Adaptive orchestration refers to orchestrators trained (often via RL) to choose policies dynamically based on the evolving task state.
Why It Matters
Hard-coded orchestration logic can be brittle. Learned policies can adapt tool usage, delegation, and stopping behavior under changing conditions.