Overview

Autonomous Agents

AI-powered agents that autonomously handle customer success tasks.

Overview

Autonomous agents execute complex, multi-step tasks with human oversight. They can monitor accounts, take actions, and make decisions within defined boundaries.

Agent Dashboard

Access from main navigation:

Click Agents in the sidebar
View all agents
Monitor execution
Configure settings

Dashboard Overview

Metric	Description
Active Agents	Currently running
Executions Today	Tasks completed
Pending Approvals	Waiting for human
Success Rate	% successful runs

Agent Types

Monitoring Agents

Continuously watch accounts:

Agent	Function
Health Monitor	Track score changes
Risk Detector	Identify churn signals
Opportunity Scanner	Find expansion signals
Engagement Tracker	Monitor activity levels

Action Agents

Perform specific tasks:

Agent	Function
Email Agent	Generate and send emails
Task Agent	Create and update tasks
Meeting Agent	Schedule and prep calls
QBR Agent	Generate quarterly reviews

Analysis Agents

Deep data analysis:

Agent	Function
Account Analyzer	Comprehensive reviews
Trend Analyzer	Pattern detection
Benchmark Agent	Compare against peers
Predictive Agent	Forecast outcomes

How Agents Work

Agent Cycle

1. MONITOR → Watch for trigger conditions
2. ANALYZE → Understand the situation
3. PLAN → Determine actions to take
4. EXECUTE → Perform actions
5. REPORT → Log results and ask for approval
6. LEARN → Improve based on outcomes

Human-in-the-Loop

Agents request approval for:

Action Type	Always Human	Conditional
Send Email	To executives	To others
Create Task	With high priority	Standard
Update Account	Sensitive fields	Basic fields
Escalate	Always	-
Dismiss Risk	High risk	Low risk

Approval Queue

Review pending actions:

Click Approvals in sidebar
See all pending items
Review agent’s reasoning
Approve, reject, or modify
Agent continues

Creating Agents

Step 1: Choose Type

Select agent category:

Monitoring
Action
Analysis

Step 2: Define Purpose

Describe what the agent does:

Example:
"Monitor at-risk accounts and automatically create 
follow-up tasks. If health drops more than 20 points 
in a week, send an alert to the CSM and create a 
high-priority call scheduling action."

Step 3: Set Boundaries

Define what agent can do:

Permission	Options
Create Actions	Yes/No, with limits
Send Notifications	Yes/No, to whom
Send Emails	Yes/No, require approval
Modify Accounts	Yes/No, which fields
Escalate	Yes/No, to whom

Step 4: Configure Triggers

When agent activates:

Trigger	Example
Condition	Health < 50
Schedule	Daily at 9 AM
Signal	New risk detected
Manual	User starts

Step 5: Set Parameters

Fine-tune behavior:

Parameter	Setting
Confidence Threshold	Only act if > 80% confident
Max Actions per Day	Limit to prevent spam
Escalation Threshold	When to involve humans
Logging Level	Verbosity of logs

Agent Configuration

AI Settings

Setting	Description
Provider	OpenAI, Anthropic, etc.
Model	GPT-4, Claude, etc.
Temperature	Creativity vs. precision
Max Tokens	Response length

Behavior Settings

Setting	Description
Autonomy Level	Fully autonomous, supervised
Confirmation Required	Before sensitive actions
Error Handling	Retry, escalate, stop
Timeout	Max time per task

Notification Settings

Setting	Options
On Start	Notify when agent runs
On Completion	Report results
On Error	Alert on failures
On Approval Needed	Request human input

Monitoring Agents

Live Activity Feed

Real-time agent actions:

[09:15] Health Monitor started
[09:15] Analyzing 45 at-risk accounts
[09:16] Created action for Acme Corp (health drop)
[09:16] Sent alert to [email protected]
[09:17] Email Agent started
[09:17] Generating check-in email for Globex
[09:18] Email drafted, awaiting approval

Execution History

View past runs:

View	Description
Timeline	Chronological view
By Agent	Per-agent history
By Account	Per-account activity
By Outcome	Successes and failures

Best Practices

Agent Design

Tip	Description
Start narrow	Simple, focused agents
Clear boundaries	Define what they can/cannot do
Human oversight	Approve sensitive actions
Monitor closely	Watch early executions
Iterate	Improve based on results

Autonomy Levels

Level	Agent Can	Human Must
Full	Do everything	Nothing
High	Most actions	Sensitive only
Medium	Standard actions	Major decisions
Low	Analyze only	Take all actions

Error Prevention

Practice	Benefit
Set limits	Prevent runaway actions
Require confirmation	Avoid mistakes
Log everything	Audit trail
Monitor closely	Catch issues early

Troubleshooting

Issue	Solution
Agent not acting	Check trigger conditions
Too many actions	Lower autonomy, add limits
Wrong decisions	Refine instructions
Failing consistently	Review agent design
Approval queue backed up	Prioritize, delegate