From Zero ML Engineers to Silicon Valley's Favorite AI Feature

In February 2023, nobody at Superhuman knew anything about AI. By June, they had launched a feature that "captured the hearts and minds of Silicon Valley."

Rahul Vohra, founder and CEO of Superhuman—the email client that promises to make you twice as fast—found himself facing an existential moment. With no machine learning engineers on staff and no time to hire, he made a radical decision: abandon all company processes and run AI development like a scrappy startup within his 110-person company.

In this interview, Vohra reveals how a small tiger team, founder fiat, and strategic chaos helped Superhuman beat the competition to market—and why building for individuals first was actually the hardest path to sustainable growth.

Rahul: I think we started last year in February, and believe it or not, nobody really knew anything about AI. We didn't have a machine learning engineer on staff, we'd never needed one. It just didn't turn out to be important for what we were building. We hadn't really used large language models ever, we were sort of vaguely aware, I mean, how could you not be aware of what was going on.

But it was around January, February when it became incredibly clear to me, as I think it did all of us, that it's an existential moment. And these moments don't happen particularly often in the history of startups, but it was obvious that it was existential, that everything would change, and that this is something that had to become one of the core pillars. It's like a top 3 priority—if we have 3 priorities, it's one of them at the company.

Rahul: Because we didn't really have anyone on staff who knew what they were doing, and we didn't have time to wait and hire for this, and also, this was something that was going to be existential to the company. We actually approached it in a very different way than we do most new initiatives.

For context, we, at the time, had around 100 people, around 110 people. Even so, we didn't assign this project a product manager, as you might at a company of that size to run it. We decided that I would actually run it personally. And fortunately, I just hired a president, so I was able to hand over day to day operation and management of the company to Paul, our president, and we created a small tiger team, if you will, of myself, a designer, and a handful of engineers.

The idea was to essentially run it like a startup, devoid of all process. We just get together regularly and build stuff and with founder fiat. I think a lot of what a founder has to do sometimes is, in the absence of data, just say, this is what we're gonna build, trust me, I will own it if it's wrong, and if it's right, we'll move on and build the next thing. But you need to be able to short circuit the decision paralysis that can sometimes happen, especially as companies get a little bit larger and debating things becomes the way in which decisions are taken.

Rahul: I wanted to be able to get at it really, really quickly. So it was small, it was fast, it was chaotic, it was stressful. We did it really, really quickly, and a few months later, I think it was June of last year, we launched Superhuman AI, the very first version, which lets you write with AI so you can jot down a few notes and we'll turn it into a fully written email that's in your own voice and tone, we match all the emails you've already sent.

This feature that we're now all very familiar with in basically every application that we use. That was the very first thing that we built. It really captured the hearts and minds of our userbase, of Silicon Valley, it was a really big deal at the time.

And that's an example of what we call on-demand AI. Relatively easy to run, cheap to run as well, because it's only operating when people actually click the button, or when they use the feature. But that gave us the confidence to move on to the next phase of AI which we call always on AI.

Rahul: So you move from on-demand to always on, and this is AI that is just constantly running in the background all the time. Much harder to build, much more expensive to run, and the very first example of that that we built was auto summarize. So imagine when you receive an email, a one-line summary that is pre-computed and always available on top of that conversation. And when new emails arrive, it instantly updates.

People really love that, and that then gave us the confidence to build instant reply, which is also an always-on feature. Imagine waking up to an inbox where every single email already has a draft reply. You would just hit edit and then send, and sometimes you wouldn't even need to edit.

And that then gave us the confidence to build the next thing. So we've been increasingly working up this confidence ladder, I would say, taking on more ambitious, more expensive projects. And now I'm not an IC on that team, but for the first 6 months I was, and it was really to kickstart it.

Rahul: This topic of competition has obviously come up so many times over the years at Superhuman. Every single time there's been a new email client, for example, in the early days, maybe some people in the company would panic, and I would talk about it, and we'd get on with building great software and delivering value to our customers, and every single time it's just come and gone and come and gone and come and gone.

And Paul Graham has a good piece of advice about this. Which is, don't obsess over the competition. It actually doesn't really matter. Just focus on making your own customers, your own users, incredibly happy. So that's what we've always done. And I think the AI wave is no different. I think the agent wave is no different.

We see superhuman as a core part of this evolution of the productivity landscape. We are working really hard towards building AI agentic future. We believe in a world where one day you will have a superhuman AI agent, and you can start to see if you use the product today, the beginnings of that with features like Ask AI.

Rahul: Let's actually think about who are the people who send and receive a ton of email. Who are the people who do email for a living? So founders, CEOs, obviously, most people in this room. If you're not already doing email for a living, as your companies grow, you soon will be. But it goes much deeper than that. It's also account executives, account managers, customer success, business development, talent, recruiting, business owners, realtors, leaders and managers of every kind that you can imagine.

They all do email for a living, and a clear part of their jobs actually, is to be responsive, because if they're not responsive, then they can block their team, miss opportunities, or in the worst case, even damage their own reputations. So there's both a carrot and a stick there. There's the danger of doing email poorly, which sadly most people do, and then there's also the benefit of doing it really, really well. Because it's clear to the job, that's why people are willing to pay.

Rahul: Just over the course of the last year we have scaled from relatively small and mid-market startups to teams of several thousand, which is really exciting to see, and that kind of revenue ramps really quickly.

It's much easier to build a strong business when you're selling to teams. When you sell to an individual, it's a double-edged sword. The good news is that you only need to convince one person to adopt the tool. However, ultimately, it's harder for that person to stick around because at some point someone else may come along and tell an equally clear and pointed story.

In general, there have been very, very few companies in software that have scaled to be extremely large off the back of single player subscriptions alone, and we could probably name them right here. They're companies like Spotify and Netflix. There are very, very few that have scaled to truly billion dollar single player subscription companies. And the things that make those companies feasible is that they can credibly and reliably, year after year, add tens if not hundreds of millions of new subscribers to the top of the funnel every single year.

So, if you're building a subscription business, it is much more robust to build something that you can sell to teams. It's harder in other ways because now you have to convince a team of people to adopt it. But once a team of people does adopt it, it is so much stickier, it is so much harder for them to churn out and to use something else.