TechCrunch: Cloud infrastructure is being redesigned for AI agents, not human users

Cloud infrastructure providers are redesigning their systems to accommodate AI agents rather than human users. AWS launched a next-generation OpenSearch Serverless that decouples compute from storage, enabling instant scaling during agent activity and scaling to zero when idle. The motivation is that agents create sudden, unpredictable traffic patterns — spawning sub-agents that simultaneously query databases and APIs — then go quiet. Traditional cloud infrastructure was optimized for the steady, predictable behavior of humans who click, stream, and search.

Cloudflare reports that non-human traffic already accounts for 31% of HTTP activity, with AI crawlers making up roughly 25% of bot requests. One analyst cited in the article predicts that non-human traffic will exceed human traffic in the first half of 2027.

For product managers building products with AI agents as first-class users — internal automation tools, agentic pipelines, or multi-agent orchestration systems — this has several implications. Products that rely on agent-driven workflows need infrastructure planning that anticipates burst demand and idle periods, rather than steady growth curves. Cost models based on steady-state usage will underestimate actual billing once agents are in production.

The article also notes that the gap between infrastructure designed for human users and infrastructure designed for agents is becoming a differentiator. Products that integrate early with agent-optimized backends — decoupled compute, vector databases, pay-per-use scaling — may face fewer capacity and cost surprises as their agent workloads grow. Pay-per-use pricing is gaining ground over fixed-cost infrastructure precisely because agent activity is so uneven.