Organizational Memory for Agentic Business Process Execution
06:00 · July 7, 2026 · arXiv cs.AI RSS
LLM-based agents offer new opportunities for automating business process execution beyond the limits of rule-based systems. However, general-purpose LLMs lack the organization-specific knowledge required for reliable execution, which is typically fragmented across human-oriented artifacts such as policies, process models, and standard operating procedures. While such knowledge can technically be encoded in individual prompts or agent-specific retrieval setups, this approach does not scale in enterprises, as it gives rise to knowledge silos and rule duplicates, and makes consistent updates and learning across agents difficult. We argue that this calls for an organizational memory for agentic business process execution: a shared, governed, and agent-consumable reference layer of evolving organization-specific procedural knowledge about how work should be executed. We derive requirements for such a memory, propose an architecture for its curation and consumption, and demonstrate its effectiveness in a proof-of-concept based on a procurement scenario.
Summary
LLM-based agents can interpret natural language and manage multi-step tasks with exception handling, offering a route to automating business processes that resist traditional rule-based automation such as robotic process automation. General-purpose models, however, lack built-in access to organization-specific knowledge such as policies, role structures, system landscapes, and exception-handling conventions that are typically scattered across BPMN models, standard operating procedures, and policy documents designed for human readers.
Encoding this knowledge in per-agent prompts or isolated retrieval setups quickly produces knowledge silos, duplicated rules, and inconsistent updates when many agents operate across evolving processes. The authors therefore propose an organizational memory: a single, human-governed, agent-consumable reference layer that integrates heterogeneous sources into a unified representation, supplies context at runtime, and evolves with organizational change while remaining subject to oversight.
The paper derives concrete requirements for curation, retrieval, governance, and lifecycle management, then outlines an architecture that separates knowledge ingestion and maintenance from agent consumption. A procurement scenario serves as the running example and proof-of-concept, showing how the shared memory enables consistent invoice-matching and purchase-order decisions across multiple agents without duplicating procedural rules.
Why it matters
This research is highly relevant for Dutch AI practitioners and researchers focusing on enterprise AI adoption and multi-agent systems. It provides a scalable, governed architecture for integrating organization-specific knowledge into LLM agents, aligning well with the Dutch market's emphasis on reliable and transparent AI deployment in business contexts.
