# Meta's 20-month transformation plan for AI agents Companies must rebuild infrastructure to meet the needs of agentic AI systems. - **Source:** [VentureBeat](https://venturebeat.com) - **Published:** 2026-07-15 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/466a26ab-3f45-43fd-8740-763f3ed76ab9 ## The rise of AI agents Agentic queries to Meta's data systems grew 30x in six months, challenging two decades of human-centric infrastructure assumptions. Automated systems now generate 51% of internet traffic, with this share growing rapidly. ## Three infrastructure breaking points Meta's Barak Yagour identifies simultaneous fractures in capacity, identity and velocity assumptions. One engineer can now spawn 10 agents with hundreds of subagents. Authentication systems weren't built for non-human entities. ## Dynamic control mechanisms Rather than blocking agent traffic, Meta is making infrastructure 'agent-aware' with hierarchical controls, cost attribution and priority-based throttling. This manages loads equivalent to 100,000 users appearing overnight. ## Trusted data environments Meta balances agent autonomy with security through 'trusted data environments'. Every output is traceable, sensitive data is masked, and access requests are evaluated in real-time under a 'broad exploration, narrow release' principle. ## Impact of reasoning models As AI shifts from correlation to causal reasoning, data demands intensify. Meta invests in real-time streaming and schema-aware storage. The old 'opaque data blob' approach led to GPU inefficiencies from overfetching. ## The urgency of transformation Yagour warns companies have about 20 months to prepare for agentic AI. 42% of Instagram users want fundamental algorithm changes. Fully conversational recommendation systems move beyond keyword matching to intent reasoning. ## Sources covering this story - https://venturebeat.com/data/we-have-maybe-20-months-to-rebuild-for-ai-agents-metas-infrastructure-vp-tells-vb-transform-2026 — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._