Meta’s AI leadership shake-up unfolds as remarkable departures, bold hires, and a high-stakes push for “personal superintelligence” reshape the company’s brain trust
Within days of Shengjia Zhao joining Meta, the company’s top AI hiring spree hit a choppy crossing: Zhao, an OpenAI co-creator of ChatGPT, threatened to quit and return to his former employer. He went so far as to sign paperwork to head back to OpenAI, before Meta quietly elevated him to the role of chief AI scientist. The move underscored Mark Zuckerberg’s aggressive bid to steer Meta through the most consequential leadership reorganization in the company’s twenty-year history, as he seeks to accelerate the race to create personal superintelligence. The episode also illuminates the broader upheaval reshaping Meta’s senior ranks, with veteran executives—once the backbone of Zuckerberg’s trusted leadership—seeing their influence wane as a new cohort of technocrats asserts control over the company’s AI destiny.
The upheaval that set the stage for Meta’s AI reorganization
Meta’s latest corporate realignment is not a routine reshuffle; it is the most consequential overhaul of its AI leadership since the platform’s ascent to a global social technology giant. Zuckerberg, one of the last prominent founder-CEOs in Big Tech, has long relied on a cadre of longtime lieutenants to run the company’s most critical divisions. Chief Product Officer Chris Cox, for instance, has been a steady hand in directing the product roadmaps that intersect with Meta’s AI ambitions. Yet the AI race with rivals such as OpenAI and others has compelled Zuckerberg to pivot toward a new generation of leaders who can move with greater speed and operate with the audacity of a startup mindset within the sprawling construct of a thousand-plus product ecosystems.
The crisis point came quickly: Zhao’s sudden public flirtations with leaving after onboarding did not merely reflect a personality clash or a routine retention scare. It exposed the underlying tension between a veteran management culture and a newly assembled leadership team that is designed to push Meta’s AI research into markets faster and with more aggressive resource allocation. Zhao’s willingness to revert to his prior employer—an institution he helped elevate to the vanguard of AI innovation—signaled that the dynamics within Meta’s AI unit were shifting in ways that could not be ignored by Zuckerberg or his new allies. Four people familiar with the matter confirmed Zhao’s intent to depart, a development that triggered the formalization of his title as chief AI scientist only after he had already accepted the new role, illustrating the friction between the urgency of the engineering push and the politics of senior leadership. The timestamps of these events underscore a broader narrative: Meta’s leadership team is being rebuilt around the belief that the company needs a different operating tempo, governance model, and incentive structure to win in AI.
The leadership shake-up took place against the backdrop of Meta’s broader strategy to redefine its AI architecture. Zuckerberg has positioned Meta’s AI ambitions as a multi-year mission to achieve a level of personal superintelligence—an AI capability that can match or surpass human cognitive functions across a wide range of tasks. This ambition is not merely about creating more impressive models; it is about embedding AI into Meta’s products and services in a way that meaningfully enhances user experiences, monetization, and platform health. The reorganizations are thus not only about talent acquisition or retention; they are about reorienting decision rights, funding mechanisms, and performance metrics to align with a longer-term, high-risk, high-reward AI strategy.
In parallel with Zhao’s situation, Meta has been restructuring its AI organization multiple times in a relatively short window. The company has already completed several reorganizations within half a year, culminating in a four-team structure under a broader entity known as the Meta Superintelligence Lab (MSL). This is a substantial departure from earlier configurations, signaling a shift from a more siloed, functionally oriented model to a cross-cutting, project-centric framework designed to accelerate experimentation and productization of AI breakthroughs. The aim is to compress timelines from concept to deployment, even if that comes with growing pains as new teams negotiate roles, budgets, and expectations.
Industry observers and insiders have described the situation using colorful metaphor about “big men on campus”—referring to the new generation of AI leaders who carry outsized influence in a company that has long thrived on the authority of its senior veterans. The metaphor, while informal, captures a real dynamic: a prominent company is rebalancing power to a cohort that sees the future of AI not just as a research frontier but as a strategic product discipline that requires direct access to the C-suite and the Sun Belt of engineering resources that scale to thousands of developers and researchers. This shift is visible in the way the company has begun to redefine reporting lines, decision-making processes, and the allocation of computing power and data resources that fuel AI research at scale.
The speed of these changes is notable. In recent months, Meta has leveraged a hiring surge to lure AI researchers away from prominent rivals, including OpenAI and Apple, dangling nine-figure sign-on bonuses and access to substantial computing resources. The strategy is to outpace competing labs not just in the quality of research but in the speed of translating breakthroughs into tangible product features and consumer experiences across Meta’s family of apps. This competitive push has intensified internal scrutiny on governance, process efficiency, and the capacity of the company’s newly minted leadership to manage a sprawling research operation with dozens of teams working on cutting-edge AI initiatives.
The Financial Times and other outlets have reported on a steady drumbeat of departures or near departures among senior AI staff. The pattern includes both high-profile insiders and long-tenured scientists who had been central to Meta’s earlier AI agendas. The resulting churn is significant not only for morale but also for the company’s ability to maintain continuity in mission-critical programs as it navigates a period of rapid experimentation and market-driven expectations. While some former insiders have publicly described the new environment as offering unprecedented opportunities for researchers to shape Meta’s future, others have warned of growing pains—bureaucracy, competing resources, and the steep learning curve of diving into a corporate structure that is newly arranged around an ambitious, and at times ambiguous, strategic objective.
The new Generation of AI leaders and the TBD initiative
At the center of Meta’s most recent reorganization is Alexandr Wang, a Silicon Valley entrepreneur known for Scale AI, which built a significant data-labeling business used by many AI teams to train large language models and other machine-learning systems. Wang, just 28 years old, has become the focal point of the company’s ambitious effort to accelerate its AI roadmap. He leads the company’s most secretive new department, known internally as “TBD”—a shorthand for “to be determined.” This unit is described as an enclave filled with marquee hires, tasked with shaping Meta’s next generation of AI capabilities beyond the company’s current flagship models. In this setup, Wang’s leadership is seen as instrumental in coordinating the work across multiple teams and ensuring the new AI strategy translates into concrete product innovations and competitive advantages.
Zuckerberg personally placed Wang at the helm of the TBD group, aligning him with the broader objective of delivering faster and more decisive progress toward superintelligence. Wang’s appointment has been framed as a strategic coup by Meta’s leadership—a signal that the company intends to deploy a fresh playbook that values speed, integration, and market alignment over the more traditional, slower, product-centric approach of earlier years. The TBD unit is presented as a high-priority, risk-tolerant team whose mandate includes exploring frontier AI capabilities, evaluating new model architectures, and pushing beyond Meta’s previous boundaries in generative AI and cognitive systems.
Inside the TBD team, Zhao’s role as scientific lead had historically signaled a deep commitment to Meta’s long-term research vision. However, with Zhao’s threatened departure and the emergence of Wang’s leadership, the dynamic within the team has shifted toward a faster-paced, results-oriented workflow. Some insiders say Zuckerberg has been deeply involved in the day-to-day decisions of the TBD group, a level of direct oversight that some see as micromanagement. Others view this as critical governance necessary to align the ambitious goals of a private research venture with the operational realities of a public, profit-driven company.
The company’s strategic pivot extends beyond Zhao and Wang. Nat Friedman, a well-known entrepreneur who previously led GitHub, has been appointed to head Products and Applied Research. Friedman’s role centers on integrating the models developed by the AI teams into Meta’s core products so that users experience tangible benefits from advances in AI. This appointment is widely viewed as a bet on blending strong product leadership with deep research capabilities, ensuring that the company’s innovations have clear alignment with user needs, monetization opportunities, and scalability considerations across Meta’s sprawling ecosystem of apps and services.
The new leadership is also navigating a delicate balance of experience and disruption. On one hand, traditional executives like Chris Cox remain integral to Meta’s product strategy, and Yann LeCun continues to serve as chief AI scientist, preserving continuity in the company’s voice on AI research. On the other hand, several others in the AI hierarchy have seen their roles redefined or left without clear line responsibilities, creating a sense of both opportunity and insecurity among longtime staff. Ahmad Al-Dahle, who had previously led Meta’s Llama and generative AI efforts, has not been named as head of any particular team, signaling the possibility that leadership responsibilities in the AI domain are fragmenting into smaller, cross-functional units rather than being consolidated under a single umbrella.
Wang’s leadership style has been described by some as requiring rapid decision-making and a preference for direct accountability. This has occasionally clashed with the company’s established internal processes and resource allocation. Critics argue that moving too quickly within a megacorporation can risk misalignment or duplicative efforts as teams navigate new reporting lines and performance expectations. Supporters, however, emphasize that the AI landscape demands a more pragmatic, speed-driven approach—one that is more tolerant of initial friction in exchange for longer-term advantages in model quality, deployment speed, and competitive differentiation.
A central point of contention concerns the public-facing focus of Meta’s flagship Llama Behemoth model. In a notable strategic shift, Meta is no longer actively pursuing a public release of its baking-stage flagship model as initially planned. Instead, the TBD team has redirected its attention toward developing newer, cutting-edge models that can stay ahead of rivals. The decision to pause certain public releases is described by insiders as a tactical move to refine and optimize the company’s latest AI architectures before considering broader dissemination or applications. The expectations for producing a widely adopted consumer or developer-facing model appear to be aligned with Meta’s new emphasis on internal integration, enterprise adoption, and controlled, strategic rollouts.
The conversation around TBD’s mandate underscores a broader theme: the tension between ambitious, long-term research ambitions and the need to demonstrate short-term impact. While the team is designed to produce transformative AI capabilities, it must also satisfy investors, regulators, and the public that Meta can deliver tangible value without exposing itself to excessive risk. This tension will likely shape the pace at which the TBD unit advances, as well as the level of transparency the company is willing to share about the experimental models and their potential applications.
In parallel, the company has publicly sought to emphasize that the attrition among some AI staff is not unusual for an organization of Meta’s size. A spokesperson acknowledged that “some attrition is normal” and that many employees who have left had long tenures at the company. Meta framed departures as a natural part of a large organization’s evolution, while stressing that the company remains focused on its core objective of delivering personal superintelligence. The public narrative seeks to reassure investors and employees that while personnel changes are significant, they do not derail the overarching mission nor the company’s strategic commitments to AI development and product integration.
The attrition wave: departures, near-leavings, and what they signify
The artificial intelligence upheaval at Meta has been accompanied by a wave of departures, including several staffers who recently chose to leave after relatively brief tenures, according to people familiar with the matter. These movements suggest that the new leadership tempo and the reorganized structure are introducing a different culture of accountability and expectations that some veteran staff members find misaligned with their own career trajectories or with the company’s historical operating rhythms.
Among those departing is Ethan Knight, a machine-learning scientist who joined Meta only weeks prior to his decision to exit. His short tenure underscores the pressures of joining a high-stakes AI program that demands rapid alignment with new team practices, governance, and project priorities. Another departure, Avi Verma—a former OpenAI researcher—went through the company’s onboarding process but never reported for a first day, signaling the kind of early disconnect that can accompany a fresh integration into Meta’s evolving AI ecosystem.
Rishabh Agarwal, who started at Meta in April as a research scientist, announced his departure via a post on a social platform, citing a compelling pitch from Zuckerberg and Wang but explaining that he felt drawn toward another form of risk. The exact nature of that risk was not disclosed, but the move highlights how some high-caliber researchers feel pulled by opportunities that offer new challenges, different risk profiles, or more direct paths to leadership and impact.
In addition to these recent exits, Chaya Nayak and Loredana Crisan—two generative AI staffers with nine and ten years at Meta, respectively—were among the veteran employees who announced they were leaving in the days following Zhao’s departure hesitation. Media reports had already begun detailing some of Zhao’s exit plans, and the broader pattern of experienced staff departures reinforces the sense that Meta’s AI reorganization is producing a reshaping of who leads and who follows within the company’s AI program.
Meta’s leadership has acknowledged the chatter around internal changes and the intense scrutiny of every move in the AI effort. The company’s public posture emphasizes a persistent focus on delivering “personal superintelligence,” while simultaneously managing the reality that internal dynamics—ranging from power realignments to tension with long-tenured staff—will continue to evoke questions from employees, investors, and industry observers. The departures, while significant on their own, are part of a broader narrative about Meta’s willingness to reallocate talent, reframe job roles, and redesign incentive structures to accelerate advancement toward its ambitious AI goals.
The four-team structure, LeCun, and the redefined leadership lines
The organizational rework culminates in Meta’s AI group being reorganized into four distinct teams within the renamed Meta Superintelligence Lab (MSL). This represents the fourth such overhaul within about six months, signaling both an urgency to align with a fast-moving AI landscape and a willingness to experiment with governance models that can support rapid iteration. The four-team configuration is intended to distribute responsibilities more coherently, maximize collaboration across functions, and reduce the bottlenecks that can arise in a flatter, less segmented structure when dealing with complex AI systems at scale.
At the very top of Meta’s AI hierarchy sits Wang, whose blend of commercial acumen and technical ambition has become a focal point for the company’s strategy. He is widely viewed as the individual who can translate cutting-edge research into commercially viable, consumer-facing solutions integrated into Meta’s apps. Wang’s leadership is coupled with his direct connection to Zuckerberg, a dynamic that has drawn attention because it signals a high level of executive engagement in the day-to-day progress of the AI program. Zuckerberg’s involvement has led some observers to describe the arrangement as a direct, hands-on collaboration in the shaping of Meta’s AI destiny—a posture that is both unusual for a top-level tech CEO and emblematic of the magnitude of the company’s AI bets.
The company has also navigated the complexity of reporting structures. Yann LeCun, Meta’s chief AI scientist, remains in the role but now reports to Wang—a shift that consolidates some strategic direction within the new TBD framework while preserving LeCun’s authority as a principal researcher and thought leader in AI. Ahmad Al-Dahle, who previously led Llama and other generative AI initiatives, has not been named as head of any particular team, indicating that the leadership structure may be moving toward a more distributed model with multiple unit heads rather than a single, centralized head of AI.
The product leadership side has seen changes as well. Chris Cox, still listed as chief product officer, does not oversee the generative AI domain directly any longer; instead, Wang has an elevated position in the internal hierarchy with direct reporting lines to Zuckerberg. Cox remains “heavily involved” in broader AI efforts and continues to oversee Meta’s recommendation systems, but the shift reduces the direct line of accountability for generative AI within Cox’s purview. This realignment has sparked conversations about how product leadership collaborates with AI teams and who has the final say on roadmap prioritization, resource allocation, and performance metrics for AI-enabled features.
In this context, Meta’s leadership contends that the internal tensions and realignments are a necessary consequence of pursuing a more aggressive AI agenda. The company argues that a high-stakes, cross-functional initiative—designed to push Meta toward superintelligence—requires a governance model that can accommodate rapid decision-making, close executive oversight, and a capacity to reallocate incentives and resources in line with strategic priorities. Yet critics caution that reconciling the ambitions of a startup-like “TBD” with the accountability expectations of a global public company is among the most delicate balancing acts in corporate governance.
One critical question being asked inside Meta and among external observers relates to the pace of hiring and the potential for cuts within the AI teams. A memo circulated among managers indicated that the company would temporarily pause hiring across all Meta Superintelligence Labs, with the exception of business-critical roles. The memo also stated that new hires would be evaluated on a case-by-case basis as leadership plans for 2026 headcount growth. This move suggests Meta wants to consolidate and calibrate its AI labor force, prioritizing roles that are essential to the core AI strategy while delaying or deferring hires that could lead to resource inefficiencies or misalignment with the redefined goals. The freeze is presented as a prudent step to facilitate careful planning and reduce the risk of overextension as the teams retool around the TBD initiative and the broader mission of delivering personal superintelligence.
The organizational changes underscore a broader strategic imperative: Meta is attempting to fuse the best elements of startup agility with the disciplined governance, scale, and accountability of a mature tech enterprise. Zhao’s fast ascent and dramatic potential exit, Wang’s bold leadership of TBD, Friedman’s product-centric governance, LeCun’s ongoing role, and the retention of Cox as a product leader—all of these elements reflect a company that wants to preserve continuity where possible while injecting new energy into the AI program. The tension between speed and governance, ambition and practicality, experimentation and measured progress will likely define Meta’s AI trajectory for the foreseeable future.
Strategic ambition, milestones, and the path ahead
Meta’s AI program is anchored in a high-stakes objective: to achieve superintelligence—the capacity for AI to outperform human capabilities across a broad spectrum of cognitive tasks. The company’s leadership asserts that the progress toward this goal requires not just technical breakthroughs but a reimagined organizational model that enables researchers and product teams to operate with the speed and focus of a startup, while still leveraging the resources, data access, and scale that only a company like Meta can provide. This dual mandate—innovative speed coupled with strategic discipline—has defined the leadership changes and the ongoing debate about how best to structure and fund simulation environments, training pipelines, and deployment processes.
The decision to pivot away from publicly releasing Llama Behemoth to the public is illustrative of Meta’s recalibration toward internal experimentation and controlled dissemination. By diverting the team’s attention to new models within TBD, Meta signals that it is prioritizing model performance, safety, governance, and integration capabilities over headline releases. The “to be determined” designation for the new department is a deliberate acknowledgment that the roadmap for Meta’s next generation of AI systems remains in flux—dependent on new insights, hardware capabilities, regulatory considerations, and the evolving competitive landscape. The focus on newer models rather than the immediate public rollout of flagship models also reflects Meta’s strategic calculus regarding adoption, user experience, and monetization potential.
Inside the leadership circle, there is visible concern about aligning Wang’s strategic vision with Zuckerberg’s expectations for speed and results. The reported friction over timelines and the balance between long-term strategic planning and the near-term pace of progress indicates that the company is wrestling with how to translate ambitious, sometimes audacious goals into concrete milestones. The dynamic is not simply about one person’s leadership style; it is about how Meta defines success in AI, what constitutes progress, and which incentives will drive teams to deliver on aggressive targets while maintaining the quality and reliability expected from a platform with hundreds of millions of users.
The broader market context intensifies these internal debates. Meta is competing with rival labs that have also been aggressively recruiting and investing in AI talent. The company’s nine-figure sign-on bonuses and the promise of significant compute resources reflect a competitive arms race in the AI domain. This external pressure amplifies the internal tensions that come with reassigning responsibilities, redefining budgets, and recalibrating measurement metrics across the AI organization. In such a climate, the leadership’s ability to maintain momentum, ensure clear accountability, and demonstrate measurable progress will be essential to sustaining confidence among investors, employees, and users alike.
A key operational implication of the reorganization is how Meta plans to manage its compute capacity and access to infrastructure. The company has publicly claimed that TBD Labs will boast the greatest compute-per-researcher ratio in the industry, and several insiders have described the unit as a major center for compute-intensive experimentation. If this promise holds, the company could realize accelerated iteration cycles, allowing for more rapid prototyping, testing, and refinement of AI models that could be deployed within Meta’s ecosystem. Yet this advantage must be balanced against the need to prevent bottlenecks—whether in hardware provisioning, data governance, or cross-team collaboration—that could arise from the concentration of power, resources, and decision rights in a single, heavily prioritized unit.
The leadership changes also imply a shift in how Meta will measure success in AI. The new structure emphasizes productization and integration with consumer apps, making it essential that the AI initiatives deliver concrete usability improvements and value across Meta’s platforms. Friedman’s appointment to lead Product and Applied Research is intended to ensure that the models developed by TBD and other AI teams translate into features and experiences that resonate with users, improve engagement, and create new monetization opportunities. The alignment of research excellence with product impact is central to Meta’s strategy, and the leadership’s ability to demonstrate this alignment will be critical in maintaining investor confidence as the company navigates a year of substantial reorganization and ambitious AI ambitions.
In summary, Meta’s AI leadership overhaul embodies a bold bet on a new generation of leaders, a redefined organizational architecture, and a heightened commitment to rapid experimentation in pursuit of personal superintelligence. Zhao’s volatile start, Wang’s high-stakes leadership of TBD, Friedman’s product-centric governance, LeCun’s continued but redistributed role, and Cox’s retention as a product authority collectively illustrate a company in the midst of a difficult but potentially transformative transition. The coming months will test Meta’s ability to reconcile speed with governance, ambition with practicality, and the allure of frontier AI with the realities of building a large, globally scrutinized platform. As the company navigates potential hiring pauses and the strategic planning required for 2026 headcount, the AI program’s trajectory will likely become a bellwether for Meta’s broader strategic positioning in a competitive and rapidly evolving tech landscape.
Conclusion
Meta is undergoing one of its most consequential leadership transformations in two decades as it presses forward with a high-stakes AI strategy centered on achieving personal superintelligence. The arrival, and in some cases the tentative departure, of top AI talent—especially Shengjia Zhao—has illuminated the friction between a veteran leadership culture and a fast-moving, startup-like cadre of new AI leaders. The creation of the Meta Superintelligence Lab, the formation of the TBD unit led by Alexandr Wang, and the appointment of Nat Friedman to head Products and Applied Research signal a deliberate shift toward integrating rigorous research with product-focused execution. While veterans such as Yann LeCun remain central to the AI vision, redefined reporting lines and governance changes reflect Meta’s commitment to ensuring that the AI programs are not only innovative but also scalable, integrated, and aligned with the company’s long-term business goals. The wave of departures among experienced AI staff underscores the inherent tension in attempting to accelerate progress within a complex corporate structure. Meta’s challenge will be to sustain momentum through the inevitable growing pains of a major organizational shift while maintaining focus on delivering meaningful user experiences and value to shareholders. The next chapter of Meta’s AI journey will hinge on how effectively the TBD initiative can produce breakthrough capabilities, how well the product leadership translates those breakthroughs into real-world features, and how the company manages talent, resources, and governance to realize the promise of superintelligence without compromising reliability, safety, or user trust.
