Skip to content

Best Practices & Patterns

Learn from teams who’ve deployed thousands of agents in production. These guides cover everything from writing effective instructions to optimizing performance and ensuring security.



Customer Support Agents

Best practices for building reliable, helpful support automation.

Coming Soon - We’re working on this guide. In the meantime, check out our Customer Support Solution.

Code Review Agents

Automate code review with consistent, thorough analysis.

Coming Soon - We’re working on this guide. In the meantime, check out our Code Automation Solution.

Data Analysis Agents

Build agents that turn data into actionable insights.

Coming Soon - We’re working on this guide. In the meantime, check out our Data Analysis Solution.


Performance Optimization

Make your agents faster and more cost-effective.

Coming Soon - We’re working on this guide. In the meantime, see our API Documentation for optimization tips.

Error Handling & Recovery

Build resilient agents that handle failures gracefully.

Coming Soon - We’re working on this guide. In the meantime, check our Testing Guide.

Monitoring & Observability

Track agent performance and identify issues early.

Coming Soon - We’re working on this guide. In the meantime, see Account Settings for usage tracking.


  • Be specific - Clear instructions get better results
  • Test thoroughly - Try edge cases before deploying
  • Start simple - Add complexity gradually
  • Monitor actively - Review conversations regularly
  • Iterate often - Improve based on real usage
  • Document everything - Keep track of what works
  • Don’t be vague - “Be helpful” isn’t enough guidance
  • Don’t skip testing - Production isn’t the place to debug
  • Don’t over-complicate - Simple agents are more reliable
  • Don’t set and forget - Agents need ongoing refinement
  • Don’t ignore feedback - Users tell you what needs fixing
  • Don’t expose secrets - Never put API keys in instructions

When to use: Support agents that need human backup

1. Agent attempts to answer
2. If confidence < 80%, ask clarifying questions
3. If still uncertain, escalate to human
4. Human reviews context and takes over

When to use: Agents that perform actions (not just chat)

1. Agent proposes action
2. Show user what will happen
3. Get explicit confirmation
4. Execute action
5. Confirm completion

When to use: Agents with multiple information sources

1. Check primary knowledge
2. If no answer, check secondary sources
3. If still no answer, search web (if enabled)
4. If nothing found, admit uncertainty

Guidelines for evaluating agent quality in your environment:

Metric Starting target Notes
Response time Under a few seconds (Fast mode) Deep mode takes longer by design
Accuracy Validate against your knowledge Depends on instructions and KB quality
Escalation rate Tune per use case Support agents should escalate when uncertain
Resolution rate Measure in your pilot Track before scaling to production

Week 1: Foundations

  1. Read Agent Instructions Guide
  2. Complete Interactive Quickstart
  3. Build your first agent

Week 2: Optimization

  1. Read Knowledge Optimization
  2. Read Testing Guide
  3. Refine your agent based on testing

Week 3: Production

  1. Read Security Best Practices
  2. Monitor your agent’s performance
  3. Deploy to production with monitoring

Week 4: Scale

  1. Optimize based on usage patterns
  2. Implement error handling strategies
  3. Scale to handle more traffic

Share what works in Discord or open a docs PR.



Have a best practice to share? We’d love to hear it!