Selecting Agents
Agents are specialized AI personas that bring distinct perspectives to your debate. Learn how to choose the right mix for your session.
What are Agents?
Agents (formerly called Knights) are AI personas with:
- Role: A specific expertise area (e.g., "Product Strategist", "Legal Counsel", "Financial Analyst")
- Prompt: Instructions that define their perspective and approach
- Preferred Model: The AI model best suited for their role
- Domain Expertise: Areas of knowledge they specialize in
Types of Agents
Official Agents
Curated by Crucible Labs, these Agents are carefully calibrated for common use cases. They're pre-configured with optimal prompts and model assignments.
Best for: Standard business decisions, general strategy questions, getting started quickly.
Marketplace Agents
Community-created Agents published on the Marketplace. These include resume/CV-sourced Agents with real-world expertise profiles.
Best for: Specialized domains, industry-specific questions, unique perspectives.
Private Agents
Agents you create yourself for your organization's specific needs. These remain private to your workspace.
Best for: Company-specific knowledge, proprietary processes, internal expertise.
How to Select Agents
1. Review Suggestions
When you create a session, Crucible automatically suggests Agents based on your question. Review these suggestions and consider:
- Do they cover the key perspectives needed?
- Are there any gaps in expertise?
- Is there appropriate diversity of viewpoints?
2. Add Marketplace Agents
Browse the Marketplace to find Agents with specialized expertise relevant to your question. Look for:
- Domain expertise matching your question
- Seniority level appropriate for your decision
- Proven track record (if available)
3. Balance Perspectives
Aim for a balanced mix that includes:
- Different functional areas (product, finance, legal, marketing, etc.)
- Varying risk tolerances (conservative, moderate, aggressive)
- Different seniority levels (strategic vs. tactical perspectives)
- Competing viewpoints on the question
4. Optimal Team Size
While you can seat up to 12 Agents, optimal team sizes are:
- 6-8 Agents: Best balance of perspectives and debate quality
- 4-5 Agents: Faster sessions, good for simpler questions
- 9-12 Agents: Comprehensive coverage, longer sessions
Agent Configuration
For each Agent, you can optionally configure:
- Model Override: Change the default model assignment
- Role Weighting: Adjust how much influence an Agent has in convergence
- Custom Instructions: Add session-specific context or constraints
Best Practices
Include a Devil's Advocate
Always include at least one Agent that will challenge assumptions and identify risks.
Match Expertise to Question
If your question involves legal issues, include a legal Agent. If it's about product strategy, include product expertise.
Avoid Redundancy
Don't seat multiple Agents with identical expertise unless you want to see how different models approach the same role.
Consider Your Audience
If the Decision Brief will be reviewed by executives, ensure Agents can speak to strategic concerns, not just tactical details.