Autonomy Modes
Control how much independence your agent has when replying to emails.
Autonomy Modes
Your agent can operate at three different levels of independence. Choose the mode that matches your comfort level and use case.
The Three Modes
1. Human in the Loop (Safest)
How it works:
- Email arrives
- Agent reads and understands it
- Agent drafts a response
- You review the draft
- You approve or reject
- Email sends (or doesn't)
Best for:
- Getting started with a new agent
- Complex or sensitive domains
- High-stakes emails (legal, contracts, complaints)
- Building confidence in the agent
Pros:
- Complete control over every response
- Lowest risk of mistakes
- Great for learning how the agent thinks
- Can provide feedback to improve responses
Cons:
- Slowest mode (you must review each email)
- Requires your active attention
- Not scalable for high-volume emails
- Can become a bottleneck
Example use case: Customer support agent handling complaints and refund requests
2. Semi-Autonomous (Balanced)
How it works:
- Email arrives
- Agent reads and understands it
- Agent drafts a response
- Agent sends after 2-minute delay
- You can cancel within 2 minutes
- Email sends (unless you cancel)
Best for:
- Routine emails with low risk
- Once you trust the agent's judgment
- High-volume support tickets
- Common questions and responses
Pros:
- Faster than Human in the Loop
- Safety net: you can still cancel
- Good balance of speed and control
- Scales better for volume
Cons:
- Requires you to monitor emails
- 2-minute window is tight
- Some emails might send before you see them
- Need to be responsive
Example use case: Support agent answering FAQs, sales agent confirming meeting times
3. Full Auto (Fastest)
How it works:
- Email arrives
- Agent reads and understands it
- Agent sends response immediately
- No review or approval needed
Best for:
- High-volume, low-risk emails
- Routine confirmations and acknowledgments
- Emails you're confident the agent handles well
- Scaling to thousands of emails
Pros:
- Fastest response time
- Fully automated, no human intervention
- Best for scaling
- Improves customer experience (instant responses)
Cons:
- Highest risk of mistakes
- No safety net
- Can't review before sending
- Mistakes go directly to customers
Example use case: Automatic confirmations, receipt acknowledgments, simple FAQ responses
Recommended Progression
Week 1: Human in the Loop
- Create your agent
- Send 20-30 test emails
- Review all responses
- Adjust instructions based on feedback
- Build confidence
Week 2: Semi-Autonomous
- Switch to Semi-Auto mode
- Monitor the 2-minute window
- Let some emails send automatically
- Review what it sends
- Adjust escalation rules if needed
Week 3+: Full Auto (if appropriate)
- Only for routine, low-risk emails
- Keep other agents in Semi-Auto or Human in the Loop
- Monitor performance weekly
- Adjust rules as needed
Real-World Examples
Support Agent: Mixed Modes
Human in the Loop:
- Complaints and negative feedback
- Refund requests
- Technical issues
- Anything escalated
Semi-Autonomous:
- Common questions (pricing, features, billing)
- Password resets
- Account status inquiries
- General information requests
Full Auto:
- Automatic confirmations ("Thanks for reaching out")
- Receipt acknowledgments
- Out-of-office replies
- Routing confirmations
Sales Agent: Mostly Semi-Auto
Human in the Loop:
- Enterprise deals (>$100k)
- Custom pricing requests
- Legal/contract questions
- Competitor mentions
Semi-Autonomous:
- Lead qualification responses
- Product question answers
- Demo scheduling confirmations
- Follow-up emails
Full Auto:
- Initial lead acknowledgments
- Meeting confirmation reminders
- Simple FAQ responses
Switching Modes
You can change autonomy mode anytime:
- Go to agent settings
- Click "Autonomy Mode"
- Select new mode
- Changes take effect immediately
Note: Changing modes only affects new emails. Existing drafts keep their original mode.
Best Practices
Start Conservative
Begin with Human in the Loop. It's easier to move to Full Auto than to recover from mistakes.
Test First
Before switching modes, send 10-20 test emails and review responses.
Monitor Escalations
If escalation rate is high (>30%), stay in Human in the Loop longer.
Use Escalation Rules
Escalation rules are your safety net. Define them clearly so risky emails don't go to Full Auto.
Review Weekly
Even in Full Auto, review a sample of sent emails weekly to ensure quality.
Document Your Rules
Write down which emails go to which mode. Share with your team.
Troubleshooting
Q: Agent is sending bad responses in Semi-Auto
A: Switch back to Human in the Loop. Review responses and update instructions. Test more emails before trying Semi-Auto again.
Q: 2-minute window is too short
A: Semi-Auto isn't right for you. Use Human in the Loop instead. You can always move to Full Auto later.
Q: Too many escalations
A: Your escalation rules might be too broad. Review and tighten them. Or the agent's instructions need clarification.
Q: Want to use Full Auto but worried
A: Use Semi-Auto first. Once you're comfortable, switch to Full Auto. You can always go back.
Next Steps
- Building Agents - Configure your agent
- Escalation Rules - Define when to escalate
- Tools & Integrations - What your agent can do