Agentic AI
New Framework Proposes Treating AI Agents as First-Class Services
Researchers introduce Agentic Services Computing to engineer and manage autonomous LLM-based agents as reusable service entities.
A new research framework aims to bring the disciplined engineering practices of services computing to autonomous AI agents, marking a conceptual shift in how intelligent systems might be built and deployed.
The Emerging Paradigm#
Services computing has a long track record—from early Web services through microservices, cloud-native platforms, and serverless architectures. Each generation refined how software functions could be described, reused, composed, and governed. Researchers now argue that LLM-powered agents represent a fundamentally new form of service that doesn't fit neatly into these existing models.
What Makes Agentic Services Different#
Unlike traditional microservices that execute predefined functions when called, autonomous agents operate differently. They accept high-level goals, maintain internal state, reason about context, and use external tools to act independently. They can collaborate with other agents and operate effectively in open, dynamic environments. This autonomy introduces novel challenges around accountability, governance, and reliable orchestration that classic services computing frameworks don't adequately address.
The Proposed Framework#
Researchers have developed Agentic Services Computing (ASC) as a service-oriented lens for autonomous agents. Rather than treating agents as a separate computational paradigm, ASC extends traditional services principles to encompass goal-driven, stateful, tool-mediated autonomous behavior. The framework organizes knowledge around a service lifecycle—covering how agents are defined, composed, deployed, monitored, audited, and improved over time.
Why This Matters#
Formalizing agents as first-class services could unlock systematic engineering practices: clear service descriptions, standardized composition patterns, reliable delivery pipelines, comprehensive observability, and governance mechanisms. This would enable organizations to manage autonomous agents with the same discipline and confidence they apply to conventional microservices architectures, ultimately accelerating the maturation of multi-agent systems in production environments.
The research draws on existing work on LLM agents and multi-agent systems while establishing a roadmap for addressing open challenges unique to this domain.
