In this expert opinion piece for The Fast Mode, Kamiwaza Field CTO James Urquhart explores the operational challenge posed by "ephemeral agents" — AI workers that appear unannounced, run for seconds or minutes, and vanish, sometimes spawning more agents as they go. This ephemerality is efficient and adaptive, but without shared rules of engagement it becomes chaotic, leaving enterprises to cobble together bespoke integrations that fragment and fail to scale. Urquhart details the real costs of this fragmentation: wasted engineering effort reinventing connectors, security teams unable to enforce policy on processes that vanish before they can be logged, compliance gaps as agents cross jurisdictions, and pilots that collapse under production demands. He draws a historical parallel to the era before TCP/IP unified networking, when incompatible vendor protocols stifled scale until standards emerged. Composability — treating agents, models, and data sources as interchangeable parts — is presented as a valuable first step that lowers experimentation cost and reduces vendor lock-in, but only a partial answer without common interfaces. Early protocols like Model Context Protocol (MCP) and agent-to-agent (A2A) hint at what standardized interfaces could look like, yet neither was designed for the many-to-many coordination that enterprise swarms will require. Urquhart argues the field needs message queues with backpressure, portable memory and context, observability for millisecond-lifespan processes, and identity and policy enforcement built for ephemerality. The piece asks who will define agentic AI's "Kubernetes moment." Read the full article on The Fast Mode.
Source: The Fast Mode — Read the full article
