Safety

What is auto-send?

Auto-send means a workflow delivers a message without someone pressing send that time. It is safe for a template you already approved, filled with fields from the system of record. It is a bad default for fresh prose.

In plain English

The ready-by text that uses the ticket number and the time from the counter system can auto-send. A reply to a complaint should not. An auto-send rule is the written line between those. Draft mode is the other side of the line. Auto-send is a privilege you earn on one narrow message. The text you have sent a hundred times without edits can be a candidate. The first reply to a new lead is not. Turn it on for one template, watch every send for a week, and keep a kill switch that a person can find without calling you. If you cannot explain the worst message it might send, it is not ready. Keep a customer you can text yourself as the canary. If they receive something odd, you hear it before a stranger does.

How an owner would use it

List the templates allowed to auto-send. Each one should have no sentence the model is inventing. Review a sample every week. Pull the permission the moment a bad one gets out.

Common mistakes

Auto-sending anything an LLM just wrote is how a made-up discount leaves the building. A threshold like 'send if the model seems sure' is not a rule you can audit. The model does not owe you a real confidence. If you cannot point at the template and the fields, do not auto-send.

A worked example

A dry cleaner

Ready texts send, stain debates do not

This is an illustration, not a customer result. When the ticket is marked ready, the customer gets the template with the name and the pickup window from the system. A reply about a damaged button is drafted and waits. The template has no blank the model is allowed to fill with an apology offer.

Related terms

Prompts and guides

Questions owners ask

What is auto-send?

Auto-send delivers a message without a fresh human approval. Limit it to approved templates filled from your systems.

How do I decide what may auto-send?

If a person would not need to edit the words, and every fact comes from a field you trust, it can be a candidate. Fresh wording stays in draft mode.

Should I auto-send when the model says it is confident?

No. Confidence language from a model is not a measurement you can bank on. Use the template rule instead.

Product guides are on AI apps for business.