Work · Community report
Process and reply to an email inbox
Process and reply to an email inbox. Have it go through an email inbox and reply where it can, with reply-all and other irreversible sends held back for your approval.
What you can get
Reaches your accounts Creates or changes things in accounts you connect, with your approval.
-
It does
- Reads and triages inbox messages
- Drafts replies for messages you've said it can handle
- Does not send a reply-all, or anything you've flagged as needing review, without your approval
-
It stops and asks
- Before it starts: set explicit rules for what it can reply to on its own.
- Before any reply-all: your approval, no exceptions.
- After a session: review what was actually sent.
-
You approve
- Your approval before any reply-all is sent
- Your approval before sending anything flagged as uncertain
Derived from this task's own brief and steps, not from the source.
- Set the rulesStops for your approval
Which message types can be auto-handled; reply-all always needs approval.
- Process the inbox
Routine messages handled; uncertain ones flagged.
- Review before sendingStops for your approval
Recipients and wording checked, especially for multi-recipient replies.
- Check the summaryStops for your approval
What was sent, to whom, and anything needing a follow-up.
Editorial, not from the source
Works when
- A chunk of your inbox is routine and low-stakes (confirmations, simple questions).
- You're willing to set rules up front for what it can send versus hold for review.
- You can check in afterward on what actually went out.
Does not work when
- Many messages in the inbox are sensitive, legal or easy to misjudge the right recipients for.
- You want it running fully unattended with no review of sent replies.
- A reply-all mistake in this inbox would be costly or embarrassing in a way you can't risk.
Process and reply to an email inbox: review template
| Item | Details | Status |
|---|---|---|
| [Auto-handled replies] | [Routine messages, per your rule] | [Sent] |
| [Flagged messages] | [Sensitive or multi-recipient] | [Held for your review] |
| [Reply-all / multi-recipient] | [Draft and recipient list] | [Approved by you before sending] |
The source case
On X, @NotZainAgain reports: "Just had Instinct go through my email inbox and deal with messages to the best of its ability" They add: "It told me it replied all on an email and some people were upset, but I’m sure it’s no big deal" and close with "Can’t wait to check in later and see how it did!" The post is written as the task is happening and doesn't report a final outcome, but it does describe a reply-all mistake that upset some recipients as it occurred. This is one person's in-progress report, checked on 2026-10-09; it has not been reproduced by this directory.
A public report by someone outside this library. Not independently reproduced here.
From the source
Source excerpt
Just had Instinct go through my email inbox and deal with messages to the best of its ability
Source excerpt
It told me it replied all on an email and some people were upset, but I’m sure it’s no big deal
Quoted from the original posts. Not written by this guide, and not a result produced here.
Suggested prompt
Bring Which message types are safe to auto-handle, and which aren't, A clear rule on reply-all specifically. Acts Reads and triages inbox messages. Drafts replies for messages you've said it can handle. Does not send a reply-all, or anything you've flagged as needing review, without your approval. Source community report, checked 2026-10-09.
Written by this guide as a starting point, not quoted from the source. Your edits stay in this browser.
Other ways to ask
Editorial rewrites of the brief above, for a different emphasis. Copy them separately.
What you need
Editorial, not from the source
You supply
- Which message types are safe to auto-handle, and which aren't
- A clear rule on reply-all specifically
Connected
- Read and send access to the inbox being processed
How to try it
Editorial, not from the source
- Give it access to the inbox and say which kinds of messages it can handle on its own versus which need you.
- Have it flag anything it's unsure about rather than guessing at a reply.
- Review the recipient list and the draft for any reply before it's sent, and require your approval for reply-all specifically.
- Check in afterward on what was sent, to whom, and whether anything needs a follow-up correction.
A mistake reported as it happened
@NotZainAgain's report is unusual among the sources behind these pages because it's written mid-task rather than as a tidy after-the-fact summary. They say they had Instinct go through their inbox and deal with messages to the best of its ability, then mention, almost in passing, that it replied all on an email and some people were upset, closing with a shrug and a plan to check in later to see how it did. That structure, a routine task that produces a visible mistake in real time, is worth taking at face value rather than smoothing over, since it's the clearest first-hand account among these reports of something going wrong.
Reply-all deserves its own rule
A reply sent to one person and a reply sent to everyone on a thread are different categories of risk, and the report's mistake is specifically the second kind. Before any inbox-processing task starts, set an explicit rule that reply-all, or any multi-recipient reply, needs your approval every time, not just as a general caution that gets forgotten after the first few sessions go well. Routine single-recipient replies, like confirming receipt of something simple, are a reasonable place to allow automatic handling; a reply visible to a whole thread is not, because the cost of a wrong tone or wrong audience is much higher and harder to walk back once it's delivered.
Check in on what actually happened
The report ends before its own outcome is known, with the author planning to check in later to see how the session actually went. Build that check-in into the task itself rather than leaving it for later: ask for a session summary of what was replied to, what was flagged for you, and whether anything, like an unwanted reply-all, needs a follow-up correction sent on your behalf. Inboxes that mix routine mail with anything sensitive or legal are better handled with a narrower scope at first, widening what's auto-handled only once you've seen a few sessions go cleanly and the summaries match what you expected to see.
Open Instinct to try it. That is the product's own site; this guide is not part of it.
Limits and confirmation points
- The report describes a reply-all mistake that upset recipients, told to the author after the fact rather than caught before sending.
- A reply-all sent to the wrong group, or with a tone meant for one person but seen by many, can be hard to undo once delivered.
- The report is written in the moment and doesn't say how the situation was ultimately resolved or whether any damage from the mistake was addressed.
- Some inbox messages (legal, financial, anything sensitive) likely need your own reply regardless of how capable the handling is for routine mail.
Editorial, not from the source
- Before it starts: set explicit rules for what it can reply to on its own.
- Before any reply-all: your approval, no exceptions.
- After a session: review what was actually sent.
Editorial, not from the source
- It replies all when a direct reply was what you wanted
- Set an explicit rule that any multi-recipient reply needs your approval before it's sent, with no exceptions.
- A reply-all mistake has already upset some recipients
- Review who received it, draft a short correction or clarification yourself, and tighten the reply-all rule before letting it continue.
- You're not sure what it's handled versus left for you
- Ask for an explicit session summary: what was replied to, what was flagged, and what's still waiting.
Common questions
Can an agent process and reply to my email inbox?
One user reports having it go through their inbox and reply where it could, but also reports a reply-all mistake that upset some recipients. This directory has not reproduced either outcome.
Will it avoid sending an unwanted reply-all?
Not by default, based on the one report available. Set an explicit rule requiring your approval for any reply-all or multi-recipient reply.
What kinds of messages are safest to auto-handle?
Routine, low-stakes ones, like simple confirmations. Anything sensitive, legal or multi-recipient is safer held for your review.
How do I know what it actually sent?
Ask for a session summary after each run: what was replied to, what was flagged, and what's still pending.
What if a mistake has already happened?
Review who received it, send a short correction yourself if needed, and tighten the rules (especially around reply-all) before letting it continue.
Next in Work
Get recurring task recommendations — ask your own agent to check for relevant updates.