Durable workflow infrastructure for enterprise agentic AI
Run long-running, computationally expensive agentic-AI workflows with important state stored in a database, allowing them to resume from completed steps and logically undo committed updates when later steps fail.
From Imagination in Action — Do Databases Have a Future? | Conversations in Action at 01:44
Problem: LLMs struggle with enterprise data because it is private, structurally heterogeneous, affected by schema rot and access controls, and often requires complex joins. Agentic workflows also need recovery and compensation when a later action fails after earlier transactions have committed.
For: Enterprises and organizations building agentic-AI workflows that combine LLMs with proprietary enterprise data and actions.
Products from this video
Examples
- Department of Transportation of the city of Munich: a citizen-complaint workflow had to determine whether a pedestrian had enough time to cross a specific intersection; traditional agentic AI failed completely.