Verified
Alphabet
Scaling- 204,757 staff
- Gemini on its own TPUs
- ~75% of new code AI-generated
- AI sales tools: $100Ms of revenue (Oct 2025)
Alphabet deploys AI inside Google mostly with tools it builds: Gemini models on its own TPUs, the Antigravity agent platform and internal coding agents such as Agent Smith, with Anthropic's Claude Opus 5 added for engineers on a quota in Sep 2026. Sundar Pichai said in Apr 2026 that about 75% of new code at Google is AI-generated and approved by engineers. Philipp Schindler said in Oct 2025 that Gemini tools for ad sellers raised their productivity over 10% and added hundreds of millions of dollars in revenue.
abc.xyzTheir approach to AI
- The platform
- Alphabet builds its internal AI tools itself: Gemini models, the Antigravity agentic development platform, coding tools such as Cider, Jetski and Agent Smith, and a Gemini chatbot trained on Google's own documentation; Koray Kavukcuoglu was uniting the coding tools under Antigravity in Apr 2026.
- The models
- Google says Gemini remains the primary model for internal development; since Sep 2026 every engineer can use Anthropic's Claude Opus 5 inside Antigravity on a per-user quota, while Claude Code and OpenAI's Codex stay barred for most staff (Business Insider, Bloomberg).
- The reach
- Google reports adoption by group, not company-wide: about 75% of new code was AI-generated in Apr 2026, and 83% of the sales team used Gemini-assisted tools weekly in Jul 2026; Alphabet had 204,757 employees on Sep 30, 2026.
- The catch
- Alphabet credits AI with returns in ad sales and engineering, but reports no firm-level AI figure, no company-wide count of weekly AI users, and headcount that grew 7.7% in the year to Sep 2026.
Snapshot
- Employees
- 204,757
- Sep 30, 2026, up from 190,167 a year earlier (Q3 2026 earnings release)
- Capital expenditure
- $91.4B (2025)
- 2026 guidance $195B to $205B, mostly AI technical infrastructure
- AI/ML headcount
- Not disclosed
- R&D expense was $61.1B in 2025, driven by compensation and depreciation (2025 Form 10-K)
- Who owns AI
- Sundar Pichai, CEO
- Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect, unifies internal coding tools
- Platform
- Antigravity and in-house tools
- Cider, Jetski, Agent Smith and an internal Gemini chatbot
- Models
- Gemini, plus Claude on a quota
- Claude Opus 5 inside Antigravity for every engineer since Sep 2026
- Reach
- ~75% of new code AI-generated
- Apr 2026; 83% of the sales team uses Gemini tools weekly (Jul 2026)
- Use cases
- Engineering, sales, support, safety, security, finance
- No company-wide count published
Key findings
Alphabet credits AI with 2 business returns reported since May 2025: Gemini tools for ad sellers raised productivity over 10% and added hundreds of millions of dollars in revenue (Oct 2025), and AI raised engineering velocity an estimated 10% (Jun 2025); an AlphaEvolve heuristic also recovers about 0.7% of Google's worldwide compute (May 2025), a capacity gain with no dollar value.
The share of new code at Google generated by AI and approved by engineers rose from more than a quarter in Oct 2024 to about half in fall 2025 and about 75% by Apr 22, 2026, and Sundar Pichai said in Oct 2026 that teams using Gemini 4 Argon internally ship 35% more code with 30% fewer errors.
Google keeps most staff on its own Gemini models and tools: Bloomberg reported in Apr 2026 that most employees are barred from Claude Code and Codex for security reasons, and Business Insider reported in Sep 2026 that every engineer can now use Anthropic's Claude Opus 5 inside Antigravity on a per-user quota.
Koray Kavukcuoglu, named Google's first Chief AI Architect in Jun 2025, became SVP of Google DeepMind on Aug 5, 2026 reporting to Sundar Pichai, while Demis Hassabis became chair of Google DeepMind and Alphabet's chief scientist.
Alphabet's headcount rose 7.7% to 204,757 in the year to Sep 30, 2026, and Alphabet has not credited any job cuts to AI, though Anat Ashkenazi cites AI-generated code, fewer management layers and a voluntary exit program among its efficiency levers (Mar 2026).
Where Google sits
- Stage
- Scaling
Alphabet credits AI with business returns in 2 functions, both reported in the last 18 months, which places it at Scaling: Philipp Schindler said on Oct 29, 2025 that Gemini tools for Google's ad sales teams raised seller productivity by over 10% and led to hundreds of millions of dollars in incremental revenue; Google estimates that AI has raised engineering velocity about 10% across the company (Sundar Pichai, Jun 2025, restated by Google in Aug 2025). Google DeepMind's May 2025 report that an AlphaEvolve scheduling heuristic recovers about 0.7% of Google's worldwide compute is a capacity measure, not a currency return. On rollout, Google reports adoption by group: about 75% of new code is AI-generated (Apr 2026) and 83% of the sales team uses Gemini-assisted tools weekly (Jul 2026), while Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect, is uniting the internal coding tools under Antigravity, and Pichai and Schindler report results on earnings calls. Alphabet falls short of Embedded because Alphabet reports no firm-level cost, revenue or headcount effect it credits to AI, has not said how many of its 204,757 employees use AI each week, and grew headcount 7.7% in the year to Sep 30, 2026.
How the stages are defined - Ownership
- CEO Sundar Pichai; Koray Kavukcuoglu, Chief AI Architect, unifies internal coding tools
- Platform
- Built: Gemini, Antigravity and internal coding agents, built in-house
- Models
- Gemini by default; Claude Opus 5 on a quota inside Antigravity
- Hosting
- Google's own data centers, predominantly on its TPUs, plus some leased capacity; Claude hosting for Antigravity not disclosed
- Reach
- ~75% of new code AI-generated; 83% of sales staff weekly
How they deploy
Customer zero for its own models
Sundar Pichai wrote in Apr 2026 that Google always wants to be customer zero for its own technologies, and Alphabet runs its internal AI on the Gemini models, TPUs and agent tools it sells. A Google spokesperson said in Sep 2026 that Gemini remains the primary foundation model for internal development and that third-party models are offered on a quota for specialized use cases.
From code completion to agents
Google moved from an internal coding model, Goose, in Feb 2024 to AI that wrote more than a quarter of new code in Oct 2024 and about 75% in Apr 2026. Sundar Pichai said in Apr 2026 that Google is shifting to agentic workflows in which engineers orchestrate autonomous task forces of agents, and Sergey Brin wrote in an internal memo reported by The Information in Apr 2026 that Google must turn its models into primary developers.
One coding platform after fragmentation
Bloomberg reported in Apr 2026 that Gemini's coding capabilities were spread across half a dozen differently branded products and that some Google engineers preferred Anthropic's Claude Code. Koray Kavukcuoglu, Google's Chief AI Architect, is working with Google's main engineering team to unite the internal AI coding tools under Antigravity, and Google DeepMind formed a team led by Sebastian Borgeaud to build better coding models.
Productivity over headcount
Sundar Pichai told a Jul 2025 all-hands meeting that in this AI moment Google has to accomplish more by driving higher productivity instead of adding headcount, and Brian Saluzzo, then head of Google's core developer team, said Google felt urgency to get AI into more coding workflows. Pichai said in Jun 2025 that Google would still hire more engineers because the opportunity space is expanding.
Adoption as a job expectation
Google has moved AI use from encouraged to expected: engineering VP Megan Kacholia emailed engineers in Jun 2025 asking them to use AI for coding, Business Insider reported in Feb 2026 that non-technical staff were told AI use would count in reviews, and a Google spokesperson said managers have discretion to evaluate employees on their use of AI.
- Deployment model
- Alphabet deploys AI through central model and infrastructure groups with adoption pushed into every function. Google DeepMind, led since Aug 2026 by Koray Kavukcuoglu, builds the Gemini models and the developer teams' tools; the AI and Infrastructure organization under Amin Vahdat runs TPUs and data centers and started Project EAT in May 2025 to standardize AI tools inside that unit before a company-wide rollout; Google's core developer engineering team builds internal coding tools such as Cider. Business functions build their own agents on these models: the sales and ads support teams under Philipp Schindler, Trust & Safety, security operations, and finance, where Anat Ashkenazi described a treasury agent in Mar 2026.
- Model providers
- Alphabet runs its internal AI on its own Gemini models, including Gemini 4 Argon and Carbon, which Sundar Pichai said in Oct 2026 Google is deploying internally, and on internal coding models trained on Google's proprietary codebase that cannot be released publicly (The Information, Apr 2026). Google's only reported outside model is Anthropic's Claude: some Google DeepMind teams used Claude Code by Apr 2026, and since Sep 2026 every engineer can select Claude Opus 5 inside Antigravity on a per-user quota, while direct use of Claude Code and OpenAI's Codex stays barred for most staff.
- Data and infrastructure
- Alphabet runs internal AI in data centers it mostly builds itself, with some leased capacity (Anat Ashkenazi, Feb 2026), and Ashkenazi said in Mar 2026 that Alphabet uses predominantly TPUs internally, alongside the NVIDIA GPUs Google Cloud offers. Sundar Pichai said in Apr 2026 that just over half of Alphabet's 2026 machine learning compute investment would go to the Cloud business, leaving the rest for Google's own services and research, and Bloomberg reported that engineers using internal coding tools often hit capacity limits. Google restricts staff to approved internal AI systems partly to protect sensitive information (Business Insider, Feb 2026). Google has not said where the Claude models offered in Antigravity are hosted.
Platforms and internal tools
Gemini models
Gemini is Google DeepMind's family of models and the primary foundation model for Google's internal development, according to a Google spokesperson in Sep 2026. Sundar Pichai said on Oct 28, 2026 that Google began deploying Gemini 4 Argon internally about 4 weeks earlier and is now deploying Gemini 4 Carbon internally.
- Models
- Gemini 4 Argon, Gemini 4 Carbon and earlier Gemini models
AntigravityNov 2025
Antigravity is Google's agentic development platform, where agents plan and execute software tasks; Google sells it externally and uses it internally, and Koray Kavukcuoglu is uniting Google's internal AI coding tools under it (Apr 2026). Google's Gemini app for macOS was built with Antigravity, going from idea to a native Swift prototype in a few days (Apr 2026).
- Users
- More than 2.4 million weekly active users (Jul 2026)
- Models
- Gemini; Claude Opus 5 on a quota for Google engineers since Sep 2026
Agent SmithEarly 2026
Agent Smith is an internal agent built on Antigravity that works in the background on coding and other tasks, reaches internal tools and documents, and takes instructions from employees' phones. Business Insider reported in Mar 2026 that access was restricted because of demand.
CiderMay 2025
Cider is an internal AI coding tool that helps Google software engineers across the development process; Brian Saluzzo told staff in Jul 2025 that 50% of its users used it weekly.
Jetski
Jetski is an internal Google AI coding tool; The Information reported in Apr 2026 that Google tracks employees' use of internal AI coding tools, runs mandatory training sessions and requires their use for certain tasks.
Goose and the internal Gemini chatbotFeb 2024
Goose is an employee-only model derived from Gemini and trained on 25 years of Google engineering knowledge to answer questions about internal technology and write code on Google's stack (Business Insider, Feb 2024). Google staff also use a customized Gemini chatbot trained on company documentation (Business Insider, Feb 2026).
- Models
- Gemini-derived
AlphaEvolveDeployed internally from about 2024; announced May 2025
AlphaEvolve is a Gemini-powered coding agent from Google DeepMind that designs algorithms; Google has deployed its results in data center scheduling, TPU circuit design and Gemini training kernels.
- Models
- Gemini
Security agents and CodeMender
Google's security teams run agents that summarize tickets and analyze malicious code (Oct 2025); Security Operations Center agents triage tens of thousands of unstructured threat reports a month, and Gemini-based agents such as CodeMender find and fix software flaws (Apr 2026).
Vendors and partners
AnthropicSep 2026
Claude Opus 5 for Google engineers inside Antigravity, on a per-user quota · Some Google DeepMind teams used Claude Code by Apr 2026 (Bloomberg); Google has invested billions of dollars in Anthropic.
WindsurfJul 2025
Talent and a nonexclusive technology license that preceded Antigravity · Google paid $2.4 billion to bring in Windsurf's chief executive and other staff and license its AI coding technology (CNBC, Jul 2025; Bloomberg, Apr 2026).
NVIDIA
GPUs alongside Google's own TPUs in its AI infrastructure · Anat Ashkenazi said in Mar 2026 that Alphabet uses predominantly TPUs internally.
Yoodli
AI avatar tool adapted for Google Cloud sales teams to rehearse pitches · Reported by Business Insider in Feb 2026.
Leadership
AI, data and platform leaders
Executives
Board
AI org map
Central model and infrastructure groups; adoption pushed into every function
Alphabet has no single internal AI office. Google DeepMind, led since Aug 5, 2026 by Koray Kavukcuoglu, who reports to Sundar Pichai, builds the Gemini models and is uniting the internal coding tools under Antigravity; the AI and Infrastructure organization under Amin Vahdat, also reporting to Pichai, runs the compute; and Google's core developer engineering team builds tools such as Cider. Business functions such as sales, ads support, Trust & Safety, security and finance build their own agents on these models.
Alphabet does not disclose how many of its 204,757 employees work on AI. Google's AI work sits in Google DeepMind, which builds the Gemini models and developer tools under Koray Kavukcuoglu; the AI and Infrastructure organization under Amin Vahdat, which runs TPUs and data centers; and the core developer engineering team, which builds internal coding tools such as Cider. Bloomberg reported in Apr 2026 that Google DeepMind, Google Cloud, Google Core, Google Labs and Android were all pushing AI coding in different ways.
Reporting lines
CEO, Alphabet and Google
Google DeepMind
Koray Kavukcuoglu, SVP and Chief AI Architect
Model development, AI research, Gemini and developer teams (Aug 2026); Demis Hassabis chairs Google DeepMind
DeepMind AI coding team
Sebastian Borgeaud
Formed in 2026 to build coding models for internal use (The Information, Bloomberg)
AI and Infrastructure (AI2)
Amin Vahdat, SVP and Chief Technologist
Data centers, TPUs and other AI infrastructure; home of Project EAT
Core developer engineering
Builds internal coding tools such as Cider inside the Google Core organization; Brian Saluzzo led it until early 2026, and the current leader and reporting line are not published
Global business organization
Philipp Schindler, Chief Business Officer
Sales teams and ads customer support using Gemini tools and agents
Finance
Anat Ashkenazi, CFO
Treasury agent across cash balances (Mar 2026)
Alphabet has not published who owns AI adoption outside engineering, or the reporting line of the core developer team after Brian Saluzzo left Google for Citi in Mar 2026. Bloomberg reported in Apr 2026 that confusion persisted over who leads AI coding inside Google.
Named people
Use cases
Alphabet has described about a dozen internal AI use cases across engineering, sales, ads support, Trust & Safety, security, marketing, recruiting and finance, without publishing a count. Alphabet credits AI with 2 returns: Gemini tools for ad sellers (hundreds of millions of dollars of incremental revenue, Oct 2025) and AI in software engineering (an estimated 10% rise in engineering velocity, Jun 2025). AlphaEvolve in data center scheduling recovers about 0.7% of worldwide compute (May 2025), an operating measure with no dollar value. Alphabet reports no firm-level return from its internal AI use; its other figures, such as 75% of ads support queries handled by agents, measure workload or speed.
| Use case | Area | Status | ROI |
|---|---|---|---|
| Gemini tools for ad sales teams Google's sales teams use Gemini enriched with ads knowledge to prepare customer interactions and pitch narratives. 83% of the sales team used Gemini-assisted tools weekly in Jul 2026, with up to a 20% higher win rate when using customized pitch narratives, and some sellers have informal weekly usage quotas for call-recording and note-taking tools (Business Insider, Feb 2026). | Global business organization | In production | Hundreds of millions of dollars of incremental revenue, seller productivity up 10%+ (Oct 2025)Source |
| AI coding tools and agents Google engineers write code with Cider, Jetski, Antigravity and the Agent Smith background agent, and about 75% of new code was AI-generated and approved by engineers in Apr 2026. Google reported a code migration completed 6 times faster than engineers alone managed a year earlier (Apr 2026), and Sundar Pichai said in Oct 2026 that teams in Chrome, Ads and Search using Gemini 4 Argon ship 35% more code with 30% fewer errors. | Engineering | In production | Engineering velocity up an estimated 10% from AI, company-wide (Jun 2025)Source |
| AlphaEvolve in data center scheduling and model training AlphaEvolve found a heuristic that helps Borg schedule work across Google's data centers, proposed a circuit change adopted in an upcoming TPU, and sped up a Gemini training kernel by 23%, cutting Gemini's training time by 1%. Google DeepMind said in May 2025 that the scheduling heuristic, in production for over a year, recovers about 0.7% of Google's worldwide compute resources on average. | AI and Infrastructure | In production | |
| Agentic ads customer support Google's ads customer support teams use Gemini-powered agents that autonomously handle 75% of support queries (Jul 2026); Google's customer support division had used Gemini-powered solutions in over 40 million customer sessions in 2025 by Oct 2025. | Global business organization | In production | |
| Gemini in ads policy enforcement Google's 2025 Ads Safety Report says Gemini-based models helped block or remove 8.3 billion ads in 2025, cut incorrect advertiser suspensions by 80%, and let teams act on more than 4 times as many user reports (Apr 2026). | Trust & Safety | In production | |
| Security operations agents and CodeMender Google's Security Operations Center agents triage tens of thousands of unstructured threat reports a month, which Google says cut threat mitigation time by more than 90%, and Gemini-based agents such as CodeMender find and fix critical software flaws (Apr 2026). | Security | In production | |
| Bug triage agent An AI agent automatically handles 12% of duplicate bugs at Google, and AI models help engineers find the source of problems faster (Aug 2025). | Engineering | In production | |
| Creative asset generation Google's marketing teams used Gemini models to generate thousands of creative variations for the Gemini in Chrome launch, with 70% faster turnaround and 20% more conversions (Apr 2026), and Imagen produced 48% of the visuals in the Google I/O 2025 keynote slides. | Marketing | In production | |
| RFP responses and lead filtering An internal AI tool contributed to a 78% year-over-year rise in completed Google Cloud requests for proposals, and an AI lead-filtering tool raised lead-to-opportunity conversion 14% over a 6-week period (Aug 2025). | Google Cloud sales | In production | |
| Treasury agent Anat Ashkenazi said in Mar 2026 that Alphabet's finance team runs a treasury agent that works across the company's cash balances to maximize the return on them, and that it is yielding real financial results, without giving a figure. | Finance | In production | |
| Recruiting and employee feedback Google's People Operations team uses AI to search its candidate pool and match skills to open roles, with a recruiter leading the process, and NotebookLM summarizes themes from thousands of internal survey responses (Aug 2025). | People Operations | In production |
Adoption
Google spread AI first through engineering: an internal coding model, Goose, in Feb 2024, AI writing more than a quarter of new code by Oct 2024, the Cider coding tool from May 2025, Antigravity from Nov 2025 and background agents such as Agent Smith in early 2026, with the AI-generated share of new code reaching about 75% in Apr 2026. Use then became an expectation: Sundar Pichai pressed staff on productivity in Jul 2025, Google launched the AI Savvy Google training site, and by Feb 2026 non-technical staff in sales and strategy were told AI use would count in reviews. Google reports adoption by group, such as 83% of the sales team using Gemini tools weekly (Jul 2026) and 50% of Cider's users using it weekly (Jul 2025), but has published no count of weekly AI users across its 204,757 employees. Business Insider reported in Mar 2026 that Agent Smith access was restricted because of demand, while 404 Media reported in Jun 2026 that some engineers mock the internal coding tools.
By the numbers
Rollout, first pilot to today
Feb 2024
Business Insider reported that Google had launched Goose, an employee-only model derived from Gemini and trained on 25 years of Google engineering knowledge.
Oct 2024
Sundar Pichai said more than a quarter of all new code at Google was generated by AI, then reviewed and accepted by engineers.
May 2025
Google DeepMind disclosed that an AlphaEvolve heuristic had run in Google's data center scheduling for over a year, recovering about 0.7% of worldwide compute.
May 2025
Staff in Google's AI and Infrastructure organization started Project EAT to help employees adopt AI products and standardize their use.
Jun 2025
Sundar Pichai said Google estimates AI has raised engineering velocity about 10% across the company and that Google still plans to hire more engineers.
Jun 2025
Google named Koray Kavukcuoglu its first Chief AI Architect, a new SVP role reporting to Sundar Pichai, to bring its models into products faster.
Jul 2025
Sundar Pichai told an all-hands meeting that Google must drive higher productivity, and Brian Saluzzo promoted the AI Savvy Google training site and said 50% of Cider's users used it weekly.
Oct 2025
Philipp Schindler said Gemini tools for Google's sales teams raised productivity by over 10% and led to hundreds of millions of dollars in incremental revenue, and Anat Ashkenazi said nearly half of Google's code was generated by AI.
Nov 2025
Google launched Antigravity, its agentic development platform, which had more than 1.5 million weekly active users 2 months later.
Feb 2026
Anat Ashkenazi said about 50% of Google's code is written by coding agents and reviewed by engineers.
Mar 2026
Business Insider reported that Google employees were using Agent Smith, an internal background agent built on Antigravity, and that access had to be restricted because of demand.
Apr 2026
Sundar Pichai said about 75% of all new code at Google is AI-generated and approved by engineers, up from 50% in fall 2025, and that engineers now orchestrate autonomous agents.
Jul 2026
Google said 83% of its sales team uses Gemini-assisted tools weekly and that agents handle 75% of ads support queries.
Sep 2026
Business Insider reported that every Google engineer can now use Anthropic's Claude Opus 5 inside Antigravity on a per-user quota.
Oct 2026
Sundar Pichai said Google is deploying Gemini 4 Argon and Carbon internally and expects to make them widely available to Googlers with more compute.
Economics & controls
Spend and ROI
Capital spending
Alphabet spent $91.4 billion on capital expenditure in 2025, mostly technical infrastructure, and kept its 2026 guidance at $195 billion to $205 billion on Oct 28, 2026, with a significant further increase expected in 2027. Alphabet does not split out what its internal AI use costs; Sundar Pichai said in Apr 2026 that just over half of the 2026 machine learning compute investment would go to Google Cloud, and Amin Vahdat told staff in Nov 2025 that Google must double its AI serving capacity every 6 months.
Compute for internal tools
Alphabet rations compute for internal AI: Bloomberg reported in Apr 2026 that engineers using internal coding tools often hit capacity limits, Business Insider reported in Mar 2026 that Agent Smith access was restricted to handle demand, and Claude Opus 5 is offered to engineers on per-user quotas (Sep 2026). Google said it cut Gemini serving unit costs by 78% over 2025 through model optimization, efficiency and utilization improvements (Feb 2026).
Reported ROI
Alphabet has published 2 returns it credits to internal AI: Philipp Schindler said on Oct 29, 2025 that Gemini tools for Google's sales teams raised productivity by over 10% and led to hundreds of millions of dollars in incremental revenue, and Google estimates AI raised engineering velocity about 10% (Jun 2025). An AlphaEvolve scheduling heuristic recovers about 0.7% of Google's worldwide compute (May 2025), a capacity gain Google has not valued in dollars. Alphabet gives no firm-level figure, and Anat Ashkenazi said of the treasury agent only that it yields real financial results (Mar 2026).
Productivity markers
Google reports AI's spread mostly as engineering output and speed: about 75% of new code AI-generated (Apr 2026), a code migration 6 times faster than a year earlier (Apr 2026), a Chrome team on track to compress a 2-year timeline into 3 months (Jul 2026), and teams using Gemini 4 Argon shipping 35% more code with 30% fewer errors (Oct 2026). Sundar Pichai said in Jun 2025 that engineering velocity, not the share of code, is the metric Google measures most carefully.
Headcount
Alphabet had 190,820 employees at Dec 31, 2025 and 204,757 at Sep 30, 2026, up 7.7% from 190,167 a year earlier. Anat Ashkenazi said in Mar 2026 that Alphabet reduced management layers, ran its first company-wide voluntary exit program in 2025 and uses AI internally to drive efficiency, and Alphabet has not credited any job cuts to AI; Sundar Pichai said in Jun 2025 that Google planned to hire more engineers.
Governance and risk
AI Principles and launch review
Google's 2026 Responsible AI Progress Report (Feb 2026) says Google applies its AI Principles through launch reviews for frontier models at Google DeepMind's Launch Review forum and for applications through centralized expert risk reviews, alongside its Frontier Safety Framework and Secure AI Framework. An AGI Futures Council of Google senior management and Alphabet directors advises the board and management on the long-term risks of AGI.
Approved tools and the data boundary
Google restricts staff to approved internal AI systems, partly to protect sensitive corporate information (Business Insider, Feb 2026). Bloomberg reported in Apr 2026 that most employees are banned from Claude Code and Codex for security reasons and may request exceptions with a business case, and Google's internal coding models are trained on its proprietary codebase and cannot be released publicly (The Information, Apr 2026).
Outside models on a quota
Google opened Anthropic's Claude Opus 5 to every engineer in Sep 2026 only inside Antigravity and on per-user quotas; a Google spokesperson said Gemini remains the primary foundation model for internal development.
AI use in performance reviews
A Google spokesperson said in Aug 2025 that Google was not factoring AI use into performance reviews, but by Feb 2026 a spokesperson told Business Insider that managers in technical and non-technical functions have discretion to evaluate employees on their use of AI, and Google has added AI use to the job requirements of certain software engineers.
Disclosed risks
Alphabet's 2025 Form 10-K says increased use of AI in its offerings and internal systems may create new problematic content and potential for misuse, and that AI may increase its exposure to cyber attacks. Google's security teams say they track agents on risk reduction, removal of repetitive work and analysts' trust in the tools (Oct 2025).
Recent signals
Oct 28, 2026
Sundar Pichai said teams in Chrome, Ads and Search using Gemini 4 Argon internally ship 35% more code with 30% fewer errors, that Google is now deploying Gemini 4 Carbon internally, and that AI is raising Google's own training and launch velocity.
Oct 28, 2026
Alphabet reported 204,757 employees at Sep 30, 2026, up from 190,167 a year earlier, and kept its 2026 capital expenditure guidance at $195 billion to $205 billion.
Sep 15, 2026
Business Insider reported that Google cleared every engineer to use Anthropic's Claude Opus 5 inside Antigravity on per-user quotas, while Claude Code and Codex stay off limits.
Aug 5, 2026
Sundar Pichai made Koray Kavukcuoglu SVP of Google DeepMind, responsible for models, Gemini and developer teams, and Demis Hassabis chair of Google DeepMind and Alphabet's chief scientist.
Jul 22, 2026
Google said 83% of its sales team uses Gemini-assisted tools weekly, agents autonomously handle 75% of ads support queries, and a Chrome team is on track to deliver 8 times faster through model-driven refactoring.
Jun 4, 2026
404 Media reported that Google engineers share internal memes mocking the AI coding tools they are told to use.
Apr 22, 2026
Sundar Pichai said about 75% of all new code at Google is AI-generated and approved by engineers, up from 50% in fall 2025, and that a complex code migration finished 6 times faster than a year earlier.
Apr 22, 2026
Bloomberg reported that Koray Kavukcuoglu is uniting Google's internal AI coding tools under Antigravity, that some engineers prefer Claude Code, and that Brian Saluzzo, who led internal AI coding adoption, had left.
Apr 16, 2026
Google's 2025 Ads Safety Report credited Gemini-based enforcement with blocking or removing 8.3 billion ads in 2025 and cutting incorrect advertiser suspensions by 80%.
Mar 27, 2026
Business Insider reported that Google employees use Agent Smith, an internal background agent built on Antigravity, and that access was restricted because of demand.
Mar 3, 2026
Anat Ashkenazi said Alphabet uses predominantly TPUs internally, runs a treasury agent that yields real financial results, and ran its first company-wide voluntary exit program in 2025.
Feb 27, 2026
Business Insider reported that Google told some non-technical staff AI use would factor into performance reviews, and a spokesperson said managers have discretion to evaluate it.
Feb 4, 2026
Anat Ashkenazi said about 50% of Google's code is written by coding agents and reviewed by engineers, and Google said it cut Gemini serving unit costs 78% over 2025.
Jan 30, 2026
Business Insider reported on Project EAT, started in May 2025 inside Amin Vahdat's AI and Infrastructure organization to make Google an AI-powered workplace.
Quick answers
- How does Alphabet deploy AI?
- Alphabet deploys AI inside Google mostly with tools it builds: Gemini models on its own TPUs, the Antigravity agent platform and internal coding agents such as Agent Smith, with Anthropic's Claude Opus 5 added for engineers on a quota in Sep 2026. Sundar Pichai said in Apr 2026 that about 75% of new code at Google is AI-generated and approved by engineers. Philipp Schindler said in Oct 2025 that Gemini tools for ad sellers raised their productivity over 10% and added hundreds of millions of dollars in revenue.
- What does Alphabet use AI for?
- Alphabet has described about a dozen internal AI use cases across engineering, sales, ads support, Trust & Safety, security, marketing, recruiting and finance, without publishing a count. Alphabet credits AI with 2 returns: Gemini tools for ad sellers (hundreds of millions of dollars of incremental revenue, Oct 2025) and AI in software engineering (an estimated 10% rise in engineering velocity, Jun 2025). AlphaEvolve in data center scheduling recovers about 0.7% of worldwide compute (May 2025), an operating measure with no dollar value. Alphabet reports no firm-level return from its internal AI use; its other figures, such as 75% of ads support queries handled by agents, measure workload or speed.
- Who leads AI at Alphabet?
- AI at Alphabet is led by Koray Kavukcuoglu (SVP, Google DeepMind, and Chief AI Architect); Demis Hassabis (Chair of Google DeepMind and Chief Scientist, Alphabet); Amin Vahdat (SVP and Chief Technologist, AI and Infrastructure). Ownership of AI on the org chart: CEO Sundar Pichai; Koray Kavukcuoglu, Chief AI Architect, unifies internal coding tools.
- What AI platform does Alphabet use?
- Alphabet builds its internal AI tools itself: Gemini models, the Antigravity agentic development platform, coding tools such as Cider, Jetski and Agent Smith, and a Gemini chatbot trained on Google's own documentation; Koray Kavukcuoglu was uniting the coding tools under Antigravity in Apr 2026.
- Which AI models does Alphabet use?
- Google says Gemini remains the primary model for internal development; since Sep 2026 every engineer can use Anthropic's Claude Opus 5 inside Antigravity on a per-user quota, while Claude Code and OpenAI's Codex stay barred for most staff (Business Insider, Bloomberg).
- How many Google employees use AI?
- Google reports adoption by group, not company-wide: about 75% of new code was AI-generated in Apr 2026, and 83% of the sales team used Gemini-assisted tools weekly in Jul 2026; Alphabet had 204,757 employees on Sep 30, 2026.
- What return has Alphabet reported from AI?
- Alphabet credits AI with business returns in 2 functions, both reported in the last 18 months, which places it at Scaling: Philipp Schindler said on Oct 29, 2025 that Gemini tools for Google's ad sales teams raised seller productivity by over 10% and led to hundreds of millions of dollars in incremental revenue; Google estimates that AI has raised engineering velocity about 10% across the company (Sundar Pichai, Jun 2025, restated by Google in Aug 2025). Google DeepMind's May 2025 report that an AlphaEvolve scheduling heuristic recovers about 0.7% of Google's worldwide compute is a capacity measure, not a currency return. On rollout, Google reports adoption by group: about 75% of new code is AI-generated (Apr 2026) and 83% of the sales team uses Gemini-assisted tools weekly (Jul 2026), while Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect, is uniting the internal coding tools under Antigravity, and Pichai and Schindler report results on earnings calls. Alphabet falls short of Embedded because Alphabet reports no firm-level cost, revenue or headcount effect it credits to AI, has not said how many of its 204,757 employees use AI each week, and grew headcount 7.7% in the year to Sep 30, 2026.
Open questions
- Will Alphabet publish a firm-level return from its internal AI use, in cost, revenue or headcount, beyond the 2 function-level returns it has reported?
- How many of Alphabet's 204,757 employees use AI tools each week, outside the sales team and Cider's users?
- Has Google updated its estimate of a 10% engineering velocity gain since Aug 2025, given that the AI-generated share of new code has since risen from about 30% to about 75%?
- What are the dollar values of the 0.7% of worldwide compute that AlphaEvolve recovers and of the treasury agent's results?
- What do the agents that handle 75% of ads support queries return in support cost or headcount?
- Who owns AI adoption outside engineering after Brian Saluzzo's departure, and has the unification of coding tools under Antigravity been completed?
- How much compute does Alphabet allocate to its own employees' AI tools, and how much does it spend on Anthropic models for its engineers?
- Does Google formally count AI use in performance reviews, given differing statements in Aug 2025 and Feb 2026?
Sources
Alphabet and Google
- Sundar Pichai: Google Cloud Next 2026 remarks (Apr 22, 2026)
- Sundar Pichai: Q3 2026 earnings remarks (Oct 28, 2026)
- Sundar Pichai: Q2 2026 earnings remarks (Jul 22, 2026)
- Sundar Pichai: Q1 2026 earnings remarks (Apr 29, 2026)
- Sundar Pichai: Q4 2025 earnings remarks (Feb 4, 2026)
- Sundar Pichai: Q3 2024 earnings remarks (Oct 2024)
- 14 ways Googlers use AI to work smarter (Aug 18, 2025)
- Our 2026 Responsible AI Progress Report (Feb 18, 2026)
- AI Responsibility Update 2026 (PDF, Feb 2026)
- Google DeepMind: AlphaEvolve (May 14, 2025)
- Google Cloud: How Google does it, building AI agents for cybersecurity (Oct 23, 2025)
Filings and investor materials
- Q3 2026 earnings release, Form 8-K exhibit (Oct 28, 2026)
- Alphabet 2025 Form 10-K (filed Feb 5, 2026)
- Q3 2026 earnings call transcript (Oct 28, 2026)
- Q2 2026 earnings call transcript (Jul 22, 2026)
- Q1 2026 earnings call transcript (Apr 29, 2026)
- Q4 2025 earnings call transcript (Feb 4, 2026)
- Q3 2025 earnings call transcript (Oct 29, 2025)
- Anat Ashkenazi at the Morgan Stanley TMT Conference, transcript (Mar 3, 2026)
- Lex Fridman Podcast #471, Sundar Pichai transcript (Jun 5, 2025)
News
- Bloomberg via The Star: Google's internal politics leave it playing catch-up on AI coding (Apr 23, 2026)
- Business Insider: Google's Agent Smith helps employees with AI-driven coding (Mar 27, 2026)
- Business Insider: Inside Google's Project EAT (Jan 30, 2026)
- AI Weekly, citing Business Insider: Google opens Claude Opus 5 to all engineers via Antigravity (Sep 15, 2026)
- Android Authority, citing The Information: Google AI coding strike team (Apr 22, 2026)
- People Matters, citing Business Insider: Google increases pressure on employees to adopt AI (Feb 27, 2026)
- CNBC: Google execs say employees have to be more AI-savvy (Jul 29, 2025)
- CNBC: Google must double AI serving capacity every 6 months (Nov 21, 2025)
- Entrepreneur: Google tells employees to use AI more for coding (Aug 21, 2025)
- 404 Media: Google employees internally share memes about its AI (Jun 4, 2026)
- Anadolu Agency: Google appoints Koray Kavukcuoglu SVP of Google DeepMind (Aug 6, 2026)
- Semafor: Google names new chief AI architect (Jun 11, 2025)
- TechCrunch: Google promotes Amin Vahdat (Dec 10, 2025)
- PPC Land: Google's 2025 Ads Safety Report (Apr 17, 2026)
- Business Today, citing Business Insider: Google launches Goose for employees (Feb 21, 2024)
- Fortune: Alphabet names Anat Ashkenazi CFO (Jun 5, 2024)
- CapitalG: Philipp Schindler biography
- CIO Dive: Citi appoints former Google executive Brian Saluzzo as CIO (Apr 28, 2026)
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
Alphabet profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 5, 2026
