Nobody talks about this, but object-centric context might be the real reason AI agents in process mining are becoming possible.
Traditional case-centric process mining forces every event into a single case ID. Clean for one process, but real operations don’t work that way.
An order touches sales, logistics, finance, procurement, deliveries, invoices and intercompany flows — each with its own objects and relationships.
That’s exactly where AI agents need richer context. An agent can’t reliably decide the next best action if the underlying process view has already flattened multiple objects into an artificial case.
This is essentially why Celonis built the Context Model — a digital twin that preserves objects, documents and their relationships, enriched with business knowledge and constraints, so AI can reason with the right context rather than guess.
My bet: object-centric data foundations and agentic AI adoption will move together, not independently.
Reliable agents need more than an LLM. They need a process model that actually reflects how the business works.
Anyone else building AI agents on top of object-centric data models? Curious what you’re seeing in practice.
