Every major wave of platform software follows a similar arc: an initial experimentation phase with competing frameworks and ad-hoc implementations, followed by the establishment of a small set of stable abstractions with well-defined semantics, and finally a consolidation around those abstractions that allows applications to be portably targeted.
For instance, POSIX did this for classical operating systems, while Kubernetes achieved it for cloud environments. Currently, agentic AI systems—autonomous, LLM-driven agents that plan, use tools, maintain memory, and collaborate—are in the experimentation phase of the third wave. Numerous frameworks and protocols have surfaced, yet there is no consensus on the core abstractions or the guarantees they entail.
Without such consensus, agentic applications cannot be written portably, platforms cannot compose reliably, and the field cannot progress beyond prototype deployments.
We argue that the way forward is to adopt the methodology of prior waves: derive new agentic abstractions by extending classical OS and cloud OS primitives to stochastic, natural-language-mediated execution, specify their semantics precisely, and consolidate around them—just as POSIX and Kubernetes consolidated their respective waves.
Blogger's Review: This article clearly identifies the lack of consensus in the agentic systems landscape and proposes a solution by leveraging the experiences of classical and cloud operating systems. This approach could significantly aid in standardizing and enhancing the portability of agent applications, making it a topic worth following.