Trust

Human control and approval-first autonomy

CaltAI is designed to earn trust over time. Autonomy should start at zero, then expand only where the team wants it.

Autonomy starts at zero

At the beginning, CaltAI should not take sensitive or client-facing action without approval.

It should show the proposed action, the context, the reasoning, and the confidence level before anything happens.

Approval is routed to the right person

Different actions need different reviewers. A founder may need to approve sensitive client messaging. An account lead may approve status updates. An operations lead may approve handoff changes.

CaltAI should respect those roles instead of treating all approvals the same.

You can approve, edit, or decline

The approval layer is not just a yes or no button. The human should be able to edit the draft, change the next action, decline it, or tell CaltAI what to do differently.

Autonomy can increase later

Once the team trusts a specific class of low-risk actions, CaltAI can be allowed to act more independently.

High-stakes actions can stay approval-only for as long as the business wants.

  • Low-risk internal updates can become more autonomous
  • Routine reminders can become more autonomous
  • Client-facing decisions can remain approval-first
  • Sensitive escalations can always require review