I spend a lot of my day in email. Not because I love it, but because ignoring it makes everything worse. Every unanswered message sits in my head like a tiny unpaid debt. So when I started building Replyay, I had a very specific fantasy: an AI that would just handle the replies for me. I wanted to open my inbox, see twenty responses already drafted and waiting, hit send, and be done with it.
Turns out, that fantasy was naive. And three specific emails are the reason I changed my mind.
Why I thought faster replies were always better
My inbox used to hover around 150 unread messages. I know, I know, some people have thousands, but for me 150 felt like drowning. Every time I opened Gmail I got this low-level dread. I’d scan subjects, ignore what looked complicated, reply to the easy ones, and close the tab feeling like I’d barely made a dent.
The idea of an AI email reply tool that could draft responses for me felt like a lifeline. I imagined myself waking up, checking my phone, seeing all replies pre-written, approving them, and moving on with actual work before 9am. I got so excited about the speed part that I forgot about the quality part. About the relationship part. About the fact that some emails are not asking for information, they are asking for attention.
The first email that stopped me
A customer wrote to me asking why their billing had gone up from $29 to $49 a month. Simple enough question. My assistant at the time drafted a polite five-line response explaining the price change had been announced two months earlier, linking to the FAQ, and offering to help with billing questions.
I read it and almost hit approve. The draft was clear, factual, and accurate. Everything a good AI email reply should be.
Then I stopped. Because I realised the customer probably already knew about the price change. That wasn’t the real question. The real question was: “I liked your product at $29, do I still like it at $49? Convince me.”
The AI draft had answered the literal question, which was “why did my bill go up”, but it had completely ignored the emotional question underneath. That customer was not confused. That customer was questioning whether to stay. A sterile, accurate reply would have just confirmed their suspicion that nobody on my side actually cared. I rewrote the whole thing. Apologised for not communicating the change more clearly, explained what we had improved since the price change, and offered to hop on a quick call if they wanted to talk about their specific usage.
They stayed. And I learned something obvious in retrospect: an AI reply tool can be fast, but fast at answering the wrong question is not helpful.
The second email made me rethink tone entirely
Someone I had never met sent me a long, frustrated email. They had been trying to integrate with our API for two days, it wasn’t working, their documentation kept contradicting itself, and they had already talked to two support people who gave conflicting answers. The email was not polite. It had all caps sentences. It used words like “frustrating” and “unacceptable.”
I tossed it into an AI assistant to see what it would come back with. The draft was impressively diplomatic. It thanked the customer for reaching out. It apologised generically. It offered to escalate to the engineering team. Perfectly professional.
But reading it, I felt cold. It was the kind of reply you send when you’re trying to cover your ass, not when you’re trying to fix a problem. Everything about the draft was designed to minimise liability and move the conversation along, not to actually address the fact that someone had wasted two days of their life because of something I should have caught.
I ended up writing back personally. I said I was sorry they’d gotten bad information, that I’d personally look into the docs, and that I would send them a working code sample within four hours. The AI email reply drafted by my tool was smoother and more correct than what I sent. But what I sent was honest, and it came from a place of responsibility instead of a place of protocol. The customer wrote back six hours later thanking me and saying they appreciated the straight talk.
That’s when I realised: tone is not something you can optimise for neutrality. The best replies in business email are not the safest ones. They are the ones that sound like a real person who owns the problem.
The third email broke my assumption entirely
This one was from a fellow founder, someone I had met once at a conference. He was asking a favour, a referral to someone in my network. His email was short, genuine, a little awkward. He basically said, “Hey, I know we don’t know each other well, but I’m trying to find a good developer for a project and I remembered you mentioned someone.”
My AI assistant drafted a perfectly polite, perfectly structured reply. It offered two names, included brief context about why each might be a fit, and closed with a nice sign off. Technically flawless.
But I didn’t send it. Because what this email actually needed was rapport, not efficiency. This person was reaching out with a small amount of vulnerability. He was asking for help from someone he barely knew. A perfectly polished response would have made the distance between us feel bigger, not smaller.
I wrote back with a single paragraph, no formatting, one typo that I didn’t bother fixing. I said something like “Yeah of course, talk to this person, they’re great for that kind of thing, let me know if you want me to introduce you properly.” It was sloppy. It was fast. But it was human. And that’s what the situation called for.
The guy called me the next week, said the referral worked out, and we ended up having a solid conversation about our respective products. That does not happen if I send the clean, polished AI email reply. That happens because I left a typo in and sounded like someone who was too busy to be formal but not too busy to help.
What I changed after these three emails
I did not stop using AI to help me reply to emails. That would be stupid. I get way too many messages, and drafting from scratch for every single one would be exhausting. But I did change how I use it.
I stopped using AI for emotionally charged emails
If someone is angry, scared, confused, or asking for something personal, I write the draft myself. I might use an AI email reply tool to check my phrasing afterwards, but I do not let it generate the first draft. The first draft sets the tone, and I want that tone to be mine, not an average of every corporate email ever written.
I started treating AI drafts as a first pass, not a final draft
It sounds obvious but it took me too long to accept. An AI reply is a starting point. It gets the facts right, it structures the information, it closes with the right call to action. But then I rewrite the first and last sentences to sound like me, and I delete any sentence that feels like it was written by a committee. If I would not say it out loud to someone standing in front of me, I do not want it in an email.
I built the tool to help me, not replace me
This is the part that matters most for what I am building with Replyay. I realised I do not want an inbox that replies itself. I want an inbox that helps me reply better. There is a difference. One takes the work off my plate entirely. The other helps me do the work more intelligently, faster, but still with my name on it.
I see a lot of AI email assistants positioning themselves as full autonomy. Just connect your inbox and let the machine handle it. Maybe that works for newsletters and spam and automated confirmations. But for the real emails, the ones from customers and colleagues and people who are trusting you with their time, I do not want automation. I want assistance.
What I still believe about AI and email
I still think the future of email is not more time spent in your inbox, it is less. I still think AI is the best thing that has happened to email since threaded conversations. And I still use an email assistant every single day to draft replies to the routine messages that used to eat up my mornings.
But I do not let it write the emails that matter. And the strange thing is, once I made that distinction, I got faster overall. Because I stopped reading every AI draft with suspicion, wondering if it was good enough to send. I stopped tweaking perfect sentences that technically said everything but felt like nothing. I let the tool handle the obvious replies and I handled the ones that needed a human brain and a human heart and sometimes even a human typo.
That is the balance I am still working on. Three emails taught me that speed without intention is just noise. And a reply written too fast, with too much polish and not enough honesty, can undo a relationship in seconds. I would rather someone wait four hours and get a messy reply from me than get a perfect one in thirty seconds that makes them wonder if I actually read their message at all.