A useful way to think about AI and email: your inbox contains two different kinds of work wearing the same clothes. There's reading work (figuring out what each message is, whether it needs you, what it's asking) and there's judgment work (deciding what you'll commit to, and how to say the things that matter). AI has gotten genuinely good at the first kind. It has no business doing the second.
That split gives you a clean rule for delegation. Here's how it plays out across the four jobs an AI email assistant actually does.
Delegate: triage
Of the ~120 emails the average professional receives daily, a small handful need a decision from you today. The rest are fyi's, receipts, newsletters, threads you're cc'd on, and scheduling noise. Sorting those piles is pattern recognition at scale, which is the thing machines beat humans at without breaking a sweat. An AI assistant reads everything and hands you the short list: these five need you, the rest is handled or filed. The overwhelm of an inbox was never really the work in it. It was not knowing which parts were work.
Delegate: extraction
Requests arrive buried. "Great meeting yesterday, quick thing, could you send the updated figures before Friday's board session?" is a task with a deadline hiding in paragraph two of a thank-you note. Humans miss these when tired or busy, which is when most email gets read. AI reliably pulls the ask out and turns it into a verb-first task with a date and a link back to the source. We wrote a full piece on the format in how to turn emails into tasks.
Delegate: first drafts
Most replies are routine: confirming, scheduling, sending a file, answering a question you've answered before. Drafting these from scratch costs minutes each and adds up to a real slice of the McKinsey figure of 2.6 hours of daily email time. An AI that has read the thread can draft the reply in your tone, and you edit or approve. The review takes seconds. The blank page took minutes.
Delegate: follow-up
Everything you send that expects a reply becomes an open loop, and human memory is a lousy tracker of open loops. AI is a perfect one. It knows what you sent, when, to whom, and whether an answer came back, so it can maintain your waiting-for list automatically and draft the polite chaser when someone goes quiet. This is the least glamorous of the four jobs and possibly the most valuable, because dropped follow-ups fail silently.
Keep: your commitments
When an email asks for your time, only you know whether the answer is yes. The AI can surface the request, estimate the effort, even note that your Thursday is already full. But saying yes creates an obligation, and obligations you didn't consciously accept are how calendars turn into other people's to-do lists. Approve every commitment yourself. It takes seconds and it's the whole game.
Keep: the messages that carry weight
Delivering hard feedback, apologizing, negotiating, writing to someone who matters to you: these live or die on judgment about the relationship, and drafts of them shouldn't be outsourced even if they could be. A reasonable test: if the recipient would feel differently knowing a machine wrote it, write it yourself.
What this looks like in practice
Pidgy is built on exactly this split. It reads your Gmail or Outlook inbox and turns it into three lists: next actions that need you, waiting-fors it tracks and chases, and reference it files. It drafts routine replies in your voice and queues follow-up chasers for one-click approval. Nothing sends without you, and no commitment enters your list without you accepting it. The reading work disappears; the judgment work gets a clean surface to happen on.
If you're evaluating any AI email tool, ask three questions before connecting your account: is your mail used to train models (it shouldn't be), does the vendor hold recognized security certification like SOC 2, and can you disconnect and delete your data cleanly. Good answers to those three are table stakes.
