Use two controls for two different problems.
A business-wide pause and a single-conversation takeover are not the same action. Pause is for situations that may affect many customers: an incorrect promotion, a service outage, a policy change or a message that should not continue. Takeover is for one conversation that needs human judgment while other approved work continues.
Both controls should be available to a named operator in the tool they already monitor. A stop control hidden in a technical dashboard is not useful when the person responsible is away from a computer.
What Kai Can Do supports today.
Kai Can Do currently gives the owner two controls through Telegram: pause automated replies and take over an individual customer conversation. These controls support live customer service work on WhatsApp and email without requiring the owner to manage a technical interface.
Kai can reply in the customer's language. If a customer asks whether Kai is human, Kai says plainly that it is AI. Clear identity matters during normal service and becomes especially important when a human takes over the conversation.
Define what “paused” means.
Before launch, write down exactly which automated replies the pause covers. Decide whether the operator will continue answering manually, how the team will recognize a paused state and who may restart the agent. Do not assume that a label such as “emergency stop” has the same meaning across every channel or workflow.
The same precision is needed for takeover. The team should know who owns the customer reply after takeover, how colleagues can see that a human is handling it and what must happen before automation resumes.
Assign authority before an incident.
Name a primary operator and a backup, then decide who may pause, restart and return an individual conversation. Keep a short written playbook outside the agent itself so the team can still follow it when something is unavailable. It should include the control channel, the people to notify and the manual route for urgent customer replies. Review access whenever a team member joins, leaves or changes role.
Use a handback checklist.
Restarting should be a deliberate decision, not the end of a timer. The operator should confirm that the original problem is understood, affected instructions have been corrected, any open customer promise has a human owner and a test conversation behaves as expected. If only one conversation was taken over, confirm that its status is clear before returning it.
Test control before testing cleverness.
A launch rehearsal should answer four questions: Can the owner pause replies? Can the owner take over one thread? Can the team tell who is currently responsible? Can the owner return control without confusing the customer? Repeat the exercise whenever channels, operators or customer rules change.
See the current owner controls and channels, review Kai's operating profile, or check which roles are available now. The related guides on customer escalation rules and teaching an agent your tone cover the decisions that should exist alongside a stop button.