research

Multi-agent research synthesis works by spawning multiple researcher agents in parallel from distinct angles, then combining their outputs with a single synthesizer agent to produce actionable recommendations.

The best way to compare approaches is triggering a 5-phase workflow with explicit angle taxonomy and phase prompts, which surfaces diverse perspectives and resolves disagreements into concrete recommendations.

You get actionable recommendations by asking a clear, multi-faceted question that triggers the workflow, allowing the synthesizer to weigh evidence quality, detect consensus, and resolve disagreements.

Yes, parallel researcher agents run from distinct angles to surface diverse perspectives, while consensus tracking and traceability provide a clear record of where conclusions differ and how disagreements are resolved.

You need stochastic consensus and debate when facing multi-faceted questions like "how should I" or "compare approaches for", ensuring structured, deduplicated outputs across codebase, architecture, and product decisions.

Parallel agent research requires clear, multi-faceted questions to trigger the full 5-phase workflow effectively; overly narrow queries may not leverage the diverse perspectives and consensus tracking needed for complex decisions.