In late May 2025, Meta split its generative-AI organization into two groups: AGI Foundations, co-led by Ahmad Al-Dahle and Amir Frenkel, and AI Products, led by Connor Hayes. The reported goal was to separate longer-term model and capability work from consumer-facing products, clarify ownership and move faster. It was an interim arrangement, not Meta’s lasting AI structure: by August, reports described another reorganization under Meta Superintelligence Labs.
What Meta changed in May 2025
The two-team structure drew a line between building AI capabilities and putting generative AI into products. The names and leadership were reported from internal communications rather than a full public organizational chart, so the divisions are best understood as an allocation of responsibility—not proof that every AI project sat in one of two boxes.
| Group | Reported leader | Reported focus |
|---|---|---|
| AGI Foundations | Ahmad Al-Dahle and Amir Frenkel | Llama, AI agents, reasoning models, media generation and capabilities associated with Meta’s longer-term AGI ambitions. |
| AI Products | Connor Hayes | Meta AI and other user-facing generative-AI features. |
| FAIR | Not stated in the May reports | Meta’s Fundamental AI Research lab remained outside this particular two-team arrangement; it was not thereby outside Meta’s wider AI strategy. |
Axios reported the group names and leaders on May 27, 2025, based on an internal memo. The Information’s account described a more granular division of work: Al-Dahle’s remit included Llama, agents and reasoning, while Frenkel’s included Meta AI, media generation and ecosystem adoption. That reporting suggests the co-leads’ boundaries were not simply a clean split between research and products. Axios’s May 2025 report and The Information’s account of the shake-up provide the reported details.
Who led the teams—and what their names meant
Ahmad Al-Dahle and Amir Frenkel
Al-Dahle had led Meta’s generative-AI group and was a prominent figure in its Llama work. He and Frenkel were reported as co-leaders of AGI Foundations. The word “AGI” described an organizational ambition; it did not establish that Meta had achieved artificial general intelligence or had a settled technical route to it.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Connor Hayes
Hayes, a longtime Meta product executive, was reported to lead AI Products. Its remit was closer to the user experience: the Meta AI assistant and generative-AI features delivered through Meta’s apps. Meta AI as an assistant is distinct from AI features embedded in Facebook, Instagram, WhatsApp and Messenger, and neither category accounts for every way Meta uses AI.
Chris Cox and FAIR
Meta chief product officer Chris Cox communicated the change internally, according to the reports. FAIR—the Fundamental AI Research lab—was reported to remain separate from the two new groups. Meta had previously described a broader reorganization in which AI-for-Product teams moved toward product engineering while FAIR remained a distinct research pillar. That history helps explain why “Meta AI” should not be treated as a single team covering all research and product engineering. Meta’s 2022 explanation of its AI organization offers that earlier context.
Why Meta said it needed a new structure
The stated rationale was operational: reduce dependencies between teams, make decision ownership clearer, allocate resources more effectively and accelerate model and product work. Cox’s reported message framed the change around speed, accountability and effectiveness. Separating longer-horizon capability development from product delivery can give each a clearer owner, but it does not automatically make the handoff between them work.
The Information also reported internal concerns about burnout, infighting, low employee-satisfaction scores and a lack of focus, citing people familiar with the matter. Those are attributed reports, not established company-wide findings. They add a possible organizational context to the change, but they do not prove that any one concern caused it.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →The competitive and product pressures behind the split
Competition and the DeepSeek moment
Meta was competing with OpenAI, Google, Microsoft and other AI developers for talent, model capability and users. Axios cast the restructuring as part of that competition. The Information reported that Meta set up internal “war rooms” after DeepSeek’s models attracted attention in early 2025. That is context reported by the outlet, not evidence that DeepSeek directly triggered the May decision.
Llama releases and credibility
The Information reported that Meta delayed parts of Llama 4 after performance concerns and later faced criticism for submitting an experimental Llama 4 Maverick version to a leaderboard rather than the exact public release. These episodes illustrate the execution and credibility questions surrounding the reorganization; they do not establish that the new organization caused, or would resolve, those problems. Benchmark results can depend on model version, prompts and tuning, so a leaderboard entry should not be treated as a like-for-like comparison without those details.
Llama was also a platform bet
The organizational story was not just about Meta AI, the consumer assistant. Meta was also trying to build a developer ecosystem around Llama, spanning downloadable models, hosted access and adoption by developers and businesses. In April 2025, Meta announced a limited free preview of the Llama API; that launch condition is historical and does not establish current pricing or availability. Meta’s LlamaCon announcement describes the preview.
Meta also announced a Llama Startup Program in May 2025 with eligible startups offered up to $6,000 per month for up to six months in hosted-API reimbursements, and a Meta-AWS program in July 2025 offering selected startups up to $200,000 in AWS promotional credits. Both are dated program terms, not evidence that either offer remains open. Meta’s Llama Startup Program announcement and its AWS program announcement set out those historical terms.
The strategic tension: models, products and research
Meta’s split addressed three distinct jobs, each with different measures of success:
Rank #4
- Foundational models and capabilities: Develop and evaluate models, reasoning, agents and media generation. This work can involve long timelines and uncertain commercial returns.
- Products: Make AI useful and dependable for people using Meta’s apps and services. Product teams must address reliability, safety, user adoption and iteration speed.
- Research: Pursue fundamental questions that may not map directly to a near-term product roadmap. FAIR’s position outside the May structure underscored that this work was not identical to either product delivery or the reported AGI Foundations remit.
Keeping these jobs close can improve feedback between users and model builders; separating them can make priorities and accountability clearer. It can also create silos: product teams may press for immediate usefulness while researchers pursue capabilities that take longer to mature. The value of a reorganization depends on how well those teams share evaluations, infrastructure and decisions—not on the labels alone.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why the May structure did not tell the whole story
By August 2025, Meta had reorganized again. The Information reported a plan for four groups under Meta Superintelligence Labs: a new group temporarily called “TBD Lab,” a product group including Meta AI, an infrastructure group and FAIR. The Information described it as Meta’s fourth AI overhaul in six months. Bloomberg separately reported on August 19 that chief AI officer Alexandr Wang circulated a memo dividing the organization into four teams, in an effort to accelerate the pursuit of superintelligence and make use of recently recruited talent.
The August reports change how the May split should be read: it was a significant experiment in dividing foundational and product work, but not a durable final operating chart. Meta’s public AI site now presents Meta Superintelligence Labs as the home of newer model and media-generation work, including Muse Spark and Muse Image. That public positioning confirms the later identity, but does not document every internal reporting line. See The Information’s report on the four-group plan, Bloomberg’s August 19 report and Meta AI’s public site.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesRepeated changes can indicate adaptation to new goals, talent and infrastructure needs; they can also create uncertainty and consume employees’ attention. The available reporting establishes that the structure changed again, not that the May design definitively failed or that the August design solved Meta’s execution challenges. “Superintelligence” is a change in strategic framing, not evidence of a technical breakthrough.
How to judge whether the reorganization worked
Organizational names alone cannot answer that question. A useful assessment would track outcomes over time and compare public releases on consistent terms:
- Model quality: Did public Llama releases improve on credible, reproducible evaluations using the actual released versions?
- Release execution: Did Meta reduce delays and make model versions and evaluation conditions clearer?
- Product outcomes: Did Meta AI become more useful and reliable, and did people keep using it?
- Developer adoption: Did model access and deployment options expand, with clear model-specific terms?
- Talent and operating health: Did the organization retain people and reduce the internal friction reported at the time?
- Economics and strategic coherence: Did product and developer adoption justify the costs of models, inference, infrastructure and compensation—and could Meta explain how FAIR, Llama, products and infrastructure fit together?
For developers and companies, the distinction between the organizational news and a purchasing decision matters. Meta AI is a consumer-facing assistant whose features and availability can vary by location, account, app, device and rollout. Llama can offer teams model control and customization, but self-hosting or partner hosting still involves compute, engineering, evaluation, safety work and compliance with the specific model’s terms. The 2025 API preview and startup offers do not establish current service pricing or eligibility. The reorganization is a reason to watch Meta’s products and platform; it is not, on its own, evidence that a particular service fits a buyer’s needs.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
