Organizations are increasingly encoding operational procedures as multi-step AI agent skills. The paper calls these Agent Operating Procedures (AOPs) and notes they must satisfy several constraints: traceability, auditability, reproducibility, access control, and bounded time and cost.
To support these requirements, the paper proposes normal forms and an algebra for AI agent workflows. The normal forms are intended to provide a canonical structure, while the algebra offers operations for combining and constraining procedures. The abstract does not detail the specific operations, but the framing suggests a formal foundation for workflow design.
The work is relevant to any organization deploying AI agents for routine operations, as it addresses the need for governance and control. The abstract is brief, so the full technical content is not yet available from the excerpt.