Can AI answer my emails? Draft, autonomous and human-only, explained

Yes, in one of three ways, set per mailbox and overridable per thread. The same three settings are worded differently in setup and in the inbox, which is worth knowing before you go looking.

Updated

Yes — and the useful question is not whether but how much. There are three settings, and the difference between them is simply who presses send.

The three settings

What it doesCalled this during setupCalled this in the inbox
AI writes and sends, no human stepAutonomous — send AI replies immediatelySend AI replies directly
AI writes, a human approvesDraft — I’ll review AI replies before they sendSave drafts for me to review
No AI at allHuman only — no AI repliesI’ll handle replies myself

Same three settings, two vocabularies. The wizard names the mode; the inbox names the behaviour. Worth knowing before you go hunting the inbox for a button labelled “Autonomous”, because there isn’t one.

The inbox also shows the current state in shorthand — Drafts, Auto-reply, or You · No AI.

Where each one is set

The mailbox default is chosen while you connect, under Default reply mode. On the Microsoft paths that field is on the setup screen itself; on the Gmail personal path it is on Finish your mailbox setup. Either way it is the same setting.

A single thread can differ. Open the conversation and the reply-mode control offers the AI branch — under “When mail arrives” — or the human one:

I’ll handle replies myself

No AI runs on this thread.

Underneath, it tells you what it is departing from: “Channel default: Support · Drafts”.

An email thread whose reply-mode control reads Marlow Support · Drafts, with the AI's reply below it waiting for review.
The current mode sits in the thread header, in shorthand. Drafts means the reply underneath waits for a person to approve it.

Which one wins

Three levels, and the most specific wins:

  1. A setting on this thread
  2. The channel default for the mailbox
  3. Failing both, autonomous

That last one deserves a sentence. If the stored mode is ever missing or unrecognised, the fallback is autonomous — the AI sends. So setting the default deliberately is better than assuming an unset field means “do nothing”.

Changing one thread without changing the rest

The control carries a checkbox:

Apply to new conversations on this channel too

Uncheck to change just this thread — new emails from other senders will keep the existing channel default.

Leave it unticked to fix one awkward thread. Tick it when you have decided the mailbox as a whole should behave differently.

Pausing is not a fourth mode

Separate control, different job. Pause AI on this thread — “Stop AI from auto-replying on this thread” — freezes the AI where it is without changing what the mode says:

AI is paused on this thread. Inbound replies will not trigger an automatic AI response until someone hands the thread back to the AI.

The picker then reads “Paused (resume to Support · Drafts)”, so you can see what it will go back to. Use pause when a human has taken over a live conversation; use Human only when the AI should never have been on it.

What choosing a mode does not change

Permission to send already exists. Both providers grant it at connect time. The mode is what decides whether that permission is ever used — not a narrower grant.

An instruction cannot stop a draft being written. Telling the agent “don’t answer marketing email” changes how a reply is worded, not whether one exists. The filters that actually stop a reply are in the code and run before the model: see what does and doesn’t get answered.

If Draft is the one you want, that flow has its own page.