Human-in-the-loop

Human-in-the-loop (HITL) is a principle where an AI agent works on its own but hands the decision to a person at key moments. The agent prepares the result, and a person reviews, edits and approves it. Automation saves time, and people stay accountable for the decision.

Where you usually want a person involved:

• Actions that are hard to undo: sending something to a customer, making a payment, deleting data
• Giving the agent new access
• Replies that affect your company's reputation
• Edge cases where the agent isn't sure about the result

In Pachca, API drafts help with this: an agent with a personal token leaves a message as a draft in the message field of the right chat, and you review and send it yourself. Drafts can't be sent through the API. The thread works as a checkpoint too: the agent reports what it did, and the team sees the result before the next step.