exploration-orchestrator

Parallel AI exploration decomposes complex questions into independent subproblems by spawning multiple workers concurrently. The orchestrator coordinates phase-driven workflows spanning decomposition, spawning, waiting, judging, iteration, and final synthesis to produce a unified analysis.

Iterative re-exploration with judge feedback is a workflow mechanism where evaluated findings trigger targeted re-spawning of workers to fill critical coverage gaps. This phase-driven coordination repeats until sufficient breadth and depth are achieved before final synthesis.

To orchestrate parallel AI workers, invoke the orchestrator with a complex question and configuration. It handles decomposition, spawns workers with controlled breadth and depth, judges findings, iterates on gaps, and synthesizes results into a unified analysis without executing code.

No, this orchestrator cannot execute code during multi-step AI investigations. It operates under safety constraints as a non-destructive coordination layer, strictly managing decomposition, spawning, judging, iteration, and synthesis without performing code execution.

You need parallel orchestration for multi-model AI exploration tasks when questions are too complex for a single pass and require controlled breadth, iterative re-exploration, and judge feedback to ensure comprehensive coverage before synthesizing a unified analysis.