Guide
Optimize Agents
Improve your AI team's performance over time.
Monitor Trust Scores
Trust scores are your primary signal for agent performance. Check them weekly:
- Rising scores → Agent is performing well. Consider increasing their budget or responsibility.
- Falling scores → Agent needs attention. Check Critic feedback to understand why.
- Stuck at medium → The agent may need better tools or clearer instructions in their role definition.
Review Critic Feedback
When the Critic rejects an agent's output, the feedback contains specific reasons. Aggregate this feedback over time to identify patterns — is the agent consistently weak in one area?
Adjust Agent Roles
If an agent consistently underperforms in certain task types, consider:
- Reassigning those tasks to a different, higher-trust agent
- Adding new tools that address the agent's weaknesses
- Replacing the agent with a new one (different model or configuration)
Budget Optimization
Analyze cost per completed task. If an agent's cost per task is high but trust score is low, consider replacing them. If cost per task is low but quality is high, increase their budget.