We didn’t talk for three days because of one button

It started with a simple question: who is in charge of a conversation with a customer, the agent or a person?

One of us believed the agent should take the customer all the way. Otherwise why build it at all: if a person still sits and watches the chat, we are selling an expensive hint. The other said no business owner will ever hand their customers to a machine completely. Customers cost money, it took a long time to find them, and one strange sentence from a bot can cost an order.

The most annoying part was that both were right. We argued one evening, then another, and then simply stopped talking about it and pretended everything was fine. For three days our shared chat had nothing but links to commits.

The answer turned out to be small. The agent replies on its own until someone on the team decides to step in. There is a stop button in the chat. A manager presses it, the agent goes quiet, and the chat shows who stopped it and when. No ten screens of rules about when to hand over to a human. A person can take the conversation at any moment.

Once we built it, it became clear why the argument took so long. We were not arguing about the machine, we were arguing about trust. You can trust an agent with a customer when you know you can stop it at any second. The button is not there to be pressed every hour. It is there so you can switch the agent on and go to sleep.

Three days of silence for one button. But now we can explain in a minute how the work is split between a person and a machine: the machine answers, the person decides. We also agreed never to stay silent for three days again. One day at most.