Agentic Orchestration: Long-Running AI Agents in Enterprise Architectures
by Bernd Ruecker
AI is changing the world – but so far, its impact often seems limited to boosting personal productivity. But what if we want to embed AI into the heart of core business processes, also in regulated industries? How can we operationalize AI without losing control, oversight, or governance?
In this talk, we explore Agentic Orchestration: enabling an AI agent to act dynamically within the well-defined guardrails of BPMN-based process models.
By exposing BPMN constructs (such as service invocations, flows, or subprocesses) as “tools” to an agent, we combine structured orchestration with autonomous decision-making. This creates a powerful blend of determinism and flexibility – making AI viable for critical, auditable, and long-running workflows.
We’ll also explore the architectural patterns behind long-running agents – whenever AI needs to interact with the outside world over time to solve complex problems. Topics include monitoring, versioning, and state management – and how BPMN and related tooling can act as a governance layer for AI behavior.
This talk is not about promoting a product. It’s about sharing architectural insights into how AI can be integrated deeply – and safely – into enterprise systems. And it aims to spark a discussion on patterns, pitfalls, and emerging best practices.