Skip to main content

​
AI Agents

AI Agents are intelligent assistants that go beyond simple automations. Powered by large language models, they can understand context, reason through problems, use tools to gather information, and take actions autonomously.

​
What Makes Agents Different?

Traditional Automations

  • Follow predefined paths
  • Execute fixed sequences
  • Rule-based decisions
  • Predictable outcomes

AI Agents

  • Reason about tasks
  • Dynamic decision-making
  • Use tools intelligently
  • Adapt to context

​
Agent Capabilities

​
Contextual Understanding

Agents understand natural language instructions and can interpret ambiguous requests based on context: 
"Prepare me for my call with Acme Corp tomorrow"
The agent will:
  • Look up the Acme Corp account
  • Review recent activities and conversations
  • Check health scores and open tasks
  • Summarize key points for the call

​
Tool Usage

Agents can use tools to:
  • Fetch data from Statisfy and integrated systems
  • Take actions like sending emails or creating tasks
  • Search through documents and knowledge bases
  • Calculate metrics and analyze trends

​
Multi-Step Reasoning

Agents break complex tasks into steps: 
Task: "Find accounts at risk of churning and draft outreach emails"

Agent reasoning:
1. Query accounts with declining health scores
2. For each account:
   a. Analyze recent activity patterns
   b. Identify key risks
   c. Draft personalized outreach
3. Compile results

​
Agent Configuration

When creating an agent, you configure:

​
Instructions (System Prompt)

Define the agent’s role, expertise, and behavior: 
You are a customer success analyst specializing in account health.
Your role is to:
- Identify early warning signs of churn
- Provide actionable recommendations
- Communicate findings clearly and concisely

Always base your analysis on data, not assumptions.

​
Model Selection

Choose the AI model that powers your agent:
ModelCharacteristicsBest For
Gemini AutoAutomatically selects based on taskGeneral use
Gemini 3 ProAdvanced reasoning, higher accuracyComplex analysis
Gemini 3 FlashFast responsesSimple tasks
Gemini 2.5 FlashBalanced speed and capabilityMost use cases

​
Reasoning Level

Control how deeply the agent thinks:
LevelDescription
AutoModel decides based on task complexity
LowQuick responses for simple tasks
MediumBalanced thinking for standard tasks
HighDeep reasoning for complex analysis

​
Toolkits

Enable groups of tools the agent can use:
ToolkitTools Included
Account DataGet account details, health history, contacts
Activity DataFetch activities, meetings, emails
Task ManagementCreate tasks, add comments, update status
CommunicationSend emails, Slack messages
CRMSalesforce operations, opportunity updates
Data AnalysisRun Python code for counting, aggregation, and calculations

​
Scope

Define what data the agent can access:
ScopeDescription
GlobalAccess all organization data
AccountLimited to specific account context

​
Creating an Agent

​
Step 1: Define the Agent’s Purpose

Start with a clear goal:
  • What problems will this agent solve?
  • Who will use it?
  • What actions should it take?

​
Step 2: Write Instructions

Craft detailed instructions that cover: 
# Role
You are a [specific role] that helps [target users] with [main task].

# Capabilities
You can:
- [Capability 1]
- [Capability 2]
- [Capability 3]

# Guidelines
- [Guideline 1]
- [Guideline 2]

# Constraints
- Never [limitation 1]
- Always [requirement 1]

​
Step 3: Select Tools

Enable only the tools the agent needs:
  • More tools = more capability but slower responses
  • Fewer tools = faster but more limited

​
Step 4: Configure Model Settings

Choose based on your needs:
  • Complex analysis → Higher reasoning, Pro model
  • Quick answers → Lower reasoning, Flash model

​
Step 5: Test Thoroughly

Test with various scenarios:
  • Typical use cases
  • Edge cases
  • Unclear or ambiguous requests

​
Example Agents

​
Account Health Analyst

Purpose: Analyze account health and provide recommendations Instructions: 
You are an account health analyst. When asked about an account:

1. Review health score history and trends
2. Analyze recent activities and engagement
3. Identify risk factors and warning signs
4. Provide specific, actionable recommendations

Always explain your reasoning and cite specific data points.
Tools: Account Data, Activity Data

​
QBR Preparation Assistant

Purpose: Prepare materials for quarterly business reviews Instructions: 
You prepare comprehensive QBR materials by:

1. Summarizing key metrics and changes
2. Highlighting achievements and successes
3. Identifying challenges and proposed solutions
4. Recommending discussion topics

Format output as a structured agenda with talking points.
Tools: Account Data, Activity Data, Task Management

​
Renewal Risk Monitor

Purpose: Proactively identify and address renewal risks Instructions: 
You monitor accounts for renewal risks by:

1. Tracking usage patterns and engagement trends
2. Identifying accounts with concerning signals
3. Drafting personalized outreach messages
4. Recommending specific retention actions

Prioritize accounts by risk level and ARR.
Tools: Account Data, Activity Data, Communication

​
Using Agents in Automations

Agents can be embedded in automation flows using the Agent V2 component: 
Trigger → Gather Context → Agent Analysis → Take Action
This combines the reliability of automations with the intelligence of agents.

​
Best Practices

​
Instruction Writing

Instead of “You are helpful”, try “You are a customer success analyst with expertise in identifying churn risk signals”
Specify how responses should be structured:
Format your analysis as:
## Summary
[1-2 sentence overview]

## Key Findings
- Finding 1
- Finding 2

## Recommendations
1. Action 1
2. Action 2
Clarify what the agent should NOT do:
Do not:
- Make up data or statistics
- Commit to timelines without verification
- Share sensitive internal information

​
Tool Configuration

Too many tools can confuse the agent and slow responses. Enable only what’s needed for the specific task.
An agent analyzing account health doesn’t need task creation tools. An agent drafting outreach doesn’t need CRM tools.

​
Testing

Use actual account data and realistic questions to validate the agent works as expected.
What happens when:
  • Data is missing?
  • The question is ambiguous?
  • Multiple accounts match?

​
Monitoring and Improvement

​
Track Agent Performance

  • Response quality: Are answers accurate and helpful?
  • Tool usage: Is the agent using tools appropriately?
  • Completion rate: How often does the agent successfully complete tasks?

​
Iterate on Instructions

Based on feedback:
  • Add guidance for common failure modes
  • Clarify ambiguous instructions
  • Add examples for complex tasks

​
Review Tool Usage

If the agent isn’t using tools effectively:
  • Check that needed tools are enabled
  • Add explicit instructions about when to use specific tools
  • Simplify by removing unnecessary tools

​
Next Steps

Creating Agents

Step-by-step guide to creating agents

AI Components in Automations

Use agents within automation flows