Turn your AI agent into a brainstorm coordinator.
Protocol, examples, and templates for structured multi-model brainstorming — get better decisions by leveraging model diversity instead of relying on a single perspective.
Every AI model has blind spots. Claude is great at nuanced reasoning but can miss implementation risks. Codex excels at concrete examples but may overlook strategic implications. Gemini brings unconventional framings that neither surfaces.
When you ask one model, you get one perspective. When you ask three and synthesize, you get something none of them would have produced alone.
We've run dozens of structured brainstorms across a range of decision topics using this protocol. In every case, the multi-model synthesis surfaced insights that any single model missed:
- A naming brainstorm where one model caught a package registry conflict that would have blocked launch
- A competitive analysis where different models found different competitors — the union was the complete picture
- A product strategy brainstorm where one model's framing ("hide the protocol, sell the workflow") became the core positioning
The protocol has three steps:
Send the same question to 2-3 AI agents with different underlying models.
Work hands-on with each agent between rounds. Steer one toward depth, push another toward concrete examples, ask a third to focus on competitive data. This isn't fire-and-forget — it's running simultaneous conversations with advisors who each have different expertise.
Take the best from each output, resolve contradictions, and produce a synthesis that captures the full picture.
The interactive direction step is what makes this different from side-by-side model comparison tools. You're not just parallelizing — you're collaborating with each model based on its strengths.
If you run multiple AI agents (Claude Code, Codex, Gemini, etc.), OACP handles the dispatch and collection. First brainstorm: 5 min setup. Every one after: 30 seconds.
1. Pick a coordinator. Choose one of your agents (Claude Code or Codex) to run the brainstorm. It will dispatch to the others and synthesize results.
2. Setup (one-time) — paste into your coordinator:
Install oacp-cli (https://github.com/kiloloop/oacp) if not already installed.
Create a new OACP project called "brainstorm" with claude, codex, and gemini
as agents. Run oacp doctor to verify everything works.
3. Dispatch — tell your coordinator what to brainstorm:
I want to brainstorm about [YOUR TOPIC HERE].
Use the brainstorm template below. Replace the topic, then dispatch to all
other agents using oacp send with --type brainstorm_request.
---
We need to decide: [YOUR TOPIC]
Context: [relevant background, constraints, goals]
Research this topic online first — search for recent discussions, benchmarks,
case studies, and expert opinions. Use real data, not just training knowledge.
Then provide:
1. Your recommended approach with rationale
2. Top 3 risks or tradeoffs
3. One contrarian take we should consider
Cite your sources.
---
4. Triage — tell your other agents to respond:
Go to each non-coordinator agent session and paste:
Check your OACP inbox for the brainstorm project. Read the request and
respond following the OACP inbox protocol
(https://github.com/kiloloop/oacp/blob/main/docs/protocol/inbox_outbox.md).
If your agents already run /loop 2m /check-inbox, skip this step — they'll pick up the request automatically.
5. Collect & synthesize:
Back in your coordinator:
Check the brainstorm project inboxes for responses. Read all responses, then
write a synthesis: where agents agree, where they disagree, unique insights
from each, and a recommended decision.
- Read the protocol doc — full step-by-step guide
- Browse the examples — real brainstorm results with commentary
- Pick a template for your decision type
- Run your first brainstorm with whatever AI tools you already have
For larger research projects, see the Multi-Agent Research Runbook — a full playbook for angle-splitting, parallel dispatch, synthesis, and signoff.
Coming soon — real brainstorm case studies designed for public sharing.
Browse the templates to get started with your own brainstorm.
Good fit:
- Architecture decisions with genuine tradeoffs
- Naming (models have different aesthetic sensibilities)
- Competitive analysis (different training data = different knowledge)
- Strategy and positioning (diverse framings surface options)
- Pricing (different reasoning about value and anchoring)
Not needed:
- Questions with a known correct answer
- Simple code generation
- Tasks where speed matters more than thoroughness
Brainstorm is part of the Kiloloop open-source ecosystem for multi-agent coordination.
- OACP — Open Agent Coordination Protocol
- oacp-skills — Skill library for OACP
- agent-estimate — AI agent effort estimation
Apache-2.0