Why we don't auto-send drafts (and why that's a feature)
Every AI email tool we tried wanted to send emails on our behalf. We think that is the wrong default — and here is why.
When we were building xNord, one of the first features every early user asked for was auto-send. "Just send the drafts automatically," they said. "I trust the AI. I do not want another review step."
We considered it seriously. We built a prototype. We tested it. And then we deliberately chose not to ship it as a default — and we think that decision is one of the most important product choices we have made.
The problem with auto-send
Email is not a low-stakes medium. An email sent cannot be unsent. An email that misrepresents your position on a deal, accepts a meeting you did not mean to accept, or responds to a legal matter in the wrong tone can have real consequences.
AI language models are very good at generating plausible text. They are not always good at understanding the full context of a business relationship, reading between the lines of what is actually being asked, or knowing when a situation is more sensitive than it appears on the surface.
We have seen draft replies that were technically correct but commercially wrong. A draft that accepted a call with a competitor. A draft that committed to a timeline the founder had not approved. A draft that was perfectly written but addressed the wrong point in a negotiation.
None of these drafts were bad. They were just missing context that only the founder had. And that context can only be applied at review time — not at generation time.
The two minutes argument
The pushback we hear is: "But the whole point is to save time. If I have to review every draft, I am still spending time on email."
This misunderstands what makes email time-consuming. The expensive part of email is not sending — it is triaging, deciding what to do, and writing. xNord handles all three of those. The review step is genuinely fast: you read a two-sentence summary, skim a draft that is usually right, and click send. Most founders are doing this in under 2 minutes a day.
Compare that to the alternative: reading every email in full, deciding what each one needs, and writing a response from scratch. That is where the time goes. Review is not the bottleneck.
Trust is earned gradually
We also think about trust differently to most AI products. The default assumption of "trust the AI" is a bad starting point for something that speaks in your name to investors, partners, and customers.
The right model is: start with propose-only, observe how the agent performs on your specific inbox, and gradually extend autonomy to categories where it consistently gets it right. Archive newsletters automatically? Fine — the risk is zero. Auto-reply to investor follow-ups? That needs more trust than two weeks of data can establish.
We are building toward more automation. The Rules engine already lets you define what gets archived automatically. Future versions will let you approve auto-send for specific sender categories or email types. But we will always make the default conservative, because the cost of a wrong send is higher than the cost of a 10-second review.
What we actually shipped
Instead of auto-send, we focused on making the review step as fast as possible. One-click approval. Keyboard shortcuts. Bulk approval for multiple drafts. The goal is that reviewing a draft takes less time than it would have taken to write a reply yourself — not that it takes zero time.
We think this is the right product. We are happy to be proven wrong by the data over time. But right now, the founders using xNord tell us that the review step feels like quality control, not overhead. And that feels like the right relationship between a founder and their AI agent.