT&T from WarnerMedia in 2020
Platform: Ask AT&T, built in-house; Launched Jun 2023; Ask AT&T Workflows agent builder added Nov 2025
Models: OpenAI, Anthropic, open-weight; Nemotron, Llama and Gemma, plus AT&T's own OTel telco models (2026)
Reach: 100,000+ employees; Ask AT&T Workflows, Feb 2026
Use cases: Care, fraud, network, coding, legal; Andy Markus said in Jan 2026 that hundreds of business cases await prioritization

Key findings

AT&T has published no dollar return it credits to AI: Andy Markus, its Chief Data and AI Officer, cited about 2 times on AI investment to CIO.com in Jan 2026 and a 5-fold in-year free cash flow return at a conference on Mar 3, 2026, conflicting multiples with no dollar base, while AT&T's earnings calls count AI as one unquantified lever in a $4 billion annual cost-savings target for the end of 2028.
AT&T rolled the Ask AT&T Workflows agent builder out to more than 100,000 employees by Feb 2026, and Ask AT&T's volume grew from about 1 billion tokens a day in Sep 2024 to about 27 billion in Feb 2026 and about 45 billion in Jul 2026.
AT&T routes employee AI requests through its own AI Gateway, which Andy Markus said in Jul 2026 cuts AI costs by as much as 90% and already saves millions, and Mark Austin told The Information in Aug 2026 that about 40% of employee requests now go to open-weight models, with a goal of 60% to 70%, to hold spending on Anthropic and OpenAI flat.
AT&T widened its AI vendor base in 2026: it renewed and doubled its work with Adaptive ML on fine-tuned fraud and call-summary models (Jun), H2O.ai's Super Agent inside Ask AT&T (Jul), OpenAI as design partner for its in-house legal team LegalEdge (Oct) and an 8-year agreement with Amdocs to modernize business and customer operations with AI (Nov 3, 2026).
AT&T had 133,030 employees at the end of 2025, and CTO Jeremy Legg told Wired, as reported in Sep 2026, that AT&T will not have the same headcount in 5 years as it automates internal work and retires copper; AT&T disputed a reported internal target of about 85,000 employees by 2030 and has not said how many job cuts it credits to AI.

Where AT&T sits

Stage: Piloting
AT&T sits at Piloting on an officially reported company-wide rollout: Chief Data and AI Officer Andy Markus said in Feb 2026 that the Ask AT&T Workflows agent builder had reached more than 100,000 employees, more than half of whom say they use it daily, and his Chief Data Office builds and owns Ask AT&T. AT&T has reported no business return it credits to AI in dollars, cost, headcount or revenue. The return multiples Markus has given are unbased executive estimates that conflict: about 2 times on generative AI investment to CIO.com in Jan 2026 and a 5-fold in-year free cash flow return at Mobile World Congress on Mar 3, 2026, neither with a dollar base or method and neither in a filing or earnings call, where AI is one unquantified lever in a $4 billion annual cost-savings target for the end of 2028. The roughly $150 million annual customer care saving appears only in a Google Cloud case study, not in AT&T's own releases. A dollar figure AT&T credits to AI, or reported returns in 2 functions, would move AT&T to Scaling.
How the stages are defined
Ownership: Andy Markus, SVP and Chief Data and AI Officer; reporting line not published
Platform: Hybrid: Ask AT&T built in-house; models and agent tools bought
Models: OpenAI, Anthropic and open-weight models behind an AI Gateway
Hosting: Dedicated Microsoft Azure tenant for Ask AT&T; Google Cloud for care agents; open-model serving location not published
Reach: 100,000+ employees on Ask AT&T Workflows (Feb 2026)

How they deploy

Build the platform, buy the models

AT&T built Ask AT&T in-house in 2023 on OpenAI's ChatGPT functionality, designed it to work with other generative AI tools, and has since added its own Ask Docs, Ask Data and Ask AT&T Workflows tools on top. Andy Markus said in Feb 2026 that AT&T relies on models that are interchangeable and selectable, never rebuilds a commodity, and will retire homegrown tools when off-the-shelf options mature.

Small models for most of the work

Andy Markus said in Feb 2026 that the future of agentic AI is many small language models, and AT&T's Ask AT&T Workflows uses a large model as a master agent directing smaller, fine-tuned models that make the decisions in each step. Markus wrote in Jul 2026 that only a small share of AT&T's tasks need the newest, most expensive models, and AT&T is training open telco models, including OTel 2.0, to handle telecom work at lower cost.

Production, not pilots

AT&T runs every proposed generative and agentic AI use case through a Generative AI Transformation Office, and each one gets a formal business case tied to the CFO's office, Andy Markus told CIO.com in Jan 2026; Markus said he looks for a free cash flow benefit in most cases, with exceptions for direct customer benefit. Markus said at Mobile World Congress in Mar 2026 that AT&T skips extended pilots and deploys AI directly into production in its network, back office and customer service.

Partners for agents and modernization

AT&T pairs its own platform with vendor agent tools: H2O.ai's Super Agent orchestrates agents inside Ask AT&T (Jul 2026), Adaptive ML's engineers help AT&T fine-tune reasoning models (Jun 2026), and Amdocs and AT&T will work as one delivery team under an 8-year agreement to modernize business and customer operations with AI (Nov 2026). CTO Jeremy Legg said in Nov 2026 that AI delivers the greatest value when grounded in the business, technology and people it supports.

AI as a network business

AT&T's leadership treats AI first as a source of network demand: John Stankey said on the Jul 2026 earnings call that agentic AI is reshaping network traffic, and COO Jeff McElfresh told Fortune in Mar 2026 that AT&T wants to be the highway for commerce and AI workloads under a $250 billion investment commitment through 2030. AT&T also sells AI networking services to businesses, such as AWS Interconnect last mile, which Shawn Hakl, head of product at AT&T Business, announced in Mar 2026.

Deployment model: AT&T runs AI from a central Chief Data Office under Andy Markus, which builds Ask AT&T, its data platforms and AI governance, with teams across the business building their own agents in Ask AT&T Workflows through no-code and pro-code paths. The first production Workflows tool was built jointly by AT&T Business and the technology development teams (Nov 2025). Use cases reach production through the Generative AI Transformation Office and business cases tied to the CFO's office, and Mark Austin, vice president of data science and AI, oversees the AI tools AT&T's employees use.
Model providers: AT&T launched Ask AT&T in 2023 with OpenAI's ChatGPT functionality, and Mark Austin told The Information in Aug 2026 that AT&T employees use models from Anthropic and OpenAI, with about 40% of requests routed through LiteLLM model routers to open-weight models such as Nvidia's Nemotron, Meta's Llama and Google's Gemma; AT&T does not use Chinese open models from DeepSeek or Moonshot. AT&T built its agent stack on LangChain, fine-tunes models with retrieval-augmented generation, and uses Microsoft's search for its vector store (Feb 2026). AT&T released OTel 2.0, a post-trained version of Google's Gemma 4 31B-IT, with the GSMA in Jul 2026, and its customer care agents run on Google's Gemini Enterprise for Customer Experience.
Data and infrastructure: AT&T runs Ask AT&T in an AT&T-dedicated Microsoft Azure tenant that it said in 2023 was pressure tested for data leakage, built Ask AT&T Workflows with Azure (2026), and trained OTel 2.0 on AMD GPUs through Microsoft Foundry Managed Compute, processing more than 1 trillion tokens for training. AT&T's customer care agents run on Google Cloud, with BigQuery routing calls by intent. Under a Feb 2026 agreement, AT&T is moving workloads from several on-premises systems to AWS Outposts, AWS's managed hybrid cloud, using Amazon Q Developer and other AWS agentic services. AT&T has not said where it serves its open-weight models; Andy Markus wrote in Jul 2026 that the OTel models can run on premises or in the cloud. Markus said in 2022 that AT&T held more than 543.7 petabytes of data and had standardized on cloud tools including H2O, Databricks and Snowflake.

Platforms and internal tools

Ask AT&TJun 2023
Ask AT&T is AT&T's internal generative AI platform, used to summarize calls and documents, write code, translate documentation, answer HR questions and query databases. AT&T's Chief Data Office built Ask AT&T in a dedicated Microsoft Azure tenant on OpenAI's ChatGPT functionality, designed it to work with other models, and routes its traffic through an AI Gateway; its volume grew from about 1 billion tokens a day in Sep 2024 to about 45 billion in Jul 2026.
Users
100,000+ employees with Ask AT&T Workflows (Feb 2026)
Models
OpenAI, Anthropic and open-weight models
Source
Ask AT&T WorkflowsNov 2025
Ask AT&T Workflows is a drag-and-drop agent builder inside Ask AT&T, built with Microsoft Azure, with a no-code path and a pro-code path in Python. Ask AT&T Workflows agents draw on AT&T's own tools for document processing, natural language to SQL and image analysis, run under human checkpoints and log every action, and Andy Markus said active users report productivity gains as high as 90% (Feb 2026).
Users
100,000+ employees; more than half say they use it daily (Feb 2026)
Source
Ask Docs and Ask DataBy Sep 2024
Ask Docs and Ask Data are AT&T's proprietary generative AI tools that hold company information for Ask AT&T and its agents; Ask Data turns plain English questions into SQL, ranked first on the BIRD leaderboard in Sep 2024, and its relational knowledge graph version topped the Spider 2.0 text-to-SQL leaderboard by Nov 2025.
Source
AI GatewayBy Jul 2026
AT&T's AI Gateway is a proprietary cache-aware router that weighs speed, cost and expected quality at each turn and sends every prompt to the most cost-effective model, switching models partway through multi-turn sessions when needed. Andy Markus wrote in Jul 2026 that the AI Gateway reduces AI costs by as much as 90% and already saves millions.
Source
OTel open telco modelsOTel 2.0, Jul 2026
OTel 2.0 is an open model AT&T post-trained from Google's Gemma 4 31B-IT on 400 billion telecom tokens selected from more than 1 trillion, with data from the GSMA and help from Red Hat, Dell, Microsoft Azure and AMD; the GSMA said OTel 2.0 topped the Open Telco AI leaderboard in Jul 2026.
Models
Gemma 4 31B-IT, post-trained
Source
AT&T Geo ModelerLive nationwide by Oct 2025
AT&T Geo Modeler is a generative AI system that uses synthetic data and a Network Foundation Model to simulate radio coverage, and it automatically adjusts nearby cell sites, for example by changing antenna angles, when a tower goes down in a storm, with employees monitoring its recommendations.
Source

Vendors and partners

Microsoft2023
Azure tenant behind Ask AT&T; Azure for Ask AT&T Workflows; Microsoft Foundry for model training · AT&T built Ask AT&T with Microsoft in 2023 and was one of the first companies to use Microsoft Foundry Managed Compute at scale to train its OTel models on AMD hardware (Jul 2026).
OpenAI2023
Model provider in Ask AT&T; design partner for AT&T LegalEdge · The first version of Ask AT&T used OpenAI's ChatGPT functionality, and AT&T Legal became a flagship in-house legal design partner for OpenAI on Oct 8, 2026.
Anthropic
Model provider for employee AI requests · Mark Austin said in Aug 2026 that AT&T aims to keep employee spending on Anthropic and OpenAI models flat by routing more work to open-weight models, according to The Information as reported by PYMNTS.
Google Cloud
Gemini Enterprise for Customer Experience and BigQuery for customer care; Gemma as the base of OTel 2.0 · Google Cloud's case study quotes Mark Austin crediting the care agents with about $150 million in annual savings; AT&T has not published that figure itself.
H2O.ai2021
Feature store partner; H2O AI Super Agent orchestrates agents inside Ask AT&T · H2O.ai said on Jul 14, 2026 that AT&T is using its Super Agent across customer experience, fraud prevention, field operations and enterprise research.
Adaptive ML2025
Reinforcement learning platform and embedded engineers for fine-tuned reasoning models · Adaptive ML said on Jun 10, 2026 that AT&T renewed after a year in production and doubled its software capacity in GPU nodes.
AmdocsNov 2026
8-year agreement pairing Amdocs aOS with AT&T's AI-enabled delivery framework
Amazon Web ServicesFeb 2026 (expanded)
AWS Outposts for migrated workloads; Amazon Q Developer and AWS agentic services for the migration
AMD and the GSMA2026
AMD GPUs and GSMA telecom data for the OTel open telco models

Leadership

AI, data and platform leaders

Jeremy Legg

Chief Technology Officer, AT&T Services

Jeremy Legg has been AT&T's CTO since May 2022. Legg announced the Nov 2026 Amdocs agreement to modernize AT&T's business and customer operations with AI, and told Wired, as reported in Sep 2026, that AT&T will not have the same headcount in 5 years.

Andy Markus

Senior Vice President and Chief Data and AI Officer

Andy Markus joined AT&T in 2020 after serving as Senior Vice President, Data and Consumer Analytics at WarnerMedia, and earlier worked at Time Inc; he holds statistics degrees from The University of Alabama. Markus leads AT&T's data platforms, automation, business intelligence, traditional and generative AI and data and AI governance, built Ask AT&T and gives the company's AI return figures in interviews and at conferences.

Mark Austin

Vice President, Data Science and AI

Mark Austin leads data science in AT&T's Chief Data Office and oversees the AI tools AT&T's employees use. Austin is the AT&T voice in the Google Cloud, Adaptive ML and open-model routing stories of 2026, including the roughly 40% of employee requests now sent to open-weight models.

Executives

John Stankey

Chairman, Chief Executive Officer and President

John Stankey joined AT&T in 1985, became CEO in Jul 2020 after serving as President and COO and running WarnerMedia, and became chairman in Feb 2025. Stankey said on the Jan 2026 earnings call that AT&T plans to speed up efficiency gains across the company by using AI and moving more customer transactions to digital.

Jeff McElfresh

Chief Operating Officer

Jeff McElfresh has been AT&T's COO since May 2022 and has spent about 30 years at the company. McElfresh told Fortune in Mar 2026 that AT&T uses AI to support technicians on home visits, help engineers place cell sites and plan fiber deployment, and that these tools are improving productivity and capital allocation.

Pascal Desroches

Senior Executive Vice President and Chief Financial Officer, retiring Dec 31, 2026

Pascal Desroches has been AT&T's CFO since Apr 2021 and announced on Jun 11, 2026 that he will retire at the end of 2026. Desroches said on the Apr 2026 earnings call that AI enablement is one of the levers behind AT&T's target of $4 billion in annual cost savings by the end of 2028.

Jennifer Biry

Deputy Chief Financial Officer; Chief Financial Officer from Jan 1, 2027

Jennifer Biry became AT&T's Deputy CFO on Jul 6, 2026 and takes over as Senior Executive Vice President and CFO on Jan 1, 2027, according to AT&T's Jun 2026 Form 8-K. Biry inherits the CFO's office to which AT&T ties every generative AI business case.

David McAtee

Senior Executive Vice President and General Counsel

David McAtee has been AT&T's General Counsel since Oct 2015. McAtee announced in Oct 2026 that AT&T Legal is working with OpenAI to build an AI-enabled legal department on top of LegalEdge, AT&T's AI-first in-house legal team, and said legal judgment remains human.

AI org map

Central Chief Data Office that builds the platform; agents built by teams across the business

AT&T centralizes data and AI in its Chief Data Office, led by Senior Vice President and Chief Data and AI Officer Andy Markus, which builds Ask AT&T and sets AI governance, while teams across AT&T build agents on the platform. Every generative and agentic AI use case passes through AT&T's Generative AI Transformation Office and needs a business case tied to the CFO's office (Jan 2026). AT&T has not published to whom Markus reports; Markus and CTO Jeremy Legg both work for AT&T Services, Inc.

AT&T's AI work sits mainly in the Chief Data Office under Andy Markus, which runs Ask AT&T, the AI Gateway, Ask Docs and Ask Data and AT&T's data science teams, with business and technology teams across AT&T building their own agents in Ask AT&T Workflows. AT&T does not publish how many data scientists or AI engineers it employs. AT&T Legal runs its own AI-first unit, LegalEdge, which had about 30 people in May 2026 according to eDiscovery Today.

Reporting lines

Chairman and CEO
John Stankey
- Chief Operating Officer
  Jeff McElfresh
  Executive officer since May 2022; reporting line inferred from the 2025 Form 10-K list of executive officers
- Finance
  Pascal Desroches, CFO; Jennifer Biry from Jan 1, 2027
  Generative AI business cases are tied to the CFO's office
- Chief Technology Office, AT&T Services
  Jeremy Legg, CTO
  Executive officer since May 2022; reporting line inferred, not published
  - Chief Data Office
    Andy Markus, Chief Data and AI Officer
    Inferred: AT&T has not published Markus's reporting line; both offices sit in AT&T Services, Inc.
    - Data Science and AI
      Mark Austin, VP
      Oversees the AI tools AT&T's employees use
- Legal Department
  David McAtee, General Counsel
  - AT&T LegalEdge
    AI-first in-house legal team launched Jan 2026; about 30 people (May 2026)

AT&T publishes no org chart for its AI work. The 2025 Form 10-K lists John Stankey, Jeff McElfresh, Pascal Desroches, Jeremy Legg and David McAtee among its executive officers but not Andy Markus, so the line from Markus to Legg is an inference. AT&T has not said where the Generative AI Transformation Office sits.

Named people

Executive sponsors

John Stankey
Chairman, CEO and President
Jeff McElfresh
Chief Operating Officer
Pascal Desroches
Chief Financial Officer, retiring Dec 31, 2026 · AI business cases are tied to the CFO's office
Jennifer Biry
Deputy CFO; CFO from Jan 1, 2027

Technology

Jeremy Legg
Chief Technology Officer, AT&T Services

Chief Data Office

Not disclosed

Andy Markus
SVP and Chief Data and AI Officer
Mark Austin
Vice President, Data Science and AI · Oversees the AI tools AT&T's employees use

Legal

About 30 in LegalEdge (May 2026)

David McAtee
Senior Executive Vice President and General Counsel
Bill Ryan
Senior Vice President and Chief Compliance Officer · Spoke for AT&T on the OpenAI legal collaboration (Oct 2026)

Use cases

AT&T runs AI in production in customer care, fraud review, network operations, software development, legal work and field operations, and Andy Markus said in Jan 2026 that hundreds more business cases await prioritization. AT&T has published no dollar return of its own for any named use case: the roughly $150 million in annual customer care savings appears only in a Google Cloud case study, the fraud review and call summary figures are task timings, and the AI Gateway and coding savings measure what AI itself costs. For AI as a whole, Markus said on Mar 3, 2026 that AT&T earns about 5 times each AI dollar back in free cash flow within the same year, up from about 2 times in Jan 2026.

Use case	Area	Status
Virtual voice and chat agents AT&T's virtual voice and chat agents run on Google's Gemini Enterprise for Customer Experience, remember up to 60 days of customer history, and resolve 20% more inquiries without a human, while BigQuery routes calls by intent. Google Cloud's undated case study quotes Mark Austin crediting the agents with about $150 million in annual savings, a figure AT&T has not published on its own channels.	Customer care	In production
Call summarization Ask AT&T summarizes inbound customer calls, which AT&T said in Sep 2024 saves 30 seconds to several minutes per call, and Adaptive ML fine-tuned a Gemma 12B model that produces the daily call transcripts 30% faster than general-purpose APIs (Jun 2026).	Customer care	In production
Customer service update requests The first production tool built with Ask AT&T Workflows, made by AT&T Business and the technology development teams, takes customer service update requests, synchronizes data across systems and installs information in real time, which AT&T says cuts wait times.	AT&T Business	In production
AI assistant in the AT&T app The new AT&T app, rolled out from Mar 18, 2026, is built around a generative AI assistant that gives customers shopping and support advice alongside account management; Kellyn Smith Kenny, chief marketing and growth officer, and Jenifer Robertson, EVP and GM for Mass Markets, introduced it.	Consumer	In production
AT&T digital receptionist The AT&T digital receptionist is a network-based AI agent that talks to unknown callers to judge whether they are spam or fraud, then connects, disconnects or takes a message while the customer watches a live transcript; AT&T first deployed it in a limited fashion in Sep 2025.	Consumer	Pilot
Fraud case review AT&T's specialized reasoning models, built with Adaptive ML, cut fraud case review time from 6 minutes to 30 seconds and give fraud analysts up to 12 times the throughput (Jun 2026), and H2O.ai's Super Agent supports AT&T's fraud and financial crime detection (Jul 2026).	Fraud prevention	In production
AT&T Geo Modeler AT&T Geo Modeler simulates radio coverage with a Network Foundation Model and automatically retunes nearby cell sites when towers fail in storms or traffic shifts, and was live nationwide by Oct 2025.	Network	In production
Network alert agents AT&T network engineers chain Ask AT&T Workflows agents that correlate telemetry to locate an alert, pull change logs, check known issues, open a trouble ticket, propose and write a code patch, and summarize the incident, with the engineer approving each step.	Network	In production
Technician support, cell siting and fiber planning AT&T uses AI to support technicians during home visits, help engineers decide where to place cell sites and plan fiber deployment, which Jeff McElfresh said in Mar 2026 is improving productivity and capital allocation, without a figure.	Field operations	In production
AI-fueled coding AT&T's developers write code with AI using function-specific build archetypes that Andy Markus said produce close to production-grade code; Markus's team built an internal curated data product in 20 minutes instead of 6 weeks (Feb 2026), and model routing cut AT&T's AI coding costs by 56% with a 2% drop in quality (Aug 2026).	Software development	In production
Workload migration with Amazon Q Developer AT&T, with AWS Professional Services, is using Amazon Q Developer and other AWS agentic services to speed the migration of network service enablement to AWS Outposts (Feb 2026).	Technology	In production
AT&T LegalEdge AT&T LegalEdge, an AI-first in-house legal team launched in Jan 2026, handled nearly 100 matters in its first 6 months and on some matters does 20% to 25% of the work that had been almost entirely outsourced to law firms; from Oct 2026 AT&T Legal is building AI legal workflows, such as natural language intake, with OpenAI.	Legal	In production
Amdocs aOS modernization AT&T and Amdocs will pair Amdocs aOS, an agentic operating system for telecom, with AT&T's AI-enabled delivery framework under an 8-year agreement to modernize business and customer operations and lower the cost of running critical systems (Nov 3, 2026).	Business and customer operations	Announced

Adoption

AT&T spread AI in layers: machine learning in network incident management, robocall blocking, fraud and field routing for years before 2023, Ask AT&T for employees from Jun 2023, and agents built in Ask AT&T Workflows from Nov 2025. Ask AT&T's volume grew from about 1 billion tokens a day in Sep 2024 to about 27 billion in Feb 2026 and about 45 billion in Jul 2026, and Andy Markus said in Feb 2026 that Ask AT&T Workflows had reached more than 100,000 employees, more than half of whom say they use it every day. AT&T has not published a weekly active user count for Ask AT&T against its 133,030 employees at the end of 2025, and the 100,000 figure is an access figure for Workflows.

By the numbers

Employees with Ask AT&T Workflows: 100,000+; Andy Markus to VentureBeat, Feb 2026; more than half of them say they use it daily

AI return multiple claimed by the Chief Data and AI Officer: ~2x to ~5x; Andy Markus: about 2x (Jan 2026), about 5x in-year free cash flow (Mar 2026); no dollar base, in no filing or call

AI tokens processed a day: ~45 billion; AT&T blog, Jul 23, 2026; about 27 billion in Feb 2026 and about 1 billion in Sep 2024

Employee AI requests on open-weight models: ~40%; Mark Austin to The Information, Aug 2026, as reported by PYMNTS; goal of 60% to 70%

Employees: 133,030; Dec 31, 2025 (2025 Form 10-K)

Rollout, first pilot to today

2020
AT&T hired Andy Markus from WarnerMedia as chief data officer to standardize its data and AI tools and data science community.
VentureBeat
Oct 2022
Andy Markus said an evaluation found AI delivers billions of dollars of value to AT&T a year, and that every AT&T model is checked for bias with its SIFT tool.
VentureBeat
Jun 2023
AT&T launched Ask AT&T for employees on OpenAI's ChatGPT functionality, in an AT&T-dedicated Microsoft Azure tenant.
AT&T
Sep 2024
Andy Markus said Ask AT&T generated about 1 billion tokens a day and that Ask Data ranked first on the BIRD text-to-SQL leaderboard.
AT&T
Oct 2025
AT&T said its generative AI Geo Modeler was live nationwide, automatically retuning cell sites during network events.
AT&T
Nov 2025
AT&T added Ask AT&T Workflows, a drag-and-drop agent builder, to Ask AT&T, with a first production agent for customer service update requests.
AT&T
Feb 2026
Andy Markus said Ask AT&T Workflows had reached more than 100,000 employees and Ask AT&T processed about 27 billion tokens a day.
VentureBeat
Mar 2026
AT&T began rolling out a new customer app built around a generative AI assistant.
AT&T
Jul 2026
AT&T said Ask AT&T averaged about 45 billion tokens a day through its AI Gateway, and released the OTel 2.0 open telco model with the GSMA.
AT&T
Aug 2026
Mark Austin said about 40% of AT&T employee AI requests go to open-weight models, according to reports of his interview with The Information.
AI Weekly
Oct 2026
AT&T Legal began building AI legal workflows with OpenAI on top of LegalEdge, its AI-first in-house legal team.
AT&T

Economics & controls

Spend and ROI

Technology budget

AT&T does not disclose a technology or AI budget. AT&T's capital expenditures were $20.8 billion in 2025 and its capital investment, including vendor financing, $22.0 billion, most of it spent on networks, and AT&T guided capital investment of $23 billion to $24 billion a year for 2026 to 2028 (Jan 2026). Jeff McElfresh described a $250 billion commitment through 2030 tied to fiber, 5G, resilience, public safety, workforce development and AI era infrastructure (Fortune, Mar 2026).

What AI itself costs

AT&T has not published what it spends on AI. Andy Markus wrote in Jul 2026 that AT&T's AI Gateway cuts AI costs by as much as 90% and already saves millions, and said in Feb 2026 that rebuilding the Ask AT&T orchestration layer, with large models directing small ones, brought up to 90% cost savings. Mark Austin told The Information in Aug 2026 that model routing cut AI coding costs by 56% with a 2% drop in quality and that AT&T aims to hold employee spending on Anthropic and OpenAI models flat by moving 60% to 70% of requests to open-weight models.

Reported ROI

Andy Markus told CIO.com in Jan 2026 that AT&T's generative AI use cases return about 2 times their investment and that he expected the figure to grow in 2026, and said at Mobile World Congress on Mar 3, 2026 that AT&T generates a 5-fold return in free cash flow within the same year on every dollar spent on AI. AT&T has not published the dollar base or method behind either figure. Google Cloud's undated case study quotes Mark Austin crediting customer care agents built on Gemini Enterprise for Customer Experience with about $150 million in annual savings and an improved Net Promoter Score, a figure that appears in no AT&T release, filing or call. Markus said in Oct 2022 that an internal evaluation found AI delivered billions of dollars of value to AT&T a year.

Cost program

AT&T achieved more than $1 billion of cost savings in 2025, John Stankey said in Jan 2026, and plans to speed efficiency gains by using AI and moving more customer transactions to digital. Pascal Desroches said in Apr 2026 that the target of $4 billion in annual cost savings by the end of 2028 rests on workforce optimization, vendor rationalization, AI enablement, digitalization and lower legacy costs, and that legacy decommissioning contributes significantly, without giving AI's share. AT&T's 2025 Form 10-K says AT&T expects savings from AI-driven efficiencies in network design and operations, software development, sales, marketing, customer support and general and administrative costs, and warns that there is no assurance AI will meaningfully enhance its business.

Headcount

AT&T had about 133,030 employees at Dec 31, 2025, about 43% of them represented by unions. AI Chat Daily, citing Wired, reported in Sep 2026 that AT&T cut about 8,000 jobs in 2025 and 2,100 in the first half of 2026, to nearly 131,000, and that CTO Jeremy Legg said AT&T will not have the same headcount in 5 years; AT&T disputed a reported internal target of about 85,000 employees by 2030 as inaccurate. AT&T has not said how much of its workforce reduction comes from AI rather than copper retirement and other changes.

Governance and risk

Use case approval

AT&T sends every proposed generative and agentic AI use case through its Generative AI Transformation Office, which weighs value to customers and the company, and each use case gets a formal business case tied to the CFO's office, Andy Markus told CIO.com in Jan 2026. Markus said he looks for a free cash flow benefit in most cases, with exceptions for use cases that directly serve customers.

Controls on agents

AT&T requires human checkpoints in every Ask AT&T Workflows process, logs every agent action, and enforces data isolation, retention policies and role-based access whenever one agent hands work to another (Nov 2025). Andy Markus said in Feb 2026 that a human on the loop checks the whole chain of agents even when steps run on their own.

Data boundary

AT&T built Ask AT&T with Microsoft in an AT&T-dedicated Azure tenant that it said was pressure tested for leakage, so employees can bring company data in without it reaching the public domain, and Ask AT&T users are responsible for checking that results are accurate (2023). AT&T does not use Chinese open-weight models from DeepSeek or Moonshot and is evaluating their risks, Mark Austin told The Information in Aug 2026.

Policy and bias

AT&T's AI Policy and AI Operating Guidelines set the rules for building and using AI, under principles of human oversight, transparency and controllability (2023). AT&T built SIFT, its System for the Integration of Fairness and Transparency, and rolled it out so every model is evaluated for bias and users are taken through mitigation steps (Oct 2022).

Legal work

AT&T Legal says lawyers decide how AI agents are used in each legal workflow, with governance, confidentiality, accountability and human review built in, and David McAtee said in Oct 2026 that legal judgment remains human.

Disclosed risks

AT&T's 2025 Form 10-K warns that generative AI models may produce incorrect output, release private or confidential information, reflect bias or infringe intellectual property, that AI regulation is uncertain, and that criminals' use of AI may make cyberattacks more frequent and severe. AT&T's Audit Committee oversees privacy and data security, with reports from the CISO and CTO.

Recent signals

Nov 3, 2026
AT&T and Amdocs signed an 8-year agreement to modernize AT&T's business and customer operations with AI, pairing Amdocs aOS with AT&T's AI-enabled delivery framework under one delivery team.
AT&T
Oct 8, 2026
AT&T Legal became a flagship in-house legal design partner for OpenAI, building AI legal workflows on top of LegalEdge, which handled nearly 100 matters in its first 6 months.
AT&T
Sep 23, 2026
AI Chat Daily, citing Wired, reported that CTO Jeremy Legg said AT&T will not have the same headcount in 5 years as it automates internal work and retires copper, and that AT&T disputed a reported internal target of about 85,000 employees by 2030.
AI Chat Daily
Aug 20, 2026
The Information reported that Mark Austin said model routing cut AT&T's AI coding costs by 56% with a 2% drop in quality and that about 40% of employee AI requests go to open-weight models, with a goal of 60% to 70%.
PYMNTS
Jul 24, 2026
AT&T and the GSMA released OTel 2.0, an open telco model post-trained from Google's Gemma 4 31B-IT on 400 billion telecom tokens using AMD GPUs and Microsoft Foundry.
GSMA
Jul 23, 2026
Andy Markus wrote that AT&T averages about 45 billion tokens a day and that its cache-aware AI Gateway cuts AI costs by as much as 90%.
AT&T
Jul 14, 2026
H2O.ai said AT&T is using its Super Agent in Ask AT&T and in customer experience, fraud prevention, field operations and enterprise research workflows.
H2O.ai release via TechIntelPro
Jun 15, 2026
AT&T named Jennifer Biry Deputy CFO from Jul 6, 2026 and CFO from Jan 1, 2027, after Pascal Desroches announced his retirement effective Dec 31, 2026.
AT&T Form 8-K
Jun 10, 2026
Adaptive ML said AT&T renewed and doubled its use of the Adaptive Engine platform after a year in production, with specialized models cutting fraud case review from 6 minutes to 30 seconds.
Adaptive ML release via The AI Journal
Apr 2026
Pascal Desroches said on the first-quarter call that AI enablement is one of the levers behind AT&T's target of $4 billion in annual cost savings by the end of 2028.
Q1 2026 earnings call transcript
Mar 18, 2026
AT&T began rolling out a new customer app built around a generative AI assistant for shopping and support.
AT&T
Mar 3, 2026
Andy Markus said at Mobile World Congress that AT&T processes 27 billion tokens a day, earns a 5-fold in-year return in free cash flow on every AI dollar, and deploys AI straight into production instead of running long pilots.
Broadband Breakfast
Feb 26, 2026
Andy Markus told VentureBeat that Ask AT&T Workflows had reached more than 100,000 employees, more than half of whom say they use it every day.
VentureBeat
Feb 4, 2026
AT&T, AWS and Amazon Leo announced that AT&T will move on-premises workloads to AWS Outposts using Amazon Q Developer and connect AWS data centers with fiber.
TelecomTV
Jan 28, 2026
Andy Markus told CIO.com that AT&T's generative AI use cases return about 2 times their investment and that each passes through a Generative AI Transformation Office with a business case tied to the CFO's office.
CIO.com

Quick answers

How does AT&T deploy AI?: AT&T deploys AI through Ask AT&T, an employee generative AI platform its Chief Data Office launched in 2023 in a dedicated Microsoft Azure tenant, which now routes about 45 billion tokens a day between OpenAI, Anthropic and open-weight models (Jul 2026). Chief Data and AI Officer Andy Markus said in Feb 2026 that the Ask AT&T Workflows agent builder had reached more than 100,000 employees, but AT&T has published no dollar return it credits to AI.
What does AT&T use AI for?: AT&T runs AI in production in customer care, fraud review, network operations, software development, legal work and field operations, and Andy Markus said in Jan 2026 that hundreds more business cases await prioritization. AT&T has published no dollar return of its own for any named use case: the roughly $150 million in annual customer care savings appears only in a Google Cloud case study, the fraud review and call summary figures are task timings, and the AI Gateway and coding savings measure what AI itself costs. For AI as a whole, Markus said on Mar 3, 2026 that AT&T earns about 5 times each AI dollar back in free cash flow within the same year, up from about 2 times in Jan 2026.
Who leads AI at AT&T?: AI at AT&T is led by Jeremy Legg (Chief Technology Officer, AT&T Services); Andy Markus (Senior Vice President and Chief Data and AI Officer); Mark Austin (Vice President, Data Science and AI). Ownership of AI on the org chart: Andy Markus, SVP and Chief Data and AI Officer; reporting line not published.
What AI platform does AT&T use?: Ask AT&T, the generative AI platform AT&T's Chief Data Office launched in Jun 2023, now bundles Ask Docs, Ask Data and the Ask AT&T Workflows agent builder, and carried about 45 billion tokens a day in Jul 2026.
Which AI models does AT&T use?: AT&T sends each request through its own AI Gateway to OpenAI or Anthropic models or to open-weight models such as Nvidia Nemotron, Meta Llama and Google Gemma, which handled about 40% of employee requests in Aug 2026.
How many AT&T employees use AI?: AT&T rolled Ask AT&T Workflows out to more than 100,000 employees, and more than half of them say they use it every day (Andy Markus, Feb 2026); AT&T had about 133,030 staff at the end of 2025.
What return has AT&T reported from AI?: AT&T sits at Piloting on an officially reported company-wide rollout: Chief Data and AI Officer Andy Markus said in Feb 2026 that the Ask AT&T Workflows agent builder had reached more than 100,000 employees, more than half of whom say they use it daily, and his Chief Data Office builds and owns Ask AT&T. AT&T has reported no business return it credits to AI in dollars, cost, headcount or revenue. The return multiples Markus has given are unbased executive estimates that conflict: about 2 times on generative AI investment to CIO.com in Jan 2026 and a 5-fold in-year free cash flow return at Mobile World Congress on Mar 3, 2026, neither with a dollar base or method and neither in a filing or earnings call, where AI is one unquantified lever in a $4 billion annual cost-savings target for the end of 2028. The roughly $150 million annual customer care saving appears only in a Google Cloud case study, not in AT&T's own releases. A dollar figure AT&T credits to AI, or reported returns in 2 functions, would move AT&T to Scaling.

Open questions

What dollar base and method sit behind Andy Markus's 5-fold in-year free cash flow return on AI (Mar 2026), what counts as a dollar spent on AI, and how does it relate to the roughly 2 times return he cited in Jan 2026?
Will AT&T confirm in its own releases, filings or calls the roughly $150 million in annual customer care savings that Google Cloud's case study attributes to Mark Austin, and over what period was it measured?
How much of AT&T's more than $1 billion in 2025 cost savings and of its $4 billion target for the end of 2028 does AT&T credit to AI rather than copper retirement, digitalization and workforce changes?
How many AT&T employees use Ask AT&T every week, given that the 100,000-plus figure counts access to Ask AT&T Workflows?
To whom does Andy Markus report, and where does the Generative AI Transformation Office sit?
Where does AT&T serve its open-weight and OTel models, on its own hardware, in Microsoft Azure or elsewhere?
What does AT&T spend on AI models and compute, and how much do its Anthropic and OpenAI bills come to?
What do deployed tools with no published return, such as LegalEdge, fraud case review, Geo Modeler, network alert agents and the AI assistant in the AT&T app, save in cost or outside spending?
What savings does AT&T expect from the 8-year Amdocs agreement, and what is it worth?
How many of AT&T's job cuts does AT&T attribute to AI, and does an internal headcount target for 2030 exist?

Sources

AT&T

Filings and investor materials

Partners

News

Data confidence

Confirmed: Stated in a primary or reputable public source.
Estimated: Derived or approximated from public reporting.
Unconfirmed: Reported, but not yet corroborated.

AT&T profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 5, 2026

More company profiles

All profiles