Agentic AI Systems Need Stable Abstractions Like POSIX and Kubernetes
A recent study published on arXiv suggests that agentic AI systems—autonomous agents powered by large language models that are capable of planning, utilizing tools, retaining memory, and collaborating—are currently in a developmental stage similar to the formative years of traditional operating systems and cloud technology. The authors compare this to POSIX for classical operating systems and Kubernetes for cloud computing, highlighting their evolution from various competing frameworks to established standards. They argue that without a shared understanding of fundamental abstractions and their assurances, the creation of portable agentic applications, reliable platform composition, and progress beyond prototype stages will be hindered. The paper, referenced as arXiv:2607.25076v1, advocates for an 'Agent Operating System' to unify the numerous existing protocols and frameworks.
Key facts
- Paper is on arXiv with ID 2607.25076v1
- Compares agentic AI to classical OS and cloud computing
- POSIX and Kubernetes are cited as examples of stable abstraction platforms
- Agentic AI systems are described as autonomous, LLM-driven agents that plan, use tools, maintain memory, and collaborate
- Dozens of frameworks and protocols exist but no community consensus on core abstractions
- Without consensus, agentic applications cannot be written portably
- Platforms cannot compose reliably without stable abstractions
- The field cannot advance beyond prototype deployments without consensus
Entities
Institutions
- arXiv