Why are DAG-focused tools poor for probabilistic agent work?

DAG-focused workflow tools are poorly suited for probabilistic agentic work. They operate with rigid, fixed structural constraints, requiring fully predefined, static execution paths that cannot accommodate the often unpredictable branching, partial failures, and long-lived state that spans extended periods needing durable resumable execution common in agent runs across typical use cases. This does not negate their significant practical value for clearly and strictly defined fixed-branch workflow scenarios.