Forrester published a reference design it calls the Agentic Runtime Architecture, a proposed shared execution layer for running AI agents that come from different vendors under central control. It argues that request-response and nightly-batch architectures cannot govern hundreds of semi-autonomous agents, and that guardrails written into individual agents force a rewrite with every policy change. The blog outlines the components; the full report is limited to Forrester clients.
What changed
Forrester's agentic guidance covered platform selection and lifecycle design rather than a runtime layer.
What it unlocks
Assessing agent plans against a named reference architecture for running agents from many vendors under one governed runtime.
- nine core components
- seven design principles
- five imperatives
What you need to act on it
- Forrester client access for the full report
Sources