Meta VP Warns of Urgent 20-Month Deadline to Overhaul Infrastructure for AI Agents
At the recent VB Transform 2026 conference, Barak Yagour, Meta’s VP of Engineering for Data Infrastructure, issued a stark warning to the tech industry: the digital foundations built over the last two decades are fundamentally ill-equipped for the era of agentic AI. With autonomous agents rapidly becoming the primary consumers of data, Yagour noted that traditional enterprise systems are already showing signs of collapse. As detailed in reports from
The urgency stems from a 30-fold surge in agentic queries hitting Meta’s systems in just six months. This trend reflects a broader global shift; according to the 2025 Bad Bot Report by Imperva, automated traffic has already overtaken human internet activity, now accounting for 51% of all traffic. Yagour highlighted that the industry is at a crossroads, stating, "We spent 20 years building infrastructure for humans. We have maybe 20 months to rebuild the whole thing for a world where humans and agents co-create at scale."
Capacity Overload: Traditional infrastructure assumes one engineer equals one unit of load. Today, a single engineer can spawn 10 agents, each capable of spawning subagents, creating a massive, unpredictable surge in resource demand that traditional metrics fail to capture.
Identity Crisis: Current access control systems are designed for human users or static services. Autonomous agents, which make independent decisions and traverse vast datasets, do not fit into these existing categories, necessitating new governance and identity frameworks.
Velocity Bottlenecks: While tools like GitHub Copilot allow agents to write nearly half of a user's code, the downstream CI/CD pipelines and testing environments have not accelerated at the same pace, creating a massive bottleneck in software delivery.
Reasoning-First Architecture: Meta is moving away from batch-processed data. To support "reasoning" models that require full behavioral history rather than summarized signals, the company is shifting toward real-time streaming pipelines and schema-aware storage.
Trusted Data Environments: To allow agents to work autonomously without risking security, Meta is implementing "trusted data environments" where agents can explore data broadly, but every action is traced, sensitive fields are masked, and outputs are scrutinized in real time.
For industry leaders and engineers tracking these shifts, independent analysis and case studies from the event can be further explored via
