At the VB Transform 2026 conference, Barak Yagour, Meta's VP of Engineering leading data infrastructure, issued an urgent warning that enterprises have only about 20 months to rebuild their entire technology infrastructure. According to Yagour, current data systems built over the last two decades were designed for humans, but the surge of autonomous AI agents is completely disrupting these foundations.
Diễn biến chi tiết
According to VentureBeat, Meta is witnessing a profound structural shift as agentic queries hitting its data systems grew 30x in a single half. This trend is not confined to Meta. Citing Imperva's 2025 Bad Bot Report, automated traffic overtook human traffic to reach 51% of total global internet traffic, growing roughly eight times faster than human traffic. Yagour emphasized that global infrastructure is at an inflection point where AI agents, rather than humans, are becoming the primary consumers of resources.
Bối cảnh & Nguyên nhân
The boom in AI agents is breaking three core assumptions that system engineers have long relied on: capacity, identity, and velocity. Regarding capacity, one engineer used to equal one unit of load, but now one engineer can spawn 10 agents, each spawning sub-agents, generating the load of 100,000 users overnight. Regarding identity, traditional access controls do not fit entities that make independent decisions but are neither human users nor deployed services. Finally, while tools like GitHub Copilot write code faster, the subsequent CI/CD pipeline does not accelerate automatically to keep up.
Phân tích kỹ thuật & Công nghệ
To address the immense pressure on the data layer, Meta is deploying "trusted data environments." Within these environments, agents can explore freely, but every output is traced back to its source, and sensitive fields are masked in real time. Concurrently, Meta is replacing batch ETL with real-time streaming for its ranking and recommendation pipelines. To prevent GPU starvation, Meta's storage is becoming schema-aware, fetching only the specific columns needed rather than loading massive opaque data blobs. The company is targeting 500 million queries per second and 1 petabyte per second of throughput for training data reads.
Ý kiến chuyên gia & Nhận định
According to Yagour, technology is not moving linearly but in a self-reinforcing flywheel: "Agents make data more accessible. Better data makes reasoning. Reasoning creates new demands that push agents and infrastructure forward." When asked whether traditional file systems and SQL will be replaced by neural storage, the Meta executive revealed the company is experimenting at every level with its multi-exabyte scale storage systems.
Tác động & Tương lai
Meta's warning sends a powerful message to the global and Vietnamese tech communities. Businesses cannot continue to run systems with legacy mindsets as the era of co-creation between humans and agents at scale approaches. Building agent-aware infrastructure with dynamic controls and cost attribution will be crucial for organizations to survive under the weight of the new AI era.