English | 正體中文 (zh-TW) | 简体中文 (zh-CN)
The Delphi Agent is an AI agent skill that uses the Delphi Technique to solve complex problems through structured, multi-round expert consensus.
Many real-world decisions involve genuine trade-offs with no single right answer. The Delphi Agent brings the power of structured expert debate into a single chat window — no multi-agent platform deployment required. The agent simulates a Coordinator and 2–5 virtual experts with opposing value stances, guiding them through blind positioning, cross-examination, red line declaration, and forced convergence, then delivers a unified Final Report with actionable recommendations.
The Delphi Agent follows a 4-phase structured process:
| Phase | Name | What Happens |
|---|---|---|
| Phase 1 | Initialization | The Coordinator deconstructs the problem, proposes expert paradigms and configuration, and waits for your confirmation. |
| Phase 2 | Debate Rounds | Experts argue across multiple rounds — challenging each other while gradually locking in sub-issue agreements. New variables can be introduced until the Red Line round. |
| Phase 3 | Forced Convergence | The Core Dilemma must be resolved. Experts either compromise or file a Formal Dissent. |
| Phase 4 | Final Report | A stand-alone report with Executive Summary, Convergence Path, Action Plan, Trade-off Map, and Residual Risks. |
The Delphi Agent is designed for situations where there is no obvious "right answer" and cross-disciplinary thinking is required. Each example below is a ready-to-use sample question — copy and paste it to start a session.
When facing a major pivot, architecture choice, or business strategy where different values or priorities conflict.
"We are deciding between a microservices and monolith architecture for our new MVP. Set up a Delphi panel representing a 'Delivery Speed Maximizer', a 'Systems Reliability Purist', and a 'Technical Debt Minimizer'."
When a challenge requires synthesizing conflicting constraints that rarely align.
"Design a data privacy policy for our new AI tool. Let's have a 'Regulatory Compliance Defender', an 'Unconstrained Data Innovator', and a 'User Sovereignty Advocate' debate the tradeoffs."
When your team is leaning toward an idea, but you want to rigorously stress-test it against fundamentally opposing paradigms to uncover hidden blind spots.
"My team wants to migrate our entire database to MongoDB. Play the role of three radically opposed data architecture paradigms to find the flaws in this plan."
When trying to predict future trends or assess risks in a highly uncertain environment where optimizations conflict.
"How will quantum computing impact current blockchain cryptography in the next 10 years? Bring together a 'Decentralization Idealist', a 'Post-Quantum Realist', and a 'State-Actor Threat Modeler'."
When a decision involves fundamental value conflicts with no objectively correct answer, and you need to surface the trade-offs transparently.
"An autonomous car's brakes fail. It must choose: (1) Save the passenger by staying its course, or (2) Save the pedestrian by swerving into a wall. Whom should the AI be programmed to save?"
You can use the Delphi Agent in two ways:
- Install as a Skill: Reference the
SKILL.mdfile to install this as an agent skill. - Use as a Gemini Gem: Refer to
gemini-gem.mdfor the properly formatted Gemini Gem configuration.
The Delphi Agent responds to a wide range of cues:
- Explicit keywords: "Delphi", "expert consensus", "structured debate", "devil's advocate", "pros and cons"
- Decision-making patterns: weighing trade-offs, comparing alternatives ("A vs B", "should I X or Y"), strategic or architectural choices
- Indecision signals: "I'm stuck", "I can't decide", "help me think through this"
Simply include any of the above in your prompt, and the agent will automatically transition into the Coordinator role and begin the setup process.