Meta VP warns of urgent need to adapt infrastructure for AI agents

Meta VP warns of urgent need to adapt infrastructure for AI agents

Barak Yagour says enterprise systems were built for humans, not autonomous agents, and the cracks are showing fast

Meta’s VP of Engineering Barak Yagour took the stage at VB Transform 2026 wearing Ray-Ban Meta AI glasses. His core message was blunt: the infrastructure powering most organizations today was designed for human users, and it simply cannot handle the demands of autonomous AI agents operating at scale.

Meta’s agent infrastructure push

In June 2026, Meta launched its Business Agent platform, designed for rapid enterprise deployment. The platform integrates with tools like Shopify and Zendesk, enabling AI agents to handle customer queries, scheduling, and other operational tasks without requiring months of custom engineering.

A month later, in July 2026, Meta released Muse Spark 1.1, an advanced multimodal model built specifically for agentic reasoning. The model improves tool usage, coding capabilities, and the ability to complete multi-step tasks autonomously.

Advertisement

Yagour has been with Meta since at least 2019, moving from Director of Data Infrastructure to his current VP role.

AgentPerf, a new industry benchmark introduced in June 2026 to specifically measure agentic workloads, has started producing eye-opening numbers. NVIDIA reported that modern hardware can support up to 20 times more agents per megawatt compared to previous generations.

Where crypto enters the picture

In February 2026, Virtuals Protocol announced updates to its Agent Commerce Protocol, focusing on on-chain agent tokenization. Instead of agents living inside a walled garden, they exist as tokenized entities on a blockchain, capable of transacting, earning, and being owned in a permissionless environment.

What this means for investors

The NVIDIA efficiency data is particularly significant for crypto. If you can run 20 times more agents per megawatt on modern hardware, the economics of running decentralized agent networks become dramatically more attractive.

AgentPerf benchmarks will eventually be applied to on-chain agent systems, and when that happens, the market will have real data to price the gap between promise and performance.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.

Meta VP warns of urgent need to adapt infrastructure for AI agents

Meta VP warns of urgent need to adapt infrastructure for AI agents

Barak Yagour says enterprise systems were built for humans, not autonomous agents, and the cracks are showing fast

Share

Add us on Google

Meta’s VP of Engineering Barak Yagour took the stage at VB Transform 2026 wearing Ray-Ban Meta AI glasses. His core message was blunt: the infrastructure powering most organizations today was designed for human users, and it simply cannot handle the demands of autonomous AI agents operating at scale.

Meta’s agent infrastructure push

In June 2026, Meta launched its Business Agent platform, designed for rapid enterprise deployment. The platform integrates with tools like Shopify and Zendesk, enabling AI agents to handle customer queries, scheduling, and other operational tasks without requiring months of custom engineering.

A month later, in July 2026, Meta released Muse Spark 1.1, an advanced multimodal model built specifically for agentic reasoning. The model improves tool usage, coding capabilities, and the ability to complete multi-step tasks autonomously.

Advertisement

Yagour has been with Meta since at least 2019, moving from Director of Data Infrastructure to his current VP role.

AgentPerf, a new industry benchmark introduced in June 2026 to specifically measure agentic workloads, has started producing eye-opening numbers. NVIDIA reported that modern hardware can support up to 20 times more agents per megawatt compared to previous generations.

Where crypto enters the picture

In February 2026, Virtuals Protocol announced updates to its Agent Commerce Protocol, focusing on on-chain agent tokenization. Instead of agents living inside a walled garden, they exist as tokenized entities on a blockchain, capable of transacting, earning, and being owned in a permissionless environment.

What this means for investors

The NVIDIA efficiency data is particularly significant for crypto. If you can run 20 times more agents per megawatt on modern hardware, the economics of running decentralized agent networks become dramatically more attractive.

AgentPerf benchmarks will eventually be applied to on-chain agent systems, and when that happens, the market will have real data to price the gap between promise and performance.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.