Verified
Meta Platforms
Piloting- 66,000+ staff
- $135-145B 2026 capex
- MetaCode and Muse Code built in-house
- Output per engineer +30% (Jan 2026)
Meta Platforms deploys AI inside the company mainly through tools it builds, MetaCode and Muse Code for engineers, Metamate for staff and agents such as Hatch in testing, while cutting staff use of Anthropic's Claude in 2026. CFO Susan Li said in Jan 2026 that output per engineer rose 30% since early 2025, mostly from agentic coding. CTO Andrew Bosworth runs internal adoption, Alexandr Wang's Meta Superintelligence Labs builds the Muse models, and Meta guides 2026 capital spending at $135-145 billion.
ai.meta.comTheir approach to AI
- The platform
- Meta Platforms builds most of its internal AI tooling: MetaCode (formerly Devmate) and Muse Code for engineers, Metamate as the staff assistant, the Confucius agent framework, and agents such as REA in ads ranking and Hatch in employee testing.
- The models
- Meta Platforms runs its own Llama and, since Apr 2026, Muse models, and has given staff tools from Anthropic, OpenAI and Google; in 2026 it began steering engineers from Claude to MetaCode and Muse Code (The Information, Jun and Oct 2026).
- The reach
- Meta Platforms has not published how many of its employees use AI; The Information reported about 30,000 Claude Code users in Oct 2026, down from about 60,000 earlier in 2026, and more than 30,000 MetaCode users.
- The catch
- Meta Platforms credits AI with one internal return, a 30% rise in output per engineer since early 2025 (Jan 2026); its ads gains are reported as conversion lifts, and an 80% cut in content-processing cost comes from offline tests.
Snapshot
- Employees
- 66,000+
- Sep 30, 2026 (Q3 2026 call); 78,865 at Dec 31, 2025 (2025 Form 10-K)
- Technology budget
- $135-145B capex (2026)
- Guidance of Oct 21, 2026; internal AI tool spending not disclosed
- AI/ML headcount
- Not disclosedest.
- Meta Superintelligence Labs was formed Jun 30, 2025; 7,000 staff were reported moved into AI-focused roles in May 2026
- Who owns AI
- Andrew Bosworth and Alexandr Wang
- Bosworth, CTO, leads AI for Work and Applied AI Engineering; Wang, Chief AI Officer, leads Meta Superintelligence Labs
- Platform
- Built in-house, with bought tools
- MetaCode, Muse Code, Metamate, Confucius agent framework; Claude Code and Gemini also in use
- Models
- Muse, Llama, Anthropic, OpenAI, Google
- Muse Spark, Meta's first Muse model, launched Apr 8, 2026
- Reach
- Not disclosed by Metaest.
- About 30,000 Claude Code and 30,000+ MetaCode users reported by The Information, Oct 2026
- Use cases
- Coding, ads ranking, content, integrity
- Plus internal agents such as REA and Hatch
Key findings
Meta Platforms CFO Susan Li said on Jan 28, 2026 that output per engineer rose 30% since the start of 2025, mostly from agentic coding, and that power users of AI coding tools raised output 80% year over year; it is the one internal return Meta credits to AI.
Meta Platforms has been moving engineers from Anthropic's Claude to its own MetaCode and Muse Code: The Information reported in Oct 2026 that Claude Code users fell from about 60,000 to about 30,000, while MetaCode had more than 30,000 users.
Meta Platforms put CTO Andrew Bosworth in charge of its AI for Work adoption program in Mar 2026, created an Applied AI Engineering organization under him led by Maher Saba, and in Jun 2026 told staff it would put AI token use under budgets after usage grew exponentially.
Meta Platforms ended Sep 2026 with over 66,000 employees, down from 78,865 at the end of 2025, after cutting about 8,000 jobs in May 2026 to offset its infrastructure spending, which it guides at $135-145 billion for 2026.
Meta Platforms made AI-driven impact a core expectation in 2026 performance reviews (Nov 2025 memo), then told staff in Sep 2026 that AI adoption dashboards and token counts would not be used to judge their impact, according to Wired.
Where Meta sits
- Stage
- Piloting
Meta Platforms is at Piloting because it has reported one business return it credits to AI in the last 18 months: CFO Susan Li said on Jan 28, 2026 that output per engineer rose 30% since the start of 2025, with most of that growth coming from agentic coding, and that power users of AI coding tools raised output 80% year over year. Meta Platforms reports its AI ads work as conversion and click lifts and revenue run-rates of AI-powered ad tools, not as revenue it credits to AI, and the more than 80% cut in per-unit content-processing cost from Muse models comes from offline evaluations (Oct 2026). On rollout, Meta Platforms has not said how many of its about 66,000 employees use its AI tools; the counts in the public record, such as about 30,000 Claude Code users and more than 30,000 MetaCode users (Oct 2026), come from press reports. CTO Andrew Bosworth has led the AI for Work program since Mar 2026. Meta Platforms falls short of Scaling because it has reported no AI-credited return in a second function and no firm-level figure; Susan Li said in Oct 2026 that AI investments deliver incremental ad revenue without putting a number on it.
How the stages are defined - Ownership
- CTO Andrew Bosworth runs internal AI adoption; Chief AI Officer Alexandr Wang builds the models
- Platform
- Hybrid: MetaCode, Muse Code and Metamate built in-house; Claude Code and Gemini bought
- Models
- In-house Muse and Llama, plus Anthropic, OpenAI and Google tools
- Hosting
- Meta's own data centers (30 owned locations) plus third-party cloud capacity from CoreWeave, Google Cloud and, from 2027, Nebius
- Reach
- Not disclosed; ~30,000 Claude Code and 30,000+ MetaCode users reported (Oct 2026)
How they deploy
Build the tools, own the stack
Mark Zuckerberg told investors in Jan 2026 that Meta Platforms is a deep technology company that needs to build and control its own models, and in Jul 2026 that Meta builds coding, development and internal productivity tools partly because it needs tools tuned for itself. Meta Platforms says it builds its own data centers, infrastructure, chips and low-level software, and in Apr 2026 moved from the Llama family to Muse Spark, the first model from Meta Superintelligence Labs.
Fewer outside tools, more in-house ones
Meta Platforms gave staff its own models alongside tools from OpenAI, Anthropic and Google, then in 2026 began steering engineers toward MetaCode, previously called Devmate, and Muse Code. A Jun 2026 memo reported by The Information said Meta would nudge staff from third-party AI tools to in-house ones while keeping access to new external models, and engineers in the Applied AI Engineering division were tasked with improving MetaCode to reduce reliance on Claude.
AI-native teams and flatter structure
Mark Zuckerberg said in Jan 2026 that 2026 is the year AI starts to dramatically change how Meta Platforms works, with investment in AI-native tooling, more weight on individual contributors and flatter teams. Zuckerberg said in Apr 2026 that 1 or 2 people are building in a week what used to take dozens of people months, and Meta piloted new titles in a Reality Labs group: AI builder, AI pod lead and AI org lead (Business Insider, Mar 2026).
From usage targets to outcomes
Meta Platforms set AI adoption goals by organization, including a first-half 2026 target for 65% of engineers in its creation organization to write more than 75% of committed code with AI (Business Insider, Mar 2026). Andrew Bosworth told staff in Apr 2026 that nobody should use AI tools just for the sake of using them, and Meta told employees in Sep 2026 that adoption dashboards and token counts would not be used to judge performance, Wired reported.
- Deployment model
- Meta Platforms splits AI between a model lab and an adoption program. Meta Superintelligence Labs, under Chief AI Officer Alexandr Wang, builds the Muse models that power Meta AI and, increasingly, internal tools. CTO Andrew Bosworth leads AI for Work, the program that drives internal adoption, and the Applied AI Engineering organization led by Maher Saba, which builds the data engine and tooling for model training. Meta Compute, co-led by Santosh Janardhan and Daniel Gross, runs the data center fleet and capacity deals. Engineering teams run their own agents on shared frameworks, such as the Ranking Engineer Agent built on Meta's Confucius framework for ads ranking work.
- Model providers
- Meta Platforms trains its own models: the Llama family through Llama 4 (Apr 2025) and, since Apr 8, 2026, the Muse family from Meta Superintelligence Labs, starting with Muse Spark. Inside the company, staff have used Anthropic's Claude Code, Google's Gemini and OpenAI tools alongside Meta's own: Janelle Gale told employees in Nov 2025 they could use Metamate and Gemini for performance reviews, and The Information reported about 60,000 Claude Code users earlier in 2026. Meta moved content understanding models onto the Muse family in Q3 2026, and reported in Oct 2026 that its third-party AI token costs were one of the drivers of its expense growth.
- Data and infrastructure
- Meta Platforms owns 30 data center locations and leases others (2025 Form 10-K), including the Prometheus cluster, which Meta expected to be the first above 1 gigawatt, and Hyperion, planned to scale to 5 gigawatts (Jul 2025), plus a 1 gigawatt El Paso, Texas data center venture with BlackRock announced in Jul 2026. Meta also buys third-party cloud capacity: about $21 billion with CoreWeave through Dec 2032 (Apr 2026), up to about $27 billion with Nebius from early 2027 (Mar 2026), and Google Cloud TPUs rented for model training (reported Feb 2026). Contractual commitments, mostly for third-party cloud capacity and infrastructure, stood at $131.05 billion at Dec 31, 2025 and rose by $107 billion in Q1 2026. Meta runs workloads on Nvidia, AMD and its own MTIA chips, and is rolling out more than 1 gigawatt of custom silicon developed with Broadcom (Apr 2026).
Platforms and internal tools
MetaCode (formerly Devmate)
MetaCode is Meta Platforms' in-house coding assistant, previously known as Devmate. The Information reported in Jun 2026 that Applied AI Engineering staff were improving it to reduce Meta's reliance on Anthropic's Claude, including by generating programming challenges as reinforcement learning data.
- Users
- 30,000+ (The Information, reported Oct 2026)
Muse Code
Muse Code is a coding agent Meta Platforms built alongside its Muse models; Meta engineers use it, and Meta began offering it to businesses through Meta Enterprise Platform in Sep 2026.
- Users
- 6,000+ employees (The Information, reported Oct 2026)
- Models
- Muse
Metamate
Metamate is Meta Platforms' internal AI assistant for employees; Chief People Officer Janelle Gale told staff in Nov 2025 they could use Metamate, as well as Google's Gemini, to write performance review content.
Confucius and the Ranking Engineer Agent (REA)By Mar 2026
Confucius is Meta Platforms' internal agent framework for multistep reasoning tasks with code generation and connections to job schedulers and experiment tracking. REA, built on it, runs ads ranking experiments end to end, from hypothesis to training jobs and debugging, within compute budgets an engineer approves.
HatchInternal testing, 2026
Hatch is an autonomous agent Meta Platforms has staff testing, which runs tasks across apps and the web such as research, form filling and bookings; Wired reported in Sep 2026 that Meta was encouraging employees to try it.
AI GatewayBy Jun 2026
AI Gateway is a central dashboard Meta Platforms built to monitor employees' AI usage and spending in real time, with alerts for spending spikes planned and budgets for tokens expected in 2027 (The Information, Jun 2026).
Vendors and partners
Anthropic
Claude Code for engineers · The Information reported in Oct 2026 that Meta's Claude Code users fell from about 60,000 earlier in 2026 to about 30,000 as Meta pushed its own tools.
GoogleTPU rental reported Feb 2026
Gemini for staff; Google Cloud TPUs for model training · Meta told staff in Nov 2025 they could use Gemini for performance reviews, and rents Google Cloud TPUs to train models (The Information, reported Feb 2026).
OpenAI
AI tools available to staff · Meta gave staff tools from OpenAI alongside Anthropic and Google before its 2026 push toward in-house tools (The Information, Jun 2026).
CoreWeaveExpanded Apr 2026
Third-party AI cloud capacity · About $21 billion of capacity through Dec 2032, including early Nvidia Vera Rubin deployments, to scale Meta's inference workloads.
NebiusMar 2026
Third-party AI cloud capacity from early 2027 · 5-year agreement worth up to about $27 billion: $12 billion of dedicated Vera Rubin capacity plus up to $15 billion of further capacity.
Nvidia, AMD and Broadcom
AI chips · Meta runs its Andromeda ads retrieval engine on Nvidia, AMD and its own MTIA chips (Jan 2026) and is deploying more than 1 gigawatt of custom silicon developed with Broadcom (Apr 2026).
Leadership
AI, data and platform leaders
Executives
AI org map
Central model lab and central adoption program; teams build their own agents on shared frameworks
Meta Platforms concentrates model building in Meta Superintelligence Labs under Chief AI Officer Alexandr Wang, formed in Jun 2025, and internal adoption under CTO Andrew Bosworth, who took over the AI for Work program in Mar 2026 and oversees the Applied AI Engineering organization led by Maher Saba. Meta Compute, announced in Jan 2026 and reporting to Mark Zuckerberg, runs the data center fleet and capacity deals under Santosh Janardhan and Daniel Gross. Alex Schultz became Meta's first Chief Data Officer in Jul 2026 to lead AI-led analytics and research.
Meta Platforms has not published how many people work on AI. Meta Superintelligence Labs, formed in Jun 2025, combined the Llama teams, FAIR, product teams and a new frontier model unit, TBD Lab. Meta reported 78,865 employees at the end of 2025 and over 66,000 at Sep 30, 2026, after a May 2026 reduction of about 8,000 people; several outlets reported that about 7,000 other employees were moved into AI-focused roles at the same time.
Reporting lines
CEO
Mark Zuckerberg
Meta Superintelligence Labs
Alexandr Wang, Chief AI Officer
Builds the Muse models; reporting line to the CEO reported, not published by Meta
Chief Technology Officer and AI for Work
Andrew Bosworth
Internal AI adoption program since Mar 2026; also leads Reality Labs; reporting line not published by Meta
Applied AI Engineering
Maher Saba
2 teams: interfaces and tooling, and tasks, data and evaluations (Mar 2026)
Meta Compute
Santosh Janardhan and Daniel Gross
Data centers, silicon and capacity partnerships (Jan 2026)
Chief Operating Officer
Javier Olivan
Chief Data Office
Alex Schultz
Analytics, data science, data engineering and research; the line to the COO was reported by Axios in Jul 2026 and is unconfirmed
Meta Platforms' executive page, checked Nov 5, 2026, lists Alexandr Wang as Chief AI Officer, Andrew Bosworth as CTO, Alex Schultz as Chief Data Officer, Santosh Janardhan as Head of Infrastructure, Daniel Gross as Co-Lead of Meta Compute and Chirantan Desai as Chief Enterprise Platform Officer, but does not publish reporting lines. Guy Rosen, the CISO who ran AI for Work before Andrew Bosworth and most recently worked on bringing AI into Meta's internal business operations, said in Jun 2026 that he would leave Meta.
Where the AI workforce sits
Meta Superintelligence Labs (models, research, AI products, AI infrastructure)
Meta has not published the lab's size.
Not disclosedApplied AI Engineering
Formed Mar 2026 with up to 50 employees per manager, according to the Wall Street Journal.
Not disclosed
Named people
Use cases
Meta Platforms uses AI inside the company for software engineering, ads ranking and retrieval, content understanding and recommendations, content enforcement, analytics and employee assistants and agents. Meta Platforms has published one return it credits to AI: output per engineer up 30% since the start of 2025, mostly from agentic coding (Jan 2026). Meta reports its ads models as conversion and click lifts, its AI ad tools as revenue run-rates, and a more than 80% cut in content-processing cost from offline tests, none of which is a return credited to AI, and it publishes no firm-level AI figure.
| Use case | Area | Status | ROI |
|---|---|---|---|
| Agentic coding tools Meta Platforms engineers write code with MetaCode, Muse Code and Anthropic's Claude Code, under organization targets such as 65% of creation-organization engineers writing more than 75% of committed code with AI in the first half of 2026 (Business Insider, Mar 2026). | Software engineering | In production | Output per engineer up 30% since the start of 2025, mostly from agentic coding (Jan 2026)Source |
| Ranking Engineer Agent (REA) REA runs ads ranking experiments end to end on Meta's Confucius framework. In its first production validation across 6 models it doubled average model accuracy over baseline, and Meta says early adopters raised model-improvement proposals from 1 to 5 in the same time, with 3 engineers covering 8 models (Mar 2026). | Ads ranking engineering | In production | |
| Ads ranking and retrieval models Meta Platforms' GEM ranking model, Andromeda retrieval engine, Lattice model unification and Meta Generative Recommender select which ads to show. Meta reported a 9.6% rise in app ad conversions from expanding its ranking model and a 1.4% rise in Facebook conversions from Andromeda in Q3 2026, and said in Jan 2026 that AI ranking work lifted Facebook ad clicks 3.5%. | Advertising | In production | |
| AI-powered ad tools Meta Platforms automates campaign setup and creative for advertisers; revenue running through its end-to-end AI-powered ad tools passed a $60 billion annual run-rate in Oct 2025, and its video generation tools reached a $10 billion run-rate in Q4 2025. Both figures count revenue flowing through the tools, not revenue Meta credits to AI. | Advertising | In production | |
| LLM content understanding Meta Platforms runs every public Instagram Reels and Feed post through an LLM that analyzes topic and tone, and in Q3 2026 began moving its content understanding models to the Muse family, which in offline evaluations outperformed prior production models and cut per-unit processing costs by more than 80%. | Content and recommendations | Pilot | |
| AI content enforcement Meta Platforms announced in Mar 2026 that it is deploying more advanced AI systems to enforce rules on terrorism, child exploitation, drugs, fraud and scams once they outperform current methods, and will reduce its reliance on third-party content enforcement vendors, with people keeping the highest-risk decisions such as appeals of account disablement. | Integrity | In production | |
| AI-native build of the Forum app A small Meta Platforms team built Forum, a standalone Groups app, with an AI agent running planning, implementation, testing and verification while engineers set goals and approved results; Forum reached core feature parity with Facebook Groups in 8 weeks, and agents now help triage user feedback (Oct 2026). | Product development | In production | |
| Metamate and the AI Performance Assistant Meta Platforms staff use Metamate, and Gemini, to draft performance review content, and Meta rolled out an AI Performance Assistant for the review cycle that began Dec 8, 2025. | Employees | In production | |
| Hatch personal agent Meta Platforms employees are testing Hatch, an agent that carries out tasks across apps and the web, which Meta encouraged staff to try from Sep 2026 (Wired). | Employees | Pilot |
Adoption
Meta Platforms moved from encouraging AI use to expecting it: Janelle Gale made AI-driven impact a core expectation for 2026 reviews (Nov 2025), organizations set targets for AI-written code (late 2025 and 2026), and an internal Claudeonomics leaderboard ranked the top 250 token users before it came down (The Information, Apr 2026). Meta then reined usage in, planning token budgets in Jun 2026 and telling staff in Sep 2026 that adoption dashboards and token counts would not judge their impact. Meta Platforms has not published an employee usage count; press reports put Claude Code at about 60,000 users earlier in 2026 and about 30,000 in Oct 2026, with more than 30,000 MetaCode users and more than 6,000 Muse Code users, against a workforce of over 66,000 at Sep 30, 2026.
By the numbers
Rollout, first pilot to today
Jun 30, 2025
Mark Zuckerberg consolidated Meta's Llama, FAIR, product and new frontier model teams into Meta Superintelligence Labs, led by Chief AI Officer Alexandr Wang.
Jul 2025
Mark Zuckerberg said Meta teams were adapting Llama 4 to build autonomous agents that help improve the Facebook algorithm, at low volume so far.
Nov 2025
Janelle Gale told Meta employees that AI-driven impact would be a core expectation from 2026 and that they could use Metamate and Gemini for performance reviews.
Jan 28, 2026
Susan Li said Meta's output per engineer rose 30% since the start of 2025, mostly from agentic coding, which saw a big jump in Q4 2025.
Mar 2026
Meta formed Applied AI Engineering under Maher Saba, put Andrew Bosworth in charge of the AI for Work adoption program, and set organization targets for AI-written code.
Mar 17, 2026
Meta described REA, an autonomous agent that runs ads ranking experiments end to end on its Confucius framework.
Apr 2026
The Information reported that Meta staff used 60.2 trillion tokens in 30 days, later 73.7 trillion, on the internal Claudeonomics leaderboard, and Andrew Bosworth announced an Agent Transformation Accelerator to unify internal AI tools.
Jun 2026
Meta told about 6,000 staff it would budget AI token use from 2027, built the AI Gateway spending dashboard, and began nudging engineers toward MetaCode.
Sep 2026
Meta stopped using AI adoption dashboards and token counts to judge employees' impact and pushed staff to test its Hatch agent, Wired reported.
Oct 2026
The Information reported that Meta's Claude Code users fell to about 30,000 from about 60,000 earlier in 2026, while MetaCode passed 30,000 users and Muse Code 6,000.
Economics & controls
Spend and ROI
Capital spending
Meta Platforms bought $69.69 billion of property and equipment in 2025, mostly servers, data centers and network infrastructure, and narrowed its 2026 capital expenditure guidance to $135-145 billion on Oct 21, 2026, from $115-135 billion in Jan 2026. Meta spent $34.9 billion in Q3 2026 alone and reported negative free cash flow of $2.1 billion for the quarter. Mark Zuckerberg said in Apr 2026 that most of that year's increase in the capital spending forecast came from higher component costs, especially memory.
What AI itself costs
Meta Platforms named third-party AI token costs among the drivers of its expense growth in Q2 and Q3 2026, without giving a figure. The Information reported in Jun 2026 that Meta's internal AI token costs were projected to run into the billions in 2026 and that Meta would manage tokens with budgets and allocation from 2027. Meta's contractual commitments, mostly third-party cloud capacity and infrastructure, were $131.05 billion at Dec 31, 2025, and multi-year cloud deals added $107 billion in Q1 2026.
Reported ROI
Meta Platforms reports one internal return it credits to AI: output per engineer up 30% since the start of 2025, mostly from agentic coding, with power users of AI coding tools up 80% year over year (Susan Li, Jan 28, 2026). Susan Li said in Oct 2026 that Meta's AI investments already deliver incremental advertising revenue through engagement, recommendations and monetization, without a figure, and that Meta Superintelligence Labs investments will take longer to return. Meta's ads gains are reported as conversion and click lifts, and its 80% lower content-processing cost on Muse models comes from offline evaluations.
Headcount
Meta Platforms cut about 8,000 jobs, about 10% of staff, from May 20, 2026 and closed about 6,000 open roles, according to TheWrap; Janelle Gale's memo announcing the cuts said they would help run the company more efficiently and offset Meta's other investments, and Susan Li said in Apr 2026 that a leaner operating model would help offset the investments Meta is making. Meta booked $1.2 billion of severance in Q2 2026 and ended Sep 2026 with over 66,000 employees, down from 78,865 at the end of 2025. Meta has not credited the reduction to AI productivity.
Governance and risk
Token budgets and cost controls
Meta Platforms built AI Gateway, a dashboard that tracks employees' AI usage and spending in real time, plans alerts for spending spikes, and told staff in Jun 2026 it would manage tokens with budgets from 2027 (The Information). Andrew Bosworth wrote in Apr 2026 that token usage alone measures nothing about impact.
AI in performance reviews
Meta Platforms made AI-driven impact a core expectation for 2026 reviews (Nov 2025), then told staff in Sep 2026 that AI adoption dashboards and token counts would not be used to evaluate performance and that managers should look at quality, velocity, problem complexity and scope (Wired and The Information, as reported by Employee Experience Magazine).
Employee data for agent training
Meta Platforms began collecting mouse movements, clicks, keystrokes and some screenshots from U.S. employees' work computers on a designated list of apps in Apr 2026 to train agents on computer use; a Meta spokesperson said safeguards protect sensitive content and the data will not be used for any other purpose (Fortune, citing Reuters).
Guardrails on autonomous agents
Meta Platforms' Ranking Engineer Agent works only on the ads ranking codebase, needs engineers to grant access through preflight checklist reviews, and confirms compute budgets up front, pausing runs when thresholds are reached (Mar 2026). Meta says people keep the highest-risk content decisions, such as appeals of account disablement and reports to law enforcement, as AI takes over more enforcement (Mar 2026).
Disclosed risks
Meta Platforms' 2025 Form 10-K says there is no assurance AI will make its products better or benefit its efficiency or profitability, and lists AI and cloud service providers among the third parties whose systems create cybersecurity risk. Meta also faces lawsuits over the material used to train its AI models.
Recent signals
Oct 21, 2026
Meta Platforms reported over 66,000 employees, narrowed 2026 capital spending to $135-145 billion, named third-party AI token costs as an expense driver, and said Muse models cut content-processing costs by over 80% in offline tests.
Oct 15, 2026
Meta engineers described building the Forum app with an AI agent running planning, implementation and testing, reaching core feature parity with Facebook Groups in 8 weeks.
Oct 2026
The Information reported that Meta's internal Claude Code users fell from about 60,000 to about 30,000 as MetaCode passed 30,000 users and Muse Code 6,000.
Sep 28, 2026
Meta launched Meta Enterprise Platform to sell Muse Code, the Muse agent, Meta Business Agent and the Muse API to businesses, led by Chief Enterprise Platform Officer Chirantan Desai.
Sep 2026
Meta told staff that AI adoption dashboards and token counts would no longer judge their performance, while pushing them to test the Hatch agent, Wired reported.
Sep 1, 2026
A memo from Alexandr Wang moved Meta's internal communications from Google Chat to Slack, calling Slack the strongest platform for agents, Business Insider reported.
Jul 29, 2026
Meta booked $1.2 billion of severance for the May job cuts, named third-party AI token costs among its expense drivers, and announced a 1 gigawatt El Paso data center venture with BlackRock.
Jul 1, 2026
Meta named CMO Alex Schultz its first Chief Data Officer to lead AI-led analytics and research.
Jun 2, 2026
Meta CISO Guy Rosen, who had most recently worked on integrating AI into Meta's internal business operations, said he would leave the company.
Jun 2026
The Information reported a Meta memo planning AI token budgets for staff, an AI Gateway spending dashboard and a push from third-party tools to MetaCode.
May 20, 2026
Meta began cutting about 8,000 jobs, about 10% of staff, and Mark Zuckerberg said no further company-wide layoffs were expected in 2026.
Apr 21, 2026
Meta began capturing U.S. employees' mouse movements, clicks and keystrokes on work apps to train AI agents, Reuters reported.
Apr 9, 2026
CoreWeave and Meta expanded their AI cloud agreement to about $21 billion through Dec 2032 to scale Meta's inference workloads.
Apr 8, 2026
Meta Superintelligence Labs released Muse Spark, its first model, succeeding Llama 4 across Meta AI and Meta's apps.
Mar 2026
Meta put CTO Andrew Bosworth in charge of internal AI adoption and created an Applied AI Engineering organization under Maher Saba.
Quick answers
- How does Meta Platforms deploy AI?
- Meta Platforms deploys AI inside the company mainly through tools it builds, MetaCode and Muse Code for engineers, Metamate for staff and agents such as Hatch in testing, while cutting staff use of Anthropic's Claude in 2026. CFO Susan Li said in Jan 2026 that output per engineer rose 30% since early 2025, mostly from agentic coding. CTO Andrew Bosworth runs internal adoption, Alexandr Wang's Meta Superintelligence Labs builds the Muse models, and Meta guides 2026 capital spending at $135-145 billion.
- What does Meta Platforms use AI for?
- Meta Platforms uses AI inside the company for software engineering, ads ranking and retrieval, content understanding and recommendations, content enforcement, analytics and employee assistants and agents. Meta Platforms has published one return it credits to AI: output per engineer up 30% since the start of 2025, mostly from agentic coding (Jan 2026). Meta reports its ads models as conversion and click lifts, its AI ad tools as revenue run-rates, and a more than 80% cut in content-processing cost from offline tests, none of which is a return credited to AI, and it publishes no firm-level AI figure.
- Who leads AI at Meta Platforms?
- AI at Meta Platforms is led by Andrew Bosworth (Chief Technology Officer; leads AI for Work and Applied AI Engineering); Alexandr Wang (Chief AI Officer; leads Meta Superintelligence Labs); Alex Schultz (Chief Data Officer). Ownership of AI on the org chart: CTO Andrew Bosworth runs internal AI adoption; Chief AI Officer Alexandr Wang builds the models.
- What AI platform does Meta Platforms use?
- Meta Platforms builds most of its internal AI tooling: MetaCode (formerly Devmate) and Muse Code for engineers, Metamate as the staff assistant, the Confucius agent framework, and agents such as REA in ads ranking and Hatch in employee testing.
- Which AI models does Meta Platforms use?
- Meta Platforms runs its own Llama and, since Apr 2026, Muse models, and has given staff tools from Anthropic, OpenAI and Google; in 2026 it began steering engineers from Claude to MetaCode and Muse Code (The Information, Jun and Oct 2026).
- How many Meta employees use AI?
- Meta Platforms has not published how many of its employees use AI; The Information reported about 30,000 Claude Code users in Oct 2026, down from about 60,000 earlier in 2026, and more than 30,000 MetaCode users.
- What return has Meta Platforms reported from AI?
- Meta Platforms is at Piloting because it has reported one business return it credits to AI in the last 18 months: CFO Susan Li said on Jan 28, 2026 that output per engineer rose 30% since the start of 2025, with most of that growth coming from agentic coding, and that power users of AI coding tools raised output 80% year over year. Meta Platforms reports its AI ads work as conversion and click lifts and revenue run-rates of AI-powered ad tools, not as revenue it credits to AI, and the more than 80% cut in per-unit content-processing cost from Muse models comes from offline evaluations (Oct 2026). On rollout, Meta Platforms has not said how many of its about 66,000 employees use its AI tools; the counts in the public record, such as about 30,000 Claude Code users and more than 30,000 MetaCode users (Oct 2026), come from press reports. CTO Andrew Bosworth has led the AI for Work program since Mar 2026. Meta Platforms falls short of Scaling because it has reported no AI-credited return in a second function and no firm-level figure; Susan Li said in Oct 2026 that AI investments deliver incremental ad revenue without putting a number on it.
Open questions
- How many Meta Platforms employees use AI tools each week, given that Meta has published no count and the public figures for Claude Code, MetaCode and Muse Code come from press reports?
- How does Meta Platforms measure output per engineer, and has the 30% rise reported in Jan 2026 grown since, as Susan Li expected?
- What does Meta Platforms spend on third-party AI tokens and internal AI tools, given that it names token costs as an expense driver without a figure?
- Will Meta Platforms put a dollar figure on the incremental ad revenue Susan Li says its AI investments deliver, or on the AI content enforcement that replaces third-party vendors?
- Do the more than 80% lower content-processing costs from Muse models hold in production, beyond the offline evaluations reported in Oct 2026?
- Who leads AI for Work after Guy Rosen's departure and the Applied AI Engineering reorganization, and how does it divide work with Meta Superintelligence Labs?
- Which models run inside MetaCode and Hatch, and what share of internal AI traffic goes to Muse, Llama, Anthropic, OpenAI and Google models?
- Which third-party clouds, beyond CoreWeave, Nebius and Google Cloud, run Meta Platforms' AI workloads, given that Amazon listed Meta among new AWS agreements in Oct 2026 without terms?
- Will Meta Platforms credit any of its 2026 job cuts to AI productivity?
Sources
Meta Platforms
Filings and earnings calls
- Q3 2026 earnings call transcript, The Motley Fool (Oct 21, 2026)
- Q2 2026 earnings call transcript (Jul 29, 2026)
- Q1 2026 earnings call transcript (Apr 29, 2026)
- Q4 2025 earnings call transcript (Jan 28, 2026)
- Q3 2025 earnings call transcript (Oct 29, 2025)
- Q2 2025 earnings call transcript (Jul 30, 2025)
- Meta Platforms 2025 Form 10-K
Partners
News
- Channel Insider: Microsoft and Meta cut internal Claude use (Oct 2026)
- The State of AI: Meta moves to cap employee AI usage (Jun 13, 2026)
- TechRadar: Meta has a new boss to encourage workers to use more AI (Mar 25, 2026)
- The Week: Leaked documents detail Meta's AI coding targets (Mar 27, 2026)
- The Decoder: Meta creates new applied AI engineering division (Mar 4, 2026)
- eWeek: Meta forms new AI engineering organization (Mar 6, 2026)
- Business Insider via AOL: Meta is about to start grading workers on their AI skills (Nov 2025)
- TechBuzz: Meta eases AI mandate, pushes new agent Hatch (Sep 3, 2026)
- The Next Web: Meta drops Google Chat for Slack, citing agents (Sep 1, 2026)
- Fortune: Meta to track employees' screens and keystrokes to train AI (Apr 21, 2026)
- Fortune: Meta cuts 10% of its workforce (May 21, 2026)
- TheWrap: Meta to cut 10% of staff as company centers AI initiatives (Apr 2026)
- Employee Experience Magazine: Behind the curtain at Meta, how AI is changing work (Sep 2026)
- Reuters via The Star: Meta names Alex Schultz first chief data officer (Jul 2, 2026)
- CDO Magazine: Meta names Alex Schultz first Chief Data Officer (Jul 2026)
- Business Today: Chirantan Desai to lead Meta's enterprise AI push (Sep 29, 2026)
- Bloomberg Law: Meta's Rosen, former head of election integrity, to depart (Jun 2, 2026)
- AFP via Khaleej Times: Meta releases first new AI model since shaking up team (Apr 8, 2026)
- Fortune: Key takeaways from Zuckerberg's superintelligence memo (Jul 2, 2025)
- Wikipedia: Meta Superintelligence Labs
- Reuters via MarketScreener: Meta establishes Meta Compute (Jan 12, 2026)
- Engadget: Zuckerberg announces Meta Compute (Jan 12, 2026)
- Dataconomy: Meta signs deal to rent Google TPUs (Feb 27, 2026)
- TechCrunch: Meta rolls out AI content enforcement systems (Mar 19, 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.
Meta Platforms profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 5, 2026
