# Autonomy

**Autonomy** is the rule that says what Connect may do without asking, set per channel and enforced before anything leaves. There are four channels — email, WhatsApp, SMS and voice — four modes from off to autonomous, and four scopes; the narrowest scope that applies wins. The screen is called What Connect May Do.

- **Status:** Available
- **Audience:** both
- **In the app:** #/autonomy
- **Last verified:** 2026-09-10
- **Canonical:** https://connectbyjbrh.com/docs/glossary/autonomy/

## Modes

| Mode | What happens | What reaches a person |
|---|---|---|
| `off` | Connect is switched off for it entirely | Nothing |
| `draft_only` | It prepares the outbound and stops — deliberately without asking | Nothing to approve; the prepared work sits in the conversation |
| `ask_before_send` | Every outbound waits for a human yes | An [approval](/docs/glossary/approval/), in Needs You |
| `autonomous` | It acts within the gates | Nothing, unless a gate stops it |

`draft_only` and `ask_before_send` are the pair people mix up, and the difference is the whole reason both exist. Ask-before-send puts the work in front of a person and waits. Draft-only writes it and stops, on purpose, so that a workspace can watch what Connect *would* say without anyone having to keep saying no. If replies are appearing nowhere and nothing is queued for approval, draft-only is the first thing to check.

## Scopes, narrowest first

The same hierarchy [memory](/docs/glossary/memory/) uses, resolved the same way: `contact`, then `endpoint`, then `channel`, then `workspace`. An *endpoint* is one mailbox or one phone number.

1. A setting against one **contact** overrides everything else for that person.
2. A setting on one **endpoint** overrides the wider one — so a support mailbox can run autonomously while a director's address never sends unaided.
3. A **channel** setting is the usual place to work.
4. The **workspace** setting is the floor everything else falls back to.

Narrowest-wins is what makes a careful account manageable without slowing the business down. One account where every word matters gets its own rule; everything else keeps moving.

## What autonomy is not

**A [rule](/docs/glossary/rule/)** — Rules constrain how Connect behaves. Autonomy decides whether the behaviour may leave the building. No rule widens autonomy, and autonomy does not exempt anything from the rules.
**An allowance** — Work that autonomy permits can still be stopped by the plan's daily [allowance](/docs/glossary/allowance/). The action is permitted and the quota is spent — two different holds with the same appearance from the conversation screen.
**A suppression list** — [Suppression](/docs/glossary/suppression/) and do-not-contact stop contact with a recipient regardless of the mode set here. Granting autonomy does not clear one.
**A promise about quality** — Autonomy governs permission, not competence. Grounding, rules and the intelligence boundary decide what the message says; autonomy decides only whether it goes.

> **Careful** Changing a mode takes effect on the next piece of work, not retroactively. Actions already held stay held until decided, and actions already taken stay taken — see the [audit](/docs/glossary/audit/) trail for what was done under the previous setting.

## Choosing a starting point

A common progression is to open a channel at `ask_before_send`, work the queue for a while, and move it — or one endpoint on it — to `autonomous` once the approvals stop being interesting. `draft_only` suits a period of observation where nobody wants a queue to work. `off` is for something the business does not use, and is not a safety measure for something it does: an unanswered line that customers still write to is its own failure, and [human approval without blocking low-risk work](/research/approval-without-blocking/) covers the trade.

## Questions

### Why is a reply not going out on a channel set to autonomous?

Check three things in order: a narrower scope overriding it, the daily allowance, and a suppression on the recipient. All three look identical from the conversation, and only the first is an autonomy setting.

### Does autonomy apply to inbound work?

It governs what Connect may do, so reading, classifying and grounding continue regardless. What changes is whether the result may leave without a person.

### Can a channel be autonomous for one customer only?

Yes, at the contact scope, which beats every wider setting. That is the usual way to run a familiar account at full speed while everyone else waits.

## Related

- [Approval](https://connectbyjbrh.com/docs/glossary/approval/)
- [Needs You](https://connectbyjbrh.com/docs/glossary/needs-you/)
- [Rule](https://connectbyjbrh.com/docs/glossary/rule/)
- [What Connect may do](https://connectbyjbrh.com/docs/autonomy/)
- [Human approval without blocking low-risk work](https://connectbyjbrh.com/research/approval-without-blocking/)

## What this page is based on

- docs-source/sources/GENERAL.md §5 — channels, modes and scopes in autonomy.py
- `docs-source/facts.py` — CAPABILITY_STATUS autonomy_modes, approvals
- docs-source/sources/GENERAL.md §4 — the What Connect May Do route