An AI email assistant that keeps you in control

Email is where work arrives and where it gets stuck. For a business owner it is customers, suppliers and invoices. For a consultant, analyst or manager it is clients, colleagues and requests that each take five minutes, until there are forty of them.

An AI email assistant promises to take some of that weight. This guide covers what a good one does, what to watch out for, and a sensible way to start.

What a good AI email assistant does

Sorts what needs you from what does not

The most valuable thing an assistant can do is not write anything at all. It is telling you which of today's messages actually need a reply from you, and which are newsletters, notifications or threads you are only copied on.

Look for sorting by need, not only by sender: "needs a reply", "has a deadline", "for your information". A good assistant explains its reasoning in plain words, so you can trust it or correct it.

Summarises long threads

A thirty-message thread with three people disagreeing about dates is a good job for an assistant. Ask for the summary: what was decided, what is still open, and what you are being asked to do.

Drafts replies you can edit

A first draft removes the blank-page cost. A good assistant drafts in the tone you ask for, whether formal, casual or friendly, and keeps it short unless you ask otherwise. You edit and send.

Some assistants can also learn your own writing style from examples of your writing. This is worth having, but it should be something you set up deliberately, with writing you choose, not something it quietly builds from everything you have ever sent.

The line it should never cross: sending on its own

Sending an email cannot be undone. An assistant that sends on your behalf without asking can put a wrong date, a wrong price or the wrong tone in front of a client, under your name.

The rule to look for: drafting is the assistant's job, sending is your call. It can prepare the reply, show you exactly what will go out and to whom, and wait for your approval. If a product leads with "fully autonomous inbox", ask what happens when it gets something wrong.

Questions to ask before you connect your inbox

1. What can it read, and what does it store? Connecting an inbox is a real permission. You should know where the content is kept and how to disconnect.
2. Does it send without asking? It should not, at least not by default.
3. Can you see why it sorted something the way it did? Opaque sorting is hard to trust.
4. Does it work with the accounts you use? Check your provider is supported today, not "coming soon".

A sensible way to start

1. Week one: sorting only. Let it tell you what needs a reply. Check its judgement against your own.
2. Week two: summaries. Use it on long threads and on the mornings after time away.
3. Week three: drafts. Let it draft replies to routine messages: confirmations, scheduling, polite declines. Edit before sending.

By the end of three weeks you will know whether it saves you time, and you will not have handed over anything you cannot take back.

How Casitka handles email

In Casitka you connect Gmail (more than one Gmail account if you need) and the inbox sorts your mail by priority, marking what needs a reply. You can ask for a summary of any email, and draft a reply in the tone you choose. You can also ask your companion in chat to draft an email for you. If you have set up your writing style in Casitka's AI Twin, those chat drafts follow it.

Sending is treated as an action that cannot be undone, so by default it waits for your approval. Inbox drafts are only shown to you, never sent on their own.

Casitka works with Gmail today. If that fits how you work, see the plans or read how it works.