Who leads AI at Uber?

AI at Uber is led by Praveen Neppalli Naga (Chief Technology Officer); Johannes Gehrke (Technical fellow, AI (joined from Microsoft in 2026)); Uday Kiran Medisetty (Distinguished Engineer, engineering productivity). Ownership of AI on the org chart: CTO Praveen Neppalli Naga, who leads engineering and science.

What AI platform does Uber use?

Michelangelo, Uber's in-house machine learning platform since 2016, plus an AI Gateway to outside and self-hosted models, an agent platform built in early 2025 and an MCP Gateway with more than 800 servers (Oct 2026).

Which AI models does Uber use?

Uber routes calls to OpenAI and Anthropic models through its AI Gateway, has fine-tuned open-weight Llama 2 models in-house, and is moving some tasks to cheaper open-weight models to hold AI spend flat (Aug 2026).

How many Uber employees use AI?

Near 100% of Uber engineers use AI coding tools (Aug 2026), and Uber began rolling out ChatGPT Work to more than 30,000 employees on Oct 22, 2026, after a 1,000-person pilot.

How far has AI spread at Uber?

Uber sits at Scaling: machine learning runs its core marketplace at about 10 million predictions a second, nearly all engineers use AI coding tools, thousands of internal agents run on its agent platform, and CTO Praveen Neppalli Naga owns engineering and science while the CFO reported AI's effect on hiring on the May and Aug 2026 earnings calls. Uber falls short of Embedded because ChatGPT Work only opened to its more than 30,000 employees on Oct 22, 2026, with no company-wide weekly use figure published, and Uber gives no dollar figure for AI's financial effect.

Verified Nov 1, 2026

Uber

Scaling

Mobility and delivery
Michelangelo, built in-house
70%+ of pull requests by agents
ChatGPT Work open to 30,000+ staff

Uber deploys AI through Michelangelo, the machine learning platform it built in 2016 that runs pricing, matching and routing at about 10 million predictions a second, and through gateways it built to connect its applications and agents to OpenAI, Anthropic and open-weight models. Engineering has gone furthest: nearly all Uber engineers use AI coding tools, and more than 70% of pull requests were attributed to AI agents by Aug 2026. Uber began rolling out OpenAI's ChatGPT Work agent to more than 30,000 employees on Oct 22, 2026.

uber.com

Their approach to AI

The platform: Michelangelo, Uber's in-house machine learning platform since 2016, plus an AI Gateway to outside and self-hosted models, an agent platform built in early 2025 and an MCP Gateway with more than 800 servers (Oct 2026).
The models: Uber routes calls to OpenAI and Anthropic models through its AI Gateway, has fine-tuned open-weight Llama 2 models in-house, and is moving some tasks to cheaper open-weight models to hold AI spend flat (Aug 2026).
The reach: Near 100% of Uber engineers use AI coding tools (Aug 2026), and Uber began rolling out ChatGPT Work to more than 30,000 employees on Oct 22, 2026, after a 1,000-person pilot.
The catch: Uber puts no dollar figure on AI's return. Uber spent its 2026 budget for AI coding tools by April, capped coding-tool spend per employee in Jun 2026, and COO Andrew Macdonald said in May 2026 that AI use did not yet show up clearly in shipped features.

Snapshot

Employees: ~34,000; Dec 31, 2025, about 20,300 of them outside the U.S. (2025 Form 10-K); about 10% cut from Sep 2026
Technology spend: $3.40B R&D (2025); Up from $3.11B in 2024; no separate technology or AI budget published
AI/ML headcount: Not disclosed; Uber does not publish engineering or AI staff counts
Who owns AI: Praveen Neppalli Naga, CTO; Leads engineering and science strategy; CFO Balaji Krishnamurthy reports AI's effect on hiring
Platform: Michelangelo, built in-house; Launched 2016; AI Gateway, agent platform and MCP Gateway added since 2023
Models: OpenAI, Anthropic, open-weight; GPT-6.1 Sol is the ChatGPT Work default (Oct 2026); Claude Code in engineering
Reach: ~100% of engineers; Aug 2026; ChatGPT Work open to 30,000+ employees from Oct 22, 2026
Use cases: Thousands of internal agents; May 2026; 5,000+ ML models in production and about 400 active ML projects in May 2024

Key findings

Uber began rolling out OpenAI's ChatGPT Work agent to more than 30,000 employees on Oct 22, 2026, after a 1,000-employee pilot, calling it Uber's first company-wide agentic platform for knowledge work, while its developers keep using OpenAI's Codex.
Uber said in May 2026 that more than 10% of production-ready code was written autonomously by AI coding agents, and its engineering blog said in Aug 2026 that more than 70% of pull requests are attributed to local or cloud agents; the two figures measure different things.
Uber used up its 2026 budget for AI coding tools within about four months, driven by engineers' use of Claude Code, according to The Information in Apr 2026; CFO Balaji Krishnamurthy said on May 6, 2026 that Uber had re-upped its AI investment in exchange for slower hiring, and Uber capped spending at $1,500 a month per employee per coding tool in Jun 2026.
Weekly active users of Uber's agentic tools, engineers and non-engineers, grew 7x from Feb to Aug 2026 while weekly agent requests grew 9.4x, and Uber said in Aug 2026 that total AI spend had broadly stabilized since April as cost per token fell.
Uber cut about 10% of its customer support team in Jul 2026, with a spokesperson citing the aim to keep embracing AI, and about 10% of all staff on Sep 2, 2026, in a memo from Dara Khosrowshahi that does not cite AI.
Uber also sells AI work: Uber AI Solutions offers data labeling, evaluation and model training services through a platform with over 10 million contributors and experts (Mar 2026), and Dara Khosrowshahi said in Nov 2025 that it could become another profitable business line.

Where Uber sits

Stage: Scaling
Uber sits at Scaling: machine learning runs its core marketplace at about 10 million predictions a second, nearly all engineers use AI coding tools, thousands of internal agents run on its agent platform, and CTO Praveen Neppalli Naga owns engineering and science while the CFO reported AI's effect on hiring on the May and Aug 2026 earnings calls. Uber falls short of Embedded because ChatGPT Work only opened to its more than 30,000 employees on Oct 22, 2026, with no company-wide weekly use figure published, and Uber gives no dollar figure for AI's financial effect.
How the stages are defined
Ownership: CTO Praveen Neppalli Naga, who leads engineering and science
Platform: Hybrid: Michelangelo and AI gateways built in-house; ChatGPT Work bought
Models: OpenAI and Anthropic behind an AI Gateway, plus open-weight models
Hosting: Co-located data centers, Oracle Cloud Infrastructure, Google Cloud and AWS (Trainium pilot); OpenAI and Anthropic APIs, personal data redacted first
Reach: Near 100% of engineers (Aug 2026); ChatGPT Work open to 30,000+ staff (Oct 2026)

How they deploy

Build the platform, buy frontier models

Uber built Michelangelo in early 2016 to standardize machine learning across the company, and its Michelangelo team later built the GenAI Gateway, which by Jul 2024 gave teams one interface to OpenAI, Google Vertex AI and Uber-hosted models. Uber buys frontier models and employee tools on top of that stack: ChatGPT Enterprise and ChatGPT Work from OpenAI, OpenAI's Codex and Anthropic's Claude Code for engineers, and Cursor, Lovable and Figma Make for prototyping and design.

Engineering first

Dara Khosrowshahi said on Uber's Q1 2023 earnings call that AI's earliest and most significant effect on Uber would be on developer productivity, according to Uber's engineering blog. Uber now calls its approach a Software Factory: agents start sessions on their own for code review, fixing failed builds, on-call alert triage, bug debugging and maintenance, with humans reviewing or escalating (Aug 2026).

Pick the cheapest model that does the job

Uber benchmarks frontier and open-weight models on its own engineering work and switches to whichever gives the best cost and quality, a choice its engineers say shifts every few weeks (Aug 2026). Balaji Krishnamurthy said in Aug 2026 that Uber sets better defaults by use case, moves some tasks to lower-cost or open-weight models and shows employees their own spend, so cost per token fell while adoption rose. Dara Khosrowshahi wrote in Aug 2026 that AI did not converge on one foundation model, that applications increasingly use the right model for each task, and that autonomous driving will likely follow the same pattern.

Agents in operations and the apps

Uber was piloting first-party agents by Feb 2026, including an assistant for drivers and couriers and consumer agents for the Uber and Uber Eats apps, and was already running Merchant Reasoning Agents in delivery operations. By Aug 2026 Uber had launched Cart Assistant for grocery orders and a support agent that resolves eligible fare disputes on its own. Dara Khosrowshahi said in Aug 2026 that Uber expects AI to pay off through many small improvements to the marketplace, not one large win.

Outside AI assistants as a channel

Uber launched integrations with ChatGPT across Rides and Eats by Feb 2026, and Dara Khosrowshahi said in May 2026 that Uber will build interfaces for outside agents such as Apple's, OpenAI's, Anthropic's Claude and Google's Gemini while expecting most transactions to stay in its own apps. Uber's 2025 Form 10-K warns that AI-enabled platforms and digital assistants could come between Uber and its consumers.

Physical AI through partners

Uber puts autonomous vehicles on its network through partners such as NVIDIA, Waymo, Wayve, WeRide, Pony.ai and Nuro, and said in Oct 2025 that it would run end-to-end operations, from remote assistance to customer support, for NVIDIA-based robotaxi fleets. Uber and NVIDIA said in Oct 2025 they are building a robotaxi data factory on NVIDIA's Cosmos platform, with Uber contributing more than 3 million hours of robotaxi-specific driving data.

Deployment model: Uber runs AI as shared platforms built by engineering, with use cases owned by product and business teams. Engineering teams under CTO Praveen Neppalli Naga build Michelangelo, the AI Gateway, the agent platform and identity service, and the MCP Gateway, while engineering productivity teams run the coding agents and uReview code review. Product teams in Mobility, Delivery and Freight and corporate functions such as finance, legal and security build their own agents on those platforms. Uber's Sep 2026 reorganization combined its Core Services engineering and science teams, matching the structure already used in Mobility and Delivery. For all staff, Uber supports ChatGPT Work with an internal AI skills library, approved integrations to Google Drive, Gmail, Google Calendar and Slack, and employee training (Oct 2026).
Model providers: Uber's GenAI Gateway connected teams to OpenAI, Google Vertex AI (PaLM 2) and Uber-hosted open models such as Llama 2 in Jul 2024, and in May 2026 Uber described its AI Gateway as mediating calls to OpenAI and Anthropic APIs. OpenAI supplies ChatGPT Enterprise and ChatGPT Work, with GPT-6.1 Sol as the default model (Oct 2026), and Codex for developers. Anthropic's Claude Code runs in Uber's engineering workflows, and Uber's uReview code reviewer worked best with Claude Sonnet 4 generating comments and OpenAI's o4-mini-high grading them (Aug 2025). Uber has fine-tuned Llama 2 models up to 70 billion parameters in-house and reports that its fine-tuned Uber Eats models reach performance similar to GPT-4 at much higher traffic.
Data and infrastructure: Uber began moving from its own data centers to Oracle Cloud Infrastructure and Google Cloud in Feb 2023, and its 2025 Form 10-K says it runs on a mix of third-party cloud services and co-located data centers in the U.S. and abroad. Uber managed more than 5,000 GPUs across its own data centers, Oracle Cloud Infrastructure and Google Cloud in May 2024, and trains large language models on NVIDIA A100 GPUs in its own clusters and H100 GPUs in Google Cloud. Kamran Zargahi, Uber's VP of Engineering for cloud, wrote in May 2026 that Uber still runs on-premises data centers and can shift workloads between them and the cloud. AWS said on Apr 7, 2026 that Uber is moving more trip-serving workloads to AWS Graviton4 instances and has begun pilot training some of its AI models, such as those that assign drivers and estimate arrival times, on AWS Trainium3. Uber's agents query an AI Context Graph of 24 million nodes built from more than 30 internal systems (Aug 2026), and the AI Gateway redacts personal data before requests reach outside model vendors.

Platforms and internal tools

MichelangeloEarly 2016
Michelangelo is Uber's in-house machine learning platform for training, deploying and serving models, extended since 2023 for large language models with a model catalog, an evaluation framework and a prompt engineering toolkit. Michelangelo had about 400 active ML projects, more than 20,000 training jobs a month and more than 5,000 models in production in May 2024, and serves about 10 million real-time predictions a second at peak.
Source
GenAI Gateway and AI GatewayBy Jul 2024
Uber's GenAI Gateway is a single OpenAI-compatible interface to outside and Uber-hosted large language models, with personal data redaction, budget management, audit logs and a security review of each use case before access, and it served 16 million queries a month for close to 30 teams in Jul 2024. Uber's May 2026 description of its agent stack names an AI Gateway that mediates calls to OpenAI and Anthropic APIs and applies AI Guard protections against prompt injection and jailbreaks; Uber's legal team still cited the GenAI Gateway for model access in Oct 2026.
Users
Close to 30 teams (Jul 2024)
Models
OpenAI, Anthropic, Google Vertex AI, Uber-hosted open models
Source
Agent platform and identity serviceEarly 2025
Uber's internal agent platform lets teams build agents in Python with an internal SDK or without code through a user interface, and gives each agent a short-lived, scoped identity checked against an Agent Registry at every hop. Thousands of Uber's internal agents had adopted the identity system by May 2026.
Source
MCP GatewayBy 2026
Uber's MCP Gateway connects AI agents to Uber's back-end services and to outside tools such as Jira and Google, with tool-level authorization, redaction of sensitive data and every tool disabled until its owning team enables it. The MCP Gateway hosted more than 800 MCP servers and 5,000 tools in Oct 2026; Uber's Aug 2026 Software Factory post put the count at over 1,000 servers, internal and outside.
Source
Software Factory coding agents
Uber's Software Factory runs local and cloud coding agents across the development life cycle on a harness that serves frontier and open-weight models, with Anthropic's Claude Code among them, more than 3,600 agent skills built by engineers and more than 30,000 skill executions a day (Aug 2026). Uber nudges engineers with Slack alerts at 50%, 80% and 100% of expected spend and defaults subagents to cheaper models. Uber's developers also use OpenAI's Codex (Oct 2026).
Users
Engineers and non-engineers; weekly active users up 7x, Feb to Aug 2026
Models
Anthropic Claude Code, OpenAI Codex, open-weight models
Source
uReviewBy Aug 2025
uReview is Uber's AI code reviewer, which comments on bugs, error handling, security issues and coding standards in every monorepo and covered over 90% of about 65,000 weekly code changes in Aug 2025. Engineers marked about 75% of uReview comments useful, and Uber says uReview costs an order of magnitude less than typical third-party tools.
Models
Claude Sonnet 4 with OpenAI o4-mini-high as grader (Aug 2025)
Source
ChatGPT WorkOct 22, 2026
ChatGPT Work is OpenAI's agent inside ChatGPT Enterprise, which Uber is rolling out company-wide for research, meeting briefs, presentations, data analysis and multi-step tasks across Google Drive, Gmail, Google Calendar and Slack. Sumanth Sukumar, an Uber VP of engineering, told TechCrunch in Oct 2026 that about 80% of the pilot's early adopters were non-engineers, and that Uber had paired 30 AI engineers with teams across the company earlier in 2026 to automate parts of their jobs. Uber and OpenAI say they will publish adoption and outcome data as the rollout progresses.
Users
Available to 30,000+ employees (Oct 2026); 1,000 in the pilot
Models
OpenAI, with GPT-6.1 Sol as the default
Source
Uber AI SolutionsExpanded Jun 2025
Uber AI Solutions is Uber's business selling data annotation, data collection, evaluation and model training work to AI labs and enterprises, using Uber's task-routing platform and a network of over 10 million contributors and specialized experts (Mar 2026). Uber expanded Uber AI Solutions to 30 countries in Jun 2025.
Source

Vendors and partners

OpenAIBy 2024
ChatGPT Enterprise and ChatGPT Work for employees, Codex for developers, models behind the AI Gateway, ChatGPT app integrations · OpenAI's GPT-4 models powered Uber's QueryGPT and invoice automation in 2024 and 2025, and Uber began rolling out ChatGPT Work to more than 30,000 employees on Oct 22, 2026.
AnthropicBy Aug 2025
Claude Code for engineers; Claude models in uReview and behind the AI Gateway · Anthropic's Claude Sonnet 4 was the comment generator in uReview's best-performing setup in Aug 2025. CTO Praveen Neppalli Naga told The Information in Apr 2026 that engineers' use of Claude Code had used up Uber's 2026 budget for AI coding tools, and Uber's Aug 2026 Software Factory post measured its token savings in Claude Code sessions.
Oracle Cloud InfrastructureFeb 2023
Public cloud for Uber's migration off its own data centers; GPUs; Ampere Arm instances · Uber worked with Oracle and Ampere on OCI's AmpereOne M A4 compute instances (Dec 2025).
Google CloudFeb 2023
Public cloud; H100 GPUs for large language model training; Vertex AI models via the GenAI Gateway · Google Cloud hosts NVIDIA H100 hosts that Uber uses alongside its own A100 clusters to train Llama-family models.
Amazon Web ServicesExpanded Apr 2026
Graviton4 instances for trip-serving workloads; pilot training of Uber AI models on Trainium3 · AWS said on Apr 7, 2026 that Uber had begun pilot training some AI models on Trainium3, including models that assign drivers or couriers and estimate arrival times.
NVIDIAJan 2025
Autonomous vehicle compute and software partner; robotaxi data factory on NVIDIA Cosmos · NVIDIA said in Mar 2026 that it will launch L4 robotaxis running its software on Uber across 28 cities by 2028.
Cursor, Lovable and Figma Make
AI prototyping and design tools used by Uber product and design teams · Uber's design systems team built uSpec, a Cursor agent that writes Figma component specs and runs locally so design data stays on Uber's network (Mar 2026).

Leadership

AI, data and platform leaders

Praveen Neppalli Naga

Chief Technology Officer

Praveen Neppalli Naga joined Uber in 2015 after seven years at LinkedIn and leads Uber's engineering and science strategy and technical execution. Neppalli Naga told The Information in Apr 2026 that Uber had already spent its 2026 budget for AI coding tools, wrote in Aug 2026 that Uber's AI costs fell as adoption rose, and announced the ChatGPT Work rollout to Uber employees in Oct 2026.

Johannes Gehrke

Technical fellow, AI (joined from Microsoft in 2026)

Unconfirmed

Johannes Gehrke left Microsoft in 2026 after 14 years, most recently as a technical fellow in AI infrastructure, to join Uber, according to GeekWire, and Uber's Aug 2026 Software Factory post thanks him for his leadership. Uber has not published his title.

Uday Kiran Medisetty

Distinguished Engineer, engineering productivity

Uday Kiran Medisetty leads Uber's engineering productivity initiatives, including agentic coding, AI debugging, code review and large-scale refactoring, and wrote Uber's Aug 2026 post on the Software Factory approach that Uber shared at the AI Engineer 2026 conference.

Kamran Zargahi

Vice President of Engineering, cloud engineering

Kamran Zargahi is responsible for Uber's cloud engineering, including the long-running move from Uber's own data centers to cloud providers and the choice of GPU and Arm hardware, according to his May 2026 post on Uber's engineering blog. Zargahi spoke for Uber in AWS's Apr 2026 announcement of Uber's Trainium pilot for AI model training.

Dan Carpenter

General Manager, Uber AI Solutions

Unconfirmed

Dan Carpenter is listed as General Manager of Uber AI Solutions on the MACHINA 2026 summit's speaker page, after commercial roles at FluidStack and Omniva and 12 years at AWS. Bloomberg reported on Jul 1, 2026 that two Uber AI Solutions leaders had left, which Uber called part of a broader leadership transition at the division.

Executives

Dara Khosrowshahi

Chief Executive Officer

Dara Khosrowshahi has been Uber's CEO since 2017 and was earlier CEO of Expedia and CFO of IAC Travel. Khosrowshahi frames AI at Uber as many small marketplace gains and sets the company's line on outside AI agents and autonomous vehicles on earnings calls.

Andrew Macdonald

President and Chief Operating Officer

Andrew Macdonald became President and COO on Jun 2, 2025, responsible for Uber's Mobility, Delivery and Autonomous businesses and cross-platform functions such as customer support. Macdonald said in a May 2026 podcast interview, reported by Fortune, that Uber could not yet draw a clear line from engineers' AI tool use to more useful features shipped to consumers.

Balaji Krishnamurthy

Chief Financial Officer

Balaji Krishnamurthy joined Uber in 2019, ran investor relations from 2020 to 2023 and strategic finance from 2023, and became CFO on Feb 16, 2026. Krishnamurthy sets out Uber's AI economics on earnings calls: AI spending re-upped in May 2026, traded against slower hiring, and held broadly stable by Aug 2026.

Jill Hazelbaker

President and Chief Corporate Affairs Officer

Jill Hazelbaker became President and Chief Corporate Affairs Officer on May 11, 2026, adding the people function and safety operations to marketing, communications and public policy; Dara Khosrowshahi tied the move to a period of rapid change from advances in AI to autonomous vehicles. Hazelbaker earlier led communications and public policy at Snap.

Tony West

Senior Vice President, Chief Legal Officer and Corporate Secretary

Tony West leads Uber's global Legal, Compliance and Ethics, and Security functions, and was earlier General Counsel of PepsiCo and Associate Attorney General at the U.S. Department of Justice. Uber's legal team co-built the Legal Redlining Agent, which Uber described in Oct 2026.

Sachin Kansal

Chief Product Officer

Sachin Kansal joined Uber in 2017 as its first product leader for safety technology and now leads product management, design and product operations for Uber's Mobility and Delivery business. Kansal was earlier VP of Product at Lookout.

Sarfraz Maredia

Head of Autonomous Mobility & Delivery (leaving Nov 16, 2026)

Sarfraz Maredia joined Uber in 2014 and ran Mobility in the U.S. and Canada and Uber Eats in the Americas before leading the autonomous business, whose marketplace, fleet operations and partnerships he oversees. Bloomberg reported on Oct 13, 2026 that Maredia will leave Uber on Nov 16, 2026 for a role outside transportation, with his duties split between two senior directors.

Board

Nikesh Arora

Independent director

Nikesh Arora, chairman and CEO of Palo Alto Networks, joined Uber's board on May 31, 2025. Uber's 2026 proxy says it added Arora in response to investor focus on cybersecurity and AI oversight.

AI org map

Shared AI platforms built by central engineering; agents and models owned by product and business teams

Uber's executive team has no dedicated AI or data chief. CTO Praveen Neppalli Naga, a member of the executive team, leads engineering and science, including the platform teams behind Michelangelo, the AI and MCP gateways, the agent platform and the Software Factory coding agents. Product and science teams in Mobility, Delivery and Freight, and functions such as finance, legal, security and customer support, build agents on those platforms, and CFO Balaji Krishnamurthy weighs AI spending against hiring. Uber AI Solutions runs as a separate unit, the autonomous business sits under the President and COO, and Uber's board oversees AI as a whole.

Uber does not publish how many people work on AI. Uber's AI work sits in three places: platform teams under CTO Praveen Neppalli Naga that build Michelangelo, the AI and MCP gateways and the coding agents; product and science teams in Mobility, Delivery and Freight; and functional teams in finance, legal and security that build their own agents. Uber's research and development expense, mostly engineering, design and product pay, was $3.40 billion in 2025, and Uber AI Solutions runs a separate network of over 10 million outside contributors and experts who are not Uber employees.

Reporting lines

Chief Executive Officer
Dara Khosrowshahi
- Engineering and Science
  Praveen Neppalli Naga, CTO
  Line to the CEO inferred from Uber's executive team listing; Core Services engineering and science teams combined in Sep 2026
  - AI Platform (Michelangelo, AI Gateway, agent platform, MCP Gateway)
    Leader not published; inferred from Uber engineering blog posts
  - Engineering productivity and Software Factory
    Uday Kiran Medisetty, Distinguished Engineer (initiative lead)
    Coding agents, uReview, AI debugging and refactoring
  - Cloud engineering
    Kamran Zargahi, VP of Engineering
- Finance
  Balaji Krishnamurthy, CFO
  Uber FinTech built the Finch data agent and invoice automation
- Mobility, Delivery and Autonomous
  Andrew Macdonald, President and COO
  Product teams behind Cart Assistant, the driver assistant and merchant agents; customer support (Jun 2025 Form 8-K)
  - Global Community Operations
    Megha Yethadka, VP
    Customer support, where the fare dispute agent runs; line inferred from the Jun 2025 Form 8-K
  - Autonomous Mobility & Delivery
    Sarfraz Maredia (leaving Nov 16, 2026)
    Duties to be split between two senior directors after Nov 16, 2026
Uber AI Solutions
Dan Carpenter, GM
Reporting line not published

Uber's leadership page, checked Nov 1, 2026, lists Praveen Neppalli Naga as CTO on the executive team, with no separate data or AI chief. Uber does not publish an org chart, so the lines from the AI platform, engineering productivity and cloud teams to the CTO are inferred from engineering blog posts that credit his leadership. Uber's Jun 2025 Form 8-K made the President and COO responsible for the Mobility, Delivery and Autonomous businesses and for customer support. Uber's 2025 Form 10-K says the Chief Privacy Officer, in the role since Nov 2025, and the CTO support the cybersecurity program.

Where the AI workforce sits

Engineering, science and AI platform teams
Uber does not publish engineering or AI headcount; research and development expense, mostly engineering, design and product pay, was $3.40 billion in 2025.
Not disclosed
Business teams building their own agents
Finance, legal, security, merchant operations and customer support teams have built agents on Uber's platforms; weekly active users of agentic tools, engineers and non-engineers, grew 7x from Feb to Aug 2026.
Not disclosed
Uber AI Solutions contributor network
Outside contributors and experts doing annotation and model training work (Mar 2026), not Uber employees.
10M+ contributors

Named people

Executive sponsors

Dara Khosrowshahi
Chief Executive Officer
Balaji Krishnamurthy
Chief Financial Officer · Reports AI spend and hiring effects on earnings calls

Engineering, AI platform and agents

Praveen Neppalli Naga
Chief Technology Officer
Sumanth Sukumar
Vice President of Engineering · Spoke for Uber on the ChatGPT Work rollout (Oct 2026); thanked for Software Factory leadership
Uday Kiran Medisetty
Distinguished Engineer, engineering productivity · Software Factory, agentic coding, code review
Alok Srivastava
Principal Engineer · Lead author on the MCP Gateway (Oct 2026)
Matt Mathew
Principal Engineer · Lead author on agent identity and authorization (May 2026)

Cloud and infrastructure

Kamran Zargahi
Vice President of Engineering, cloud engineering

Business functions

Frank Tenente
Senior Legal Director, Product Legal (Mobility & Platform) · Co-author on the Legal Redlining Agent (Oct 2026)
Austin Harrison
Director of Finance Data Analytics & AI, Uber FinTech · Named in the Finch post (Jul 2025)
Tiho Nedkov
Director of finance · Described Uber finance's AI use at a Gartner conference (May 2026)
Megha Yethadka
Vice President, Global Community Operations · Customer support; wrote the Jul 2026 memo on support cuts; named Head of Uber AI Solutions in Jun 2025

AI and autonomy businesses

Over 10 million outside contributors and experts on Uber AI Solutions (Mar 2026)

Dan Carpenter
General Manager, Uber AI Solutions
Sarfraz Maredia
Head of Autonomous Mobility & Delivery · Leaving Nov 16, 2026, with duties split between two senior directors

Use cases

Uber runs machine learning in pricing, matching, routing, arrival times and fraud, with about 400 active ML projects and more than 5,000 production models in May 2024, and since 2023 has added generative AI and agents in engineering, customer support, finance, legal, security, merchant tools, Uber Eats and Freight. Uber has published a business return for two named uses, AI coding tools and invoice processing. For AI as a whole, Uber gives no dollar figure; CFO Balaji Krishnamurthy says AI productivity gains are letting Uber moderate hiring (Aug 2026).

Use case	Area	Status	ROI
AI coding agents (Software Factory) Uber engineers use local and cloud coding agents, including Claude Code and Codex, for writing code, fixing failed builds, on-call triage and refactoring; more than 70% of pull requests were attributed to agents in Aug 2026.	Engineering	In production	Code output per engineer doubled on Uber's current measure, slowing engineering hiring (Aug 2026)Source
uReview AI code review uReview reviews code changes in all of Uber's monorepos for bugs, security issues and standards violations, covering over 90% of about 65,000 weekly changes in Aug 2025.	Engineering	In production
Genie on-call copilot Genie answers engineers' questions in Slack help channels with cited answers from internal documentation, and Uber was extending an improved version across its security and privacy channels in May 2025.	Engineering	In production
AI prototyping and design specs Uber product managers prototype with Lovable, Figma Make, Claude Code and Cursor, and nearly 40% of submissions to Uber's global hackathon used a prototyping tool (Apr 2026); the design systems team generates Figma specs with a Cursor agent (Mar 2026).	Product and design	In production
ChatGPT Work for employees Uber is rolling out OpenAI's ChatGPT Work agent to more than 30,000 employees for research, meeting briefs, presentations and data analysis across Google Workspace and Slack.	Firmwide	In production
Invoice processing Uber's TextSense platform reads supplier invoices in more than 25 languages with OpenAI GPT-4 models, extracts header fields and line items, and sends them to human reviewers before payment. An Uber finance director said in May 2026 that AI now processes more than 96% of Uber's invoices.	Finance	In production	Invoice processing cost 25-30% below the manual process, credited to GenAI (Apr 2025)
Finch finance data agent Finch is a Slack agent that turns finance questions into SQL against curated data marts, with role-based access to sensitive financial data, for Uber's analysts and accountants.	Finance	In production
Legal Redlining Agent The Legal Redlining Agent is a Microsoft Word add-in that reviews counterparties' contract edits against Uber policy and drafts responses for lawyers to approve; Uber reported over 20% shorter average review time (Oct 2026).	Legal	In production
File Semantic Analyzer Uber's File Semantic Analyzer uses a fine-tuned in-house language model to classify files leaving Uber by meaning, for data loss prevention, with human analysts checking findings before action.	Security	Pilot
Pricing, matching, routing and arrival times Machine learning on Michelangelo prices, dispatches and routes Uber trips and deliveries at more than 10 million predictions a second (Nov 2025).	Marketplace	In production
Fare dispute agent An Uber support agent calls mapping and GPS tools to reconstruct a trip and resolves eligible fare disputes end to end in English-speaking markets (Aug 2026).	Customer support	In production
Cart Assistant Cart Assistant on Uber Eats turns a prompt, recipe or photo into a draft grocery cart for the shopper to edit; Uber's Q2 2026 remarks said shoppers using it build baskets about twice the size of a typical grocery order.	Delivery	In production
Merchant Reasoning Agents and merchant tools Uber's Merchant Reasoning Agents cut false-positive store closures by more than 15% (Feb 2026), and since Nov 2025 Uber Eats has given merchants AI tools to analyze reviews, write menu descriptions and enhance food photos.	Delivery	In production
Driver and courier assistant Uber's Earner Assistant turns earnings trends and heatmaps into personalized guidance for drivers and couriers and handles questions and app settings; Uber rolled it out across the U.S. by May 2026.	Mobility and Delivery	In production
Voice Bookings Voice Bookings is a conversational assistant in the Uber app that takes a spoken request, understands destination and preferences, and offers ride options, announced at GO-GET in Apr 2026.	Mobility	Announced
ChatGPT app integrations Uber's integrations with ChatGPT let users discover rides and restaurants inside ChatGPT and complete checkout in Uber's apps (Feb 2026).	Mobility and Delivery	In production
Shipment risk and workflow automation Uber Freight uses AI to flag shipment risks earlier and to automate shipment inquiries, document checks and data corrections (Aug 2026).	Freight	In production
Data labeling and model training for outside clients Uber AI Solutions sells annotation, evaluation, data collection and agent training work to AI labs and enterprises, and Uber said in Nov 2025 that it would pilot AI data tasks, such as tagging photos, for drivers in the Uber Driver app.	AI Solutions	In production
Robotaxi data factory with NVIDIA Uber and NVIDIA are building a data factory on NVIDIA Cosmos to curate, label and simulate driving data for training autonomous driving models, with Uber contributing more than 3 million hours of robotaxi-specific data.	Autonomous Mobility & Delivery	Announced

Adoption

Uber spread AI in layers: machine learning in its marketplace since Michelangelo launched in 2016, a generative AI hackathon with 713 engineers in 2023, the GenAI Gateway serving close to 30 teams by Jul 2024, an agent platform from early 2025, and coding agents across engineering in 2026. Uber engineers moved from 95% monthly use of AI coding tools in May 2026 to near 100% adoption by Aug 2026, and weekly active users of Uber's agentic tools, engineers and non-engineers, grew 7x from Feb to Aug 2026. ChatGPT Work went from a 1,000-employee pilot to a company-wide rollout covering more than 30,000 employees on Oct 22, 2026, and Uber has not yet published company-wide weekly use.

By the numbers

Employees given ChatGPT Work access: 30,000+; Company-wide rollout from Oct 22, 2026, after a 1,000-employee pilot; no use figure yet

Pull requests attributed to AI agents: 70%+; Local or cloud agents, Uber engineering blog, Aug 27, 2026

Engineers using AI coding tools: ~100%; CFO Balaji Krishnamurthy, Aug 5, 2026; 95% used them monthly in May 2026

Real-time ML predictions: 10M+ a second; Pricing, dispatch and routing, Q3 2025 prepared remarks (Nov 2025)

Employees: ~34,000; Dec 31, 2025 (2025 Form 10-K), before a cut of about 10% announced Sep 2, 2026

Research and development expense (2025): $3.40B; 7% of revenue, up 9% on higher headcount costs; Uber publishes no separate AI budget

Rollout, first pilot to today

Early 2016
Uber launched Michelangelo to standardize machine learning workflows across the company.
Uber engineering blog
Feb 2023
Uber began moving from its own data centers to Oracle Cloud Infrastructure and Google Cloud.
Uber engineering blog
Aug 2023
Uber reported a generative AI hackathon in which 713 engineers built 98 working demos for products, developer productivity and business operations.
Uber engineering blog
May 2024
Uber said Michelangelo ran more than 5,000 production models on more than 5,000 GPUs across its own data centers, Oracle Cloud Infrastructure and Google Cloud, and had been extended for large language models.
Uber engineering blog
Jul 2024
Uber's GenAI Gateway served 16 million queries a month for close to 30 teams across more than 60 identified LLM use cases.
Uber engineering blog
Sep 2024
Uber's QueryGPT text-to-SQL tool reached about 300 daily active users in a limited release to operations and support teams.
Uber engineering blog
Early 2025
Uber built its internal agent platform, later adopted by thousands of internal agents.
Uber engineering blog
Aug 2025
Uber's uReview AI code reviewer covered over 90% of about 65,000 weekly code changes.
Uber engineering blog
Nov 2025
Uber named generative AI one of its six strategic focus areas and said machine learning makes more than 10 million predictions a second.
Q3 2025 prepared remarks
Feb 2026
Uber said it was piloting first-party agents for drivers, couriers and consumers and had launched ChatGPT integrations for Rides and Eats.
Q4 2025 prepared remarks
May 2026
Uber said 95% of its engineers used AI coding tools monthly and more than 10% of production-ready code was written autonomously by AI coding agents.
Q1 2026 prepared remarks
Aug 2026
Uber's CFO reported near 100% adoption of AI coding tools among Uber engineers.
Q2 2026 earnings call transcript
Aug 2026
Uber said more than 70% of pull requests were attributed to local or cloud agents and weekly active users of its agentic tools had grown 7x since Feb 2026.
Uber engineering blog
Oct 2026
Uber began rolling out ChatGPT Work to more than 30,000 employees, its first company-wide agentic platform for knowledge work.
Uber engineering blog

Economics & controls

Spend and ROI

Technology spend

Uber does not publish a technology or AI budget. Uber's research and development expense, mostly pay for engineering, design and product staff, rose 9% to $3.40 billion in 2025 from $3.11 billion in 2024, about 7% of revenue, driven by a $313 million rise in headcount costs (2025 Form 10-K). Uber's Q1 2026 prepared remarks said higher investment in technology headcount moderated profit flow-through at the corporate level in the first quarter.

What AI itself costs

Uber does not disclose its AI spending. CTO Praveen Neppalli Naga told The Information in Apr 2026 that Uber had already spent its 2026 budget for AI coding tools, driven by engineers' use of Claude Code. Balaji Krishnamurthy said on May 6, 2026 that Uber underestimated the impact of AI tools when it set its 2026 budgets in Nov 2025, that new models arrived in Dec 2025, and that Uber re-upped its AI investment while trading it against slower headcount growth. Bloomberg reported on Jun 2, 2026 that Uber capped each employee's spending at $1,500 a month per AI coding tool, with approval needed to go over. Uber's engineering blog said in Aug 2026 that total AI spend had broadly stabilized since Apr 2026, with cost per 1,000 model requests down almost 34% from its peak and cost per session down 52% from its Jun peak when the model is held constant.

Reported ROI

Uber has not put a dollar figure on AI's return. Balaji Krishnamurthy said on Aug 5, 2026 that, on the measure Uber currently tracks, code output per engineer has doubled with near-universal use of AI coding tools, which at a minimum let Uber moderate engineering hiring, and that Uber expects AI to raise the quality and lower the cost of customer support. Uber's finance team reported in Apr 2025 that generative AI invoice processing costs 25-30% less than the manual process. Andrew Macdonald, Uber's President and COO, said in a May 2026 podcast interview, reported by Fortune, that Uber could not yet draw a clear line from engineers' AI tool use to more useful features shipped to consumers.

Headcount

Uber had about 34,000 employees at Dec 31, 2025. Balaji Krishnamurthy said in May and Aug 2026 that AI productivity gains are letting Uber moderate hiring against its plan, and that separate organizational reviews cut headcount by about 10% to 20% in a couple of organizations in the second quarter of 2026. Uber said a cut of 23% of its People and Places division in Jun 2026 was unrelated to AI. Uber cut about 10% of its community operations team, its customer support arm, on Jul 22, 2026, with a spokesperson saying Uber would simplify operations and continue to embrace AI. Dara Khosrowshahi announced a cut of about 10% of Uber's staff on Sep 2, 2026, citing management layers, coordination roles and fragmented ownership, with no mention of AI.

AI as a business

Uber AI Solutions sells data labeling, evaluation and model training work to AI labs and enterprises, and Dara Khosrowshahi said in Nov 2025 that Uber AI Solutions was landing many customers and could become another profitable business line. Uber has not disclosed revenue for Uber AI Solutions.

Governance and risk

Board oversight

Uber's 2026 proxy assigns oversight of AI to the full board, which receives regular updates on Uber's use and governance of AI, including strategy, risk management and regulatory engagement. Uber's Chief Privacy Officer reports to the board at least annually on topics including AI governance risk, and Uber added Nikesh Arora to the board in May 2025 for his cybersecurity and AI expertise.

Published principles and audits

Uber released AI principles, launched a Responsible AI webpage and published an algorithmic audit by O'Neil Risk Consulting & Algorithmic Auditing in 2024, and in 2025 published its first European Algorithmic Transparency Report and a broader AI webpage, according to its 2026 proxy.

Data boundary for outside models

Uber's GenAI Gateway replaces personal data with placeholders before requests go to outside model vendors and restores it in responses, and Uber's Engineering Security team reviews each use case against Uber's data handling standard before granting access (Jul 2024). Uber's AI Gateway applies AI Guard protections against prompt injection, jailbreaks and unsafe content (May 2026), and Uber's File Semantic Analyzer screens files leaving the company by meaning.

Controls on agents

Uber gives every internal agent a short-lived, single-hop token minted by a central Security Token Service after checking an Agent Registry, records the full chain from the human user through each agent, and enforces tool-level policy at the MCP Gateway, mandatory for high-risk systems (May 2026). Every MCP tool starts disabled until its owning team turns it on, and changes to tool descriptions need owner approval (Oct 2026). Lawyers review every suggestion from the Legal Redlining Agent before it goes out.

Disclosed risks and regulation

Uber's 2025 Form 10-K warns that its growing use of AI could produce flawed, biased or discriminatory outcomes, that third-party AI may carry unclear intellectual property rights, and that laws such as the EU AI Act may require more investment. Uber's 2025 Form 10-K also warns that AI-enabled platforms and digital assistants could disintermediate Uber's relationship with consumers.

Watching AI costs

Uber caps each employee's spending at $1,500 a month per AI coding tool, with approval needed to exceed it (Jun 2026). Uber also sends engineers Slack alerts at 50%, 80% and 100% of their expected AI spend, flags 16 wasteful patterns on a session dashboard, and shows employees their own spend so they can manage it (Aug 2026).

Recent signals

Oct 22, 2026
Uber began rolling out OpenAI's ChatGPT Work to more than 30,000 employees after a 1,000-employee pilot, with GPT-6.1 Sol as the default model, while developers keep using Codex.
Uber engineering blog
Oct 13, 2026
Bloomberg reported that Sarfraz Maredia, who leads Uber's autonomous mobility and delivery business, will leave on Nov 16, 2026, with his duties split between two senior directors.
Bloomberg
Oct 8, 2026
Uber described its Legal Redlining Agent, a Word add-in built on the GenAI Gateway that cut average contract review time by over 20% and was part of Uber Legal's award-winning entry at ALM Legalweek in Mar 2026.
Uber engineering blog
Oct 1, 2026
Uber said its MCP Gateway hosts more than 800 MCP servers and 5,000 tools for AI agents, and that Code Mode, which routes tool calls through Uber's agent command line, is now the default for its coding agents.
Uber engineering blog
Sep 2, 2026
Dara Khosrowshahi told Uber staff the company would cut about 10% of its team to remove management layers and coordination roles, in a memo that does not mention AI.
Uber newsroom
Aug 27, 2026
Uber's engineering blog said more than 70% of pull requests are attributed to AI agents, weekly active users of agentic tools grew 7x from Feb to Aug 2026, and total AI spend has stabilized since April.
Uber engineering blog
Aug 5, 2026
Balaji Krishnamurthy said Uber engineers' adoption of AI coding tools is near 100%, code output per engineer has doubled on Uber's measure, and Uber is moving some AI work to cheaper or open-weight models to keep AI spend broadly stable.
Q2 2026 earnings call transcript
Jul 22, 2026
Uber cut about 10% of its community operations team, which handles customer support, in a memo from Megha Yethadka, with a spokesperson saying Uber would continue to embrace AI.
HCAMag, citing Bloomberg
Jul 1, 2026
Bloomberg reported that a senior engineering director and a product director at Uber AI Solutions had left, part of what Uber called a broader leadership transition at the division.
Bloomberg
Jun 2, 2026
Bloomberg reported that Uber capped employee spending at $1,500 a month per AI coding tool, after The Information reported in Apr 2026 that Uber had used up its 2026 budget for AI coding tools.
PYMNTS, citing Bloomberg
May 29, 2026
Tiho Nedkov, an Uber director of finance, said more than 90% of Uber's finance staff routinely use AI tools and AI processes more than 96% of invoices, and that engineering has overtaken finance as Uber's leading AI adopter.
CFO Dive
May 26, 2026
Fortune reported COO Andrew Macdonald's comment, on the Rapid Response podcast, that the link between Uber's AI tool use and more useful features shipped is not there yet.
Fortune
May 6, 2026
Uber said 95% of engineers use AI coding tools monthly and more than 10% of code is written autonomously by agents, and its CFO said Uber re-upped its 2026 AI budget while slowing hiring.
Q1 2026 earnings call transcript
Apr 7, 2026
AWS said Uber is moving more trip-serving workloads to Graviton4 and has begun pilot training some AI models, including dispatch and arrival-time models, on AWS Trainium3.
Amazon press release
Feb 4, 2026
Uber said it was piloting first-party AI agents for drivers, couriers and consumers, that Merchant Reasoning Agents cut false-positive store closures by more than 15%, and that it had launched ChatGPT integrations for Rides and Eats.
Q4 2025 prepared remarks

Quick answers

How does Uber deploy AI?: Uber deploys AI through Michelangelo, the machine learning platform it built in 2016 that runs pricing, matching and routing at about 10 million predictions a second, and through gateways it built to connect its applications and agents to OpenAI, Anthropic and open-weight models. Engineering has gone furthest: nearly all Uber engineers use AI coding tools, and more than 70% of pull requests were attributed to AI agents by Aug 2026. Uber began rolling out OpenAI's ChatGPT Work agent to more than 30,000 employees on Oct 22, 2026.
What does Uber use AI for?: Uber runs machine learning in pricing, matching, routing, arrival times and fraud, with about 400 active ML projects and more than 5,000 production models in May 2024, and since 2023 has added generative AI and agents in engineering, customer support, finance, legal, security, merchant tools, Uber Eats and Freight. Uber has published a business return for two named uses, AI coding tools and invoice processing. For AI as a whole, Uber gives no dollar figure; CFO Balaji Krishnamurthy says AI productivity gains are letting Uber moderate hiring (Aug 2026).
Who leads AI at Uber?: AI at Uber is led by Praveen Neppalli Naga (Chief Technology Officer); Johannes Gehrke (Technical fellow, AI (joined from Microsoft in 2026)); Uday Kiran Medisetty (Distinguished Engineer, engineering productivity). Ownership of AI on the org chart: CTO Praveen Neppalli Naga, who leads engineering and science.
What AI platform does Uber use?: Michelangelo, Uber's in-house machine learning platform since 2016, plus an AI Gateway to outside and self-hosted models, an agent platform built in early 2025 and an MCP Gateway with more than 800 servers (Oct 2026).
Which AI models does Uber use?: Uber routes calls to OpenAI and Anthropic models through its AI Gateway, has fine-tuned open-weight Llama 2 models in-house, and is moving some tasks to cheaper open-weight models to hold AI spend flat (Aug 2026).
How many Uber employees use AI?: Near 100% of Uber engineers use AI coding tools (Aug 2026), and Uber began rolling out ChatGPT Work to more than 30,000 employees on Oct 22, 2026, after a 1,000-person pilot.
How far has AI spread at Uber?: Uber sits at Scaling: machine learning runs its core marketplace at about 10 million predictions a second, nearly all engineers use AI coding tools, thousands of internal agents run on its agent platform, and CTO Praveen Neppalli Naga owns engineering and science while the CFO reported AI's effect on hiring on the May and Aug 2026 earnings calls. Uber falls short of Embedded because ChatGPT Work only opened to its more than 30,000 employees on Oct 22, 2026, with no company-wide weekly use figure published, and Uber gives no dollar figure for AI's financial effect.

Open questions

How much does Uber spend on AI tools and tokens in 2026 after spending its original coding-tool budget by April and re-upping it, and how is that spend split between OpenAI, Anthropic and open-weight models?
How does Uber measure code output per engineer, and how many engineering hires did Uber skip because of the doubling its CFO reported?
What weekly use will ChatGPT Work reach across Uber's more than 30,000 employees, and what outcomes will Uber and OpenAI publish?
Who leads Uber's AI platform and agent platform teams today, and where do they report after the Sep 2026 reorganization?
How many AI and machine learning engineers and scientists does Uber employ?
What revenue does Uber AI Solutions earn, and is it profitable yet?
What business return do Uber's deployed agents deliver where none is published, such as the fare dispute agent, Cart Assistant, uReview, Finch, Genie and the Legal Redlining Agent?
How much of Uber's AI training and inference has moved from its own data centers to Oracle Cloud Infrastructure, Google Cloud and AWS, and will the Trainium pilot move into production?
Did AI productivity play any part in the roles cut in Sep 2026, given that Dara Khosrowshahi's memo cites only management layers and coordination?
How much of Uber's customer support work do AI agents now handle, and what did the Jul 2026 support cuts save?
Will Uber's Q3 2026 results on Nov 4, 2026 update its AI adoption, spend or hiring figures?

Sources

Uber

Uber engineering blog

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.

Uber profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 1, 2026

More company profiles

All profiles