Human Confirmation
Craftly Robot is designed to help automate coordination without removing people from important decisions.
This is especially important when an AI interaction can lead to a consequential real world action.
Examples include:
- Hiring or engaging a person
- Spending money
- Booking a service
- Sharing sensitive information
- Accepting an important agreement
- Making a commitment on behalf of a user
For these situations, the system should distinguish between coordination and authorization.
What Robot Can Do
Robot can assist with activities such as:
Understanding the request → Discovering relevant agents → Asking questions → Comparing information → Communicating → Negotiating possible conditions
These activities reduce the amount of manual coordination required from the user.
What Requires Human Control
When the next step would create a meaningful real-world commitment, the user remains responsible for giving the appropriate confirmation.
The workflow can therefore become:
AI Coordination ↓ Potential Outcome ↓ Human Review ↓ Explicit Confirmation ↓ Consequential Action
This separation is important because an agent should not interpret a user's request as unlimited permission to take every possible action.
The Model Context Protocol's security guidance similarly recommends keeping a human in the loop with the ability to deny tool invocations, particularly when AI systems can call external tools or perform actions.
Why Human Confirmation Matters
Human confirmation provides an important boundary between:
“Robot found a possible solution.”
and:
“The user approved this solution and wants it to happen.”
This helps preserve:
- User control
- Accountability
- Privacy
- Clear authorization
- Reviewability
- Trust
Craftly's Principle
Craftly's goal is not to make agents blindly autonomous.
The goal is to make them usefully autonomous within clearly defined boundaries.
Let AI handle the coordination. Keep humans in control of consequential decisions.