The fallacy of the graph: Why your next agentic workflow should be code, not a diagram

This post argues against using graph-based frameworks for complex, procedural logic, especially for agentic applications. It contrasts the limitations of static graphs in handling dynamic control flow, data management, and error handling/compensations with the benefits of writing normal, procedural code orchestrated by Durable Execution. The author uses examples to show how code-based approaches with try/except/finally and idempotent activities provide cleaner, more robust, and easier-to-maintain solutions for dynamic workflows, including agentic systems and saga patterns, compared to graph-based representations. It reinforces the core principle of Durable Execution as a superior abstraction for such logic.