An AI can write a perfect email in seconds. Having a real back-and-forth conversation is much harder.
Sarah runs a salon. She has an AI assistant that emails her customers when a slot opens up.
Last Friday, a customer canceled his booking. The assistant sent an email: "We have an opening tomorrow at 2 PM. Want it?"
The customer replied in a minute: "Yes, book it!"
The assistant never saw that reply. The slot stayed open. The customer never got a confirmation.
This happens more than people realise — not because it's hard to receive email, but because most setups were never wired to close the loop.
Sending is easy. Wiring the whole loop isn't.
To be fair, receiving and parsing email isn't some unsolved problem — providers like SendGrid, Mailgun, and Postmark have offered inbound email parsing for years. Point your domain at them, and they'll hand you the clean message.
But those webhooks only push the message once. There's no inbox to check back later, and no built-in way to link a reply to the right conversation. You have to build that part yourself — and you still can't run any of it on your own servers.
There's a second issue too. AI assistants sometimes send a slightly odd reply — nothing harmful, just a little off. Many managed email providers watch for exactly that pattern, and can suspend an account fast. One strange sentence, and Sarah's whole booking system could go dark with no warning.
What a real AI assistant needs
For an assistant like Sarah's to actually hold a conversation, a few things need to work together:
- Replies need to land somewhere the AI can read them
- They need to arrive clean, not messy
- They need to stay linked to the right conversation
- The AI needs to reply back from the same email address
- All of it needs to run on infrastructure you control, not three different vendors
This is what we built Reloop for
Reloop puts that whole loop in one place, self-hosted.
When Sarah's customer replied "Yes, book it!", Reloop caught the reply, cleaned it up, and kept it linked to the right conversation. Sarah's assistant read it, checked the calendar, and sent back a confirmation — from the same salon email, so it felt like one normal conversation.
Reloop doesn't write anything itself. It just makes sure a reply doesn't disappear, and that a response can go out without a compliance system flagging it by mistake.
One honest note: right now, Sarah's assistant checks in for new replies rather than getting notified the instant one arrives — established providers like SendGrid already push replies via webhook. A push-based version for Reloop is coming next.
Owning it means watching it
None of this makes an AI assistant automatically safe.
If it ever sends something odd at scale, big providers like Gmail can still block the domain — that risk doesn't go away just because you self-host. What changes is who's responsible for catching it. You decide which addresses it can email, how many it can send per hour, and when a human needs to approve something before it goes out.
That's the real trade. Not "nothing can go wrong" — just "you're the one deciding what's allowed."
The takeaway
The AI writing a good reply was never the hard part. Holding a real, two-way conversation is.
If you're building something like Sarah's assistant, that's the part to get right first.
Check out Reloop on GitHub — it's open source and free to try.
Top comments (1)
I like that you focused on the conversation loop instead of the email itself. That’s where most AI assistants break down.
One thing I’d add is idempotency. Email is an eventually consistent system—retries, duplicate replies, and delayed messages are normal. Treating every incoming email as an event with deduplication and replay support makes the system much more reliable.
At that point, it starts looking less like an email tool and more like an event-driven workflow engine, which I think is the more interesting direction.