Drafting and sending

Two acts, compared#

Read the last row twice, because it is the one people invert. A draft waiting for approval feels like progress and is not: to the customer at the other end, a held reply is no reply, not a slow one. The queue is where that cost is paid, and it is why an approval queue nobody works is worse than a narrower autonomy rule honestly chosen.

Why the gate goes here#

A gate is only worth having where the two sides of it differ. Putting one before reading, researching or drafting would buy nothing — none of those acts can hurt anybody, and a person asked to approve them learns to approve without reading, which is the worst outcome of all. Putting one after sending is not a gate; it is a post-mortem.

1. Everything before the send is recoverable, so nothing before it needs a human decision.
2. The send is the irreversible step, so it is the only place a person's attention changes an outcome.
3. The same boundary serves every route. A send from the Assistant, from a follow-up or from the approvals queue passes the same check.
4. Sending is where evidence is required. A message is *sent* when the provider says so, not when Connect decides it is finished writing.

What a draft is worth on its own#

It is easy to treat drafting as the incomplete half of sending. In practice the draft carries most of the work: reading the thread, finding the customer's history, checking what the business has said before, deciding what the reply should contain. The send is a button.

• A draft you edit teaches nothing by itself. If the change is a lesson rather than a one-off, say so as guidance or as a memory — an edit is a decision about one message.
• A rejection is also about one message. It is recorded, and the engine does not immediately produce the same reply again, but it is not a rule.
• A stale draft is a real risk. If the thread moved while the draft waited, approving it sends an answer to an older message, which reads as though nobody was listening.
• Nothing is in flight while it waits. No provider has been called, so there is no read receipt, no delivery attempt and nothing in a sent folder.

Choosing where your own line sits#

1. Start narrower than feels necessary on any channel where a mistake is public.

   Result You see what Connect writes before anybody else does, and the cost is your reading time rather than a customer's impression.

2. Widen by scope rather than all at once — one channel, or one kind of message.

   Result Autonomy resolves at several scopes with the narrowest winning, so a single demanding account can stay held while everything else moves.

3. Watch the queue's age, not its length.

   Result A queue of forty worked hourly is healthy; a queue of six with items three days old is a customer waiting.

4. Revisit the rule when you find yourself approving without reading.

   Result That is the signal that the gate has stopped doing anything, and either the rule should widen or the approvals should get real attention.

The comparison that goes with this one is autonomy and approval: autonomy is the standing rule, an approval is one decision taken under it.

Questions#

Related#

• The four autonomy modes
• Held drafts
• Needs You
• Proof that a message was sent
• Autonomy and approval
• Human approval without blocking low-risk work