AI Agents Give Enterprises 20 Months to Rebuild Infrastructure

What happened

Barak Yagour, Meta's VP of Engineering responsible for data infrastructure, used a keynote slot at VB Transform 2026 to issue a pointed warning: enterprise systems were designed for human users, and the rapid rise of AI agents is exposing that assumption as dangerously outdated. Yagour estimated organisations have roughly 20 months to rebuild their infrastructure before the mismatch becomes critical.

The scale of the shift he described is striking. Agentic queries hitting Meta's own data systems grew 30-fold in a single half-year period. Yagour also cited broader internet traffic data showing that automated traffic surpassed human traffic in 2024, reaching 51% of total internet traffic according to Imperva's 2025 Bad Bot Report, and is expanding approximately eight times faster than human-generated traffic, per HUMAN Security's 2026 State of AI Traffic report.

Why it matters

For customer experience practitioners, the infrastructure conversation is not abstract. Every service touchpoint — from a chatbot resolving a complaint to an AI agent booking a reservation or processing a return — runs on the same enterprise plumbing Yagour is describing. If that plumbing was optimised for human click-rates and human session lengths, it will behave unpredictably when agents begin acting on customers' behalf at machine speed and volume. The failure modes are not theoretical; they show up as dropped transactions, inconsistent personalisation and eroded trust.

From a behavioural economics perspective, the more consequential shift is one of agency itself. Customers are increasingly delegating decisions to AI intermediaries, which means the moment of emotional engagement — the point where loyalty is won or lost — may no longer involve a human at all. Service designers who still frame journeys around human attention and human decision cadence are, in effect, designing for a shrinking share of actual interactions.

By the numbers

• 30x — growth in agentic queries hitting Meta's data infrastructure in a single half-year period
• 51% — share of total internet traffic that was automated in 2024, according to Imperva's 2025 Bad Bot Report
• ~8x faster — the rate at which automated traffic is growing compared with human traffic, per HUMAN Security's 2026 State of AI Traffic report
• ~20 months — Yagour's estimated window for organisations to meaningfully rebuild their infrastructure for agentic demands

The Renascence take

Most CX leaders will read this story as an IT story and hand it to their engineering teams. That would be a mistake. The 20-month clock Yagour is describing is also a customer-strategy clock — and the organisations that treat it purely as a plumbing problem will find themselves with fast infrastructure and no coherent experience to run on top of it.

The deeper issue is not whether your servers can handle agent traffic — it is whether your service model was designed with agent-mediated customers in mind at all. When an AI agent acts on a customer's behalf, the traditional levers of experience design — friction, delight, emotional resonance — need to be re-engineered for a non-human decision-maker that optimises on different signals entirely. Customer-obsessed operators should be mapping their journeys right now to identify every step that currently relies on human attention, human patience or human emotional response, and asking honestly: what happens to this moment when an agent is doing the navigating? That is where the real redesign work begins.