committee-plan

To create an executable project plan despite conflicting perspectives, this approach coordinates four AI subagents to draft parallel plans, surface contested steps, and iterate through structured discussion rounds to converge on a single consensus. It applies a structured Delphi-like method to resolve disagreements and produces a final plan with explicit verification steps.

Multi-model consensus planning works by orchestrating four AI subagents—Opus, Sonnet, Gemini, and GPT—to draft plans in parallel for complex roadmaps. These models compare their outputs, identify agreements versus contested steps, and iterate through structured discussion rounds to converge on a single executable plan with risk flags and provenance.

Yes, you can use multi-model planning for risk-aware sequencing and dependency management. It is specifically designed for complex projects needing multiple perspectives and explicit verification, such as multi-phase roadmaps. The final output includes explicit risk flags, verification steps, and provenance to ensure practical delivery checks.

The best way to draft and compare parallel plans for uncertain project scoping is initiating a planning brief to a four-model committee. This method grounds the planning in codebase patterns and existing constraints, allowing four AI subagents to draft independently, compare outputs, and converge on an executable plan with a fixed output format.

Yes, you need to provide codebase patterns and existing constraints before generating a plan. The multi-model planning process requires grounding in these practical delivery checks and existing project limitations to ensure the four AI subagents can accurately draft, compare, and converge on a realistic executable plan.

You should not use a multi-model committee for simple project planning tasks that lack complex dependencies, multi-phase roadmaps, or significant uncertainty. If your project does not require multiple perspectives, explicit verification steps, or risk-aware sequencing, the overhead of coordinating four AI subagents to reach consensus is unnecessary.