Skip to Content
SettingsAutonomy & AI Controls

Autonomy & AI Controls

The Autonomy & Action Policy page gives you complete control over what your AI agents can do independently versus what requires human approval. This is your primary governance tool for AI behavior.

Accessing Autonomy Controls

Go to Settings > Autonomy & Action Policy in the sidebar.

Emergency Kill Switch

At the top of the page, the Emergency Stop button immediately halts all AI outreach:

  • Activate — All AI activity pauses instantly. Queued jobs are cancelled.
  • Deactivate — AI resumes normal operation.

When the kill switch is active, a warning banner appears across the page. The timestamp and user who last changed the kill switch are recorded.

Email Sending Mode

Controls whether AI-drafted emails can send automatically or must go through approval:

ModeBehavior
WarmupAll AI-drafted emails go to the Approval Queue regardless of risk score. Recommended when starting out or warming up a new sending domain.
NormalThe Global Autonomy Mode and per-action settings control sending behavior. High-risk emails may still require approval based on your thresholds.

Global Autonomy Mode

Three modes control the overall autonomy level:

ModeDescriptionPer-Action Settings
AssistedAI drafts every message, but a human must review and approve before anything sends. Maximum control.Overridden — everything requires approval
ControlledPer-action settings apply. Configure each action independently as Autonomous, Approval, or Disabled.Active
Full AutonomyAI handles everything without asking. Best for high volume once you trust output quality.Overridden — everything runs autonomously

Per-Action Settings (Controlled Mode)

When in Controlled mode, you can set each action independently:

ActionRisk LevelOptions
Send replyLowAutonomous / Approval / Disabled
Send email (new outreach)MediumAutonomous / Approval / Disabled
Send SMSMediumAutonomous / Approval / Disabled
Send WhatsAppMediumAutonomous / Approval / Disabled
Book meetingMediumAutonomous / Approval / Disabled
Send calendar inviteMediumAutonomous / Approval / Disabled
Create opportunityLowAutonomous / Approval / Disabled
Advance opportunity stageHighAutonomous / Approval / Disabled
Enroll in sequenceLowAutonomous / Approval / Disabled
Mark do-not-contactHighAutonomous / Approval / Disabled
Transfer to agentLowAutonomous / Approval / Disabled
Log CRM noteLowAutonomous / Approval / Disabled

Risk levels are informational — they indicate how impactful the action is. High-risk actions (like advancing deals or marking do-not-contact) default to Approval mode.

Global Thresholds

Two additional thresholds available in Controlled mode:

Deal Value Approval Threshold

Require human approval when a deal exceeds a dollar amount. For example, setting this to $10,000 means any AI action on a deal worth more than $10K goes to the approval queue.

Max Actions Per Signal

Limit how many actions the AI can create from a single inbound signal (1-20). This prevents a single email from triggering an excessive number of automated responses.

Governance Limits

Daily Email Cap

Set the maximum number of emails the AI can send per day (1-1000). Once the cap is reached, additional emails are held until the next day.

Working Hours

Set the time window during which AI agents can send messages:

  • Default: 9:00 AM to 5:00 PM
  • Outside these hours, messages are queued and sent when the window opens

See Working Hours for more details.

AI Reply Triggers

Configure which types of inbound messages the AI automatically responds to:

IntentDescription
Expressed interestContact shows interest in your product
Meeting requestsContact wants to book a call or demo
QuestionsContact asks about pricing, features, or details
ObjectionsContact raises a concern or pushback
Other / unclassifiedReplies that do not match a clear category

Uncheck an intent to prevent the AI from auto-responding to that type of message.

Meeting Reminders

Toggle whether ScendCore sends automated reminder emails 24 hours before meetings. Disable this if you use Calendly, Cal.com, or Google Calendar reminders to avoid duplicates.

Confidence Scoring

Behind the scenes, ScendCore computes a confidence score (0-100) based on:

  • Draft quality (average rating)
  • Draft accuracy (how often humans edit AI drafts)
  • Win rate (revenue performance)
  • Deliverability (bounce rate)
  • Reputation (complaint rate)

This score influences the dynamic risk threshold that determines which actions auto-send versus require approval.

Troubleshooting

All emails going to approval queue

  • Check if Email Sending Mode is set to Warmup.
  • Check if Global Autonomy Mode is set to Assisted.
  • In Controlled mode, verify that “Send email” is set to Autonomous.

AI not responding to inbound messages

  • Verify the relevant AI Reply Trigger intents are checked.
  • Make sure the kill switch is not active.
  • Check that the channel has an Intake AI assigned.
Last updated on