In this InfoWorld feature, Kamiwaza Field CTO James Urquhart argues that agentic AI fundamentally breaks the assumptions underpinning traditional IT operations. For decades, ops has been organized around keeping applications running, with success measured in uptime. Agentic AI upends that: agents are ephemeral, spinning up in response to a prompt or another agent's request, performing a task, and then disappearing — sometimes spawning further agents in cascading, short-lived swarms. Urquhart memorably likens this to "Spirit Halloween, not Macy's." Because tools like Kubernetes were built for long-running, persistent workloads, they are optimized for a different problem than managing constant appearance and disappearance. His proposed mental model separates "capacity" (infrastructure and data) from "consumption" (the agents and models that use it), with "inference" as the interface between them. Framing operations this way, he argues, prevents brittle one-to-one bindings between agents and specific infrastructure, enables enterprises to adopt new models without ripping apart their stack, and lets swarms of agents emerge, work, and dissolve without a bespoke playbook each time. The article explores the hard open challenges this raises — shared memory and context across ephemeral processes, real-time observability for workloads that may not persist long enough to log, and jurisdiction-specific compliance through composability. It's a clear articulation of the operational thinking behind Kamiwaza's platform. Read the full article on InfoWorld for the complete argument.
Source: InfoWorld — Read the full article