Overview
Agents are AI-powered workflow steps that can make intelligent decisions and take actions based on your instructions.How it works
- Define custom instructions for the agent
- Grant the agent access to tools it can use
- The agent uses those tools to accomplish your goals
- Data flows from previous steps through the agent to subsequent steps
Agent Capabilities
Custom Instructions
Provide agents with specific instructions on how to behave and what to prioritize. Examples:- “Be concise and professional in all responses”
- “Always verify data before taking action”
- “Prioritize security and accuracy over speed”
Tool Access
Give agents access to:- External Integrations
- Web Search
- Knowledge Bases
Any action from your connected applications (with your permission).Examples:
- Send Slack messages
- Create Google Calendar events
- Write to Google Sheets
- Post to social media
- Send emails
When to Use Agents
Conversational Chatbot Workflows
Build interactive workflows that have multi-turn conversations with users. Example: Create a customer service chatbot that can:- Answer common questions
- Look up customer information
- Create support tickets
- Provide personalized recommendations
Uncertain Logic
Use agents when it’s unclear what set of actions should be taken each time the workflow runs. Example: An email processing agent that:- Reads incoming emails
- Decides on priority level
- Routes to appropriate team
- Takes different actions based on content
Complex Decision Making
Agents can evaluate complex scenarios and make nuanced decisions. Example: A data validation agent that:- Checks data quality
- Identifies inconsistencies
- Researches corrections using web search
- Updates records accordingly
Best Practices
- Be specific with instructions - Detailed instructions lead to better results
- Start simple - Begin with basic instructions and add complexity as needed
- Test thoroughly - Run test workflows to verify agent behavior
- Monitor outputs - Check the run logs to see what decisions the agent made
- Refine over time - Adjust instructions based on real-world performance
- Limit tool access - Only grant necessary permissions for security and clarity
Tips for Better Agent Performance
- Use clear language - Avoid ambiguous terms or instructions
- Provide context - Tell the agent what information matters and why
- Set boundaries - Specify what the agent should NOT do
- Include examples - Provide examples of desired behavior when possible
- Leverage knowledge bases - Upload relevant documentation for the agent to reference