Cloud infrastructure designed for predictable human behaviour—search queries, clicks, scrolling—is fundamentally misaligned with how AI agents operate. Agents generate unpredictable traffic spikes, spawn sub-agents that query hundreds of databases simultaneously, then vanish just as quickly, leaving infrastructure idle and expensive. AWS's redesigned OpenSearch Serverless addresses this by decoupling compute from storage, allowing systems to scale to zero when agents aren't active rather than maintaining reserved capacity. The broader industry is responding in kind: Databricks and Snowflake are repositioning as AI memory systems, Microsoft is updating Azure for agent bursts, and Cloudflare has launched infrastructure for persistent agent environments. This infrastructure redesign matters because it directly affects deployment costs and feasibility at scale. As non-human traffic is projected to exceed human traffic by mid-2027, and with AI crawlers already comprising roughly a quarter of all bot requests, the economics of running agents will shift dramatically. For CX teams already deploying agents—whether through Zendesk's Custom Agents or competing platforms—this infrastructure evolution means the cost-per-interaction will decline, making agent deployment viable for use cases previously considered too expensive.
The critical question for CX professionals is whether your current platform vendors are architected to benefit from this infrastructure shift, or whether they'll become cost-inefficient as agent workloads scale. Teams running Zendesk's agent investments or considering Salesforce's Agentforce need to understand whether these platforms are built on infrastructure that scales elastically with agent traffic, or whether they'll inherit the idle-compute tax that plagued earlier generations. Vendors who've invested in agent-native architecture—those integrating with platforms like Vercel or building on serverless-first infrastructure—will have a structural advantage as agent deployment accelerates. Smaller CX platforms without deep infrastructure partnerships may find themselves at a disadvantage, unable to offer the cost efficiency that enterprise buyers will increasingly demand. The infrastructure layer is becoming a competitive moat, and it's one that's largely invisible to end users until it affects pricing and performance at scale.
As AI agents move from experiments to production, AWS, Cloudflare, and others are redesigning cloud infrastructure for a future dominated by machine-generated internet traffic instead of human users.