​
Quick Overview

Agent steps let your AI make autonomous decisions about how to complete complex tasks. Instead of following fixed workflows, they work from guidelines you provide and choose which skills to use based on the situation. Think of them as hiring a smart assistant who can adapt to unexpected scenarios. They’ll keep working until they meet your exit conditions, using whatever combination of skills they need to get the job done. Agent steps are powerful but more expensive than standard actions. Use them only when the next steps are genuinely uncertain and you need intelligent decision-making.
AI Agents are powerful but more expensive and potentially less reliable than standard actions. Use them only when the next steps are truly uncertain. When possible, use actions and conditions instead.

​
Understanding AI Agent Components

AI Agents consist of four essential parts that work together:
  • Prompt: The core instructions that guide your agent’s behavior. Think of this as the agent’s mission statement - what it should accomplish and how.
  • Model: The AI model (e.g., GPT-4, Claude) that powers the agent’s decision-making capabilities.
  • Skills: The actions and capabilities available to the agent. These are the tools it can use to accomplish its goals.
  • Exit Conditions: Criteria that tell the agent when its task is complete and it should proceed to the next workflow step.

​
Adding an AI Agent

IMAGE NEEDED: Screenshot of clicking 'Enter AI agent' option when adding a step
Click the plus button in your workflow and select “Enter AI agent” to add the AI Agent step.

​
Configuring Your AI Agent

​
Setting the Agent Prompt

IMAGE NEEDED: Screenshot of AI Agent configuration panel showing the prompt field
Write clear, detailed instructions that define your agent’s role, approach, and expectations. A good prompt should establish the agent’s identity, explain available tools, and set quality standards. Example research agent prompt: 
You are a professional research assistant specializing in company intelligence gathering. Your role is to thoroughly research companies and compile accurate, up-to-date information.

RESEARCH TARGETS:
For each company, find and verify:
- Official company name and any aliases
- Founding date and brief history
- Current employee headcount (approximate range is acceptable)
- Primary business focus and industry
- LinkedIn company profile URL
- Website URL

RESEARCH APPROACH:
1. Start with web searches using company name and variations
2. Cross-reference information across multiple sources
3. Prioritize official sources (company websites, LinkedIn, Crunchbase)
4. When exact data isn't available, note "approximately" or "estimated"
5. If you can't find specific information, clearly state "not found"

QUALITY STANDARDS:
- Verify information across 2+ sources when possible
- Include confidence level for uncertain data
- Organize findings in a clear, structured format
- Note your research methodology for transparency

EXIT CONDITIONS:
Stop your research when you meet any of these criteria:
1. You have successfully found and verified all required company information
2. You have conducted thorough research using multiple strategies and documented what you found, clearly noting any unavailable data
3. You have performed at least 8 research actions and gathered substantial information
Key elements of effective agent prompts:
  • Role definition: Who is the agent and what’s their expertise?
  • Clear objectives: What specific outcomes do you need?
  • Process guidance: How should they approach the task?
  • Quality standards: What level of accuracy and detail is expected?
  • Exit conditions: When is the task complete?

​
Adding Skills

Skills are the tools and actions your agent can use to accomplish its goals. Think of them as giving your agent specific capabilities — like the ability to search the web, send emails, or analyze data.
How skills work:
  • Your agent can use skills multiple times and in different combinations
  • Skills can be used in sequence (search, then crawl results) or iteratively (search → analyze → search more if needed)
  • The agent will explain its reasoning for using specific skills
For our research agent example, add these complementary skills:
  • Web Search: Finds relevant web pages, articles, and resources across the internet. Your agent will craft search queries, evaluate results, and follow promising leads.
  • Website Content Crawler: Extracts and analyzes content from web pages. Once the agent finds relevant sites through search, it can crawl them to gather detailed information about companies, verify facts, and extract structured data.
Choose skills that work well together. For research tasks, pairing search capabilities with content extraction lets your agent both find and thoroughly analyze information sources.
Skill selection guidelines:
  • Start with 2-4 essential skills rather than overloading with options
  • Consider how skills complement each other in your workflow

​
Defining Exit Conditions

Exit conditions tell your agent when its work is complete and it should move to the next step in your workflow. Think of them as success criteria that define “done” for your agent’s task.
How exit conditions work:
  • Your agent will keep working until at least one exit condition is satisfied
  • You can have multiple exit conditions — the agent stops when ANY of them is met
  • Exit conditions should be specific and measurable
  • The agent evaluates conditions after each action it takes
Example exit conditions for our research agent:
  1. Primary success condition: 
    You have successfully found and verified all required company information: company name, founding date, headcount, business focus, LinkedIn URL, and website URL.
  2. Fallback condition: 
    You have conducted thorough research using multiple search strategies and documented what information you were able to find, clearly noting any data that remains unavailable after extensive searching.
  3. Time/effort limit: 
    You have performed at least 8 research actions (searches and crawls) and gathered substantial information even if some details remain incomplete.
Best practices for exit conditions:
  • Include a fallback option for when perfect results aren’t achievable
  • Be specific about what constitutes “completion”
  • Consider both ideal outcomes and acceptable partial results
  • Use measurable criteria when possible (e.g., “found at least 80% of required data”)

​
Best Practices

If costs are high or results inconsistent, consider using conditions and standard actions instead of AI Agents.

​
Next Steps

Master these related concepts to build more sophisticated automations:

Conditions

Use conditions for predictable branching logic instead of AI Agents

Human in the Loop

Add confirmation steps for critical AI Agent decisions

Knowledge Base

Combine AI Agents with knowledge bases for informed decisions

Testing Workflows

Learn advanced testing strategies for AI Agent workflows