The $350M Startup Creating Perfect Memory

Yeah, the Obama campaign in 2007–2008 very much felt like a startup. There was definitely that esprit de corps and closeness, especially when I joined back in November 2007. Barack Obama was then third in the polls, behind Hillary Clinton and John Edwards. So everyone at that campaign was really there because they believed in the candidate — not because they thought he could win, but because they really believed in him. For me, it was a question of: would I rather give it a shot? And if he failed and didn't make it, would I regret that more, or would I stay on the sidelines and regret that? That's why I did the campaign.

And what I learned was this risk-reward mentality. David Plouffe, the campaign manager, coined that phrase — the risk-reward mentality. The idea is that if we did everything everyone else did, we'd end up how everyone expected: third. So to actually make some decisions that broke the norms was kind of expected. That's why we did this crazy thing called A/B testing — which was actually not at all crazy in the private sector. I'd seen it up close and personal at Google. In many ways, Google search and ads are fundamentally built on this concept of A/B testing. I just brought that concept to the campaign, and a lot of our decisions used data to inform them, which is a pretty abnormal way to run a political campaign. Typically the important decisions get made by the grizzled veteran with the white hair, smoking a cigar in the room. It's not this analytical, data-driven approach. And it turned out to work really well.

Absolutely. I've written a book about A/B testing, and a big part of it is trying to teach people that it isn't a panacea, and it's not a solution to every problem. When I was running Optimizely, I used to tell CMOs and other executives that their job is to ask the right questions, not to have all the answers. A lot of CMOs — and I mean chief marketing officers, not chief medical officers — think their job is to have the great tagline, to have some intuition around the market. But really, as a leader, both of a small startup and as a big-company executive, your job is to ask the right questions. A/B testing and randomized controlled trials are a great way to get answers to those questions. But it starts with a hypothesis. You need a hypothesis that's rooted, often, in some intuition.

And the best examples of A/B tests and experiments are when you run the experiment, you had an intuition, and you realize your intuition was wrong. The most breakthrough experiments aren't the ones where you validate your existing hunch — it's when you learn something new about the world and use that to inform the exploration. So that's how I view A/B testing on marketing and websites, but also being a startup. At the end of the day, as a startup, you're navigating this foggy world. Imagine climbing a hill: you don't see perfectly how to get to the summit, but you know that every step you take, as long as you keep moving, helps you understand better what the topography looks like. You might go down one path, learn, iterate, put it in front of users or customers or patients, learn it doesn't work, come back, and go down a different path. And ultimately, as long as you don't give up and you keep learning, you'll eventually get to the summit. That's the thing I've seen universally through my experience at Google, at the campaign, and now at Rewind.

Yeah. The best book I've read on this topic is called The Mom Test. It helps you talk to your customers in a way where you get the truth, because your customers are lying to you. If you go to your customers and say, "Hey, here's my demo, here's my product, what do you think?" — you're never going to get an honest answer. People don't want to hurt your feelings or your ego. So navigating this metaphorical mountain of building something in the world is, in many ways, about learning from customers and building intuition around where to focus and where not to.

I have had times where I had to just listen to my intuition, and that leads down a path that's unpopular. For better or worse, more times than not, when I go against the grain, I've turned out to be right — which is not necessarily a good feedback loop. I'm probably going to make a bunch of dumb decisions because of that. But a good example: we raised our Series A in public. We put a pitch deck out — seven minutes on Twitter — it went viral, millions of people saw it, we got thousands of offers to invest, and ended up with a great term sheet and a $350 million valuation. Even that first idea of putting the pitch deck out publicly wasn't a consensus-good idea; a lot of startups don't do that.

More recently, I published all of my 360 performance reviews. Every quarter we do 360s at Rewind focused on how we live up to our cultural values, and I put that out there publicly. Most CEOs wouldn't have the courage to do that. Time and time again, especially with transparency, I've learned that going against the grain and following your intuition ties back to that Obama campaign risk-reward mentality. The reward has been there, so I'll keep doing it.

I'll start by saying privacy is an important thing to be concerned about, so I'm actually really happy people are asking these questions, because that's going to force us and our competitors to really get it right. If you look at the history of technology, most new technology goes through four stages of evolution. They start with people being surprised or shocked; then they're in awe — how is this possible; then they get to fear — could this kill me? And eventually it becomes accepted. I've coined this acronym: SAFE — Shock, Awe, Fear, Expected. There are so many examples throughout the history of technology, from electricity to airplanes to self-driving cars to Rewind, where because they're so new and different, people rightly react with concern.

So it behooves companies like us to really get it right from the beginning — to not make privacy a trade-off with convenience, to have a product that is privacy-first and still delivers the value of the technology. How are we doing that specifically? Well, we see the product we're building today as a personalized AI. It is not a covert surveillance or recording tool. At its core it's a product for individual users — for you to have better memory, to go back to any moment in your past, to use that to live a more productive, more present life, because that AI can do things you'd otherwise spend your time on. And indirectly, if there are things in your world you want to capture — conversations you have with others — we have tools to make that easy while also respecting the privacy of others.

Then we have some specific features we're contemplating that I think will make a big difference for the Pendant in particular, which is a wearable. One example we're excited about is voice fingerprinting and speaker diarization. So if you and I are having a conversation and it's being recorded by the Rewind Pendant, everything I say is captured, because I've opted in and it's my device. But everything you say is ignored — until you say the phrase, "Yes, I'm okay being recorded." It's basically as if it didn't happen. That's possible through voice fingerprinting and speaker diarization. It's one of many examples of how technology can give you the benefit and convenience and power without the trade-off on privacy.

We're actually contemplating not storing them. Right now they are, for the purposes of summarization. There's a good product that just came out from Zoom called Zoom AI, which tries to strike the right balance — giving you the benefit of Zoom recordings without the verbatim recordings. Interestingly, they launched without transcripts. It obviously generates a transcript, but at the end of the day you get a summary. And that strikes a pretty good balance, because you get the content of the meeting, but if in the meeting you say something you might regret, or that might get you canceled, you have the ultimate plausible deniability — there's no recording, no transcript of what you said. So we think that approach could work really well for a product like the Pendant too.

Yeah. I think there are two points. One is plausible deniability, because what's captured or stored isn't the verbatim transcript recording. The second — which I think is actually going to be more prevalent — is that today, voice-cloning technologies are so good that you can literally create a recording of anyone on the planet saying anything. That already exists today, it's already really good, and it's only going to get better. You can do it trained on just a few simple sets of words.

When that becomes so prevalent, it creates plausible deniability for anything you've said. Any recording of you is just as legitimate as anyone who had written down what you said — or even summarized what you said — because it could just as easily have been generated by an AI. So I think that might also change norms dramatically, as people realize: I can't trust anything anyone said if it's a recording. It's indistinguishable from what an AI could say. We're not building that, obviously, but others are. And because that will be out there, I think people's knee-jerk reaction to a recording that feels official and could be used against them will very quickly go away.

Yeah, and I don't think it's societal change writ large. I think there's probably 5–10% of society who will have to change their behavior, and those people probably should. I operate under the assumption that everything I say is recorded, that anyone could say it or use it. I don't ever tell one person one thing and another person a different thing. I don't lie, I don't gaslight — and it's not because I have some higher moral character. I just know at some point maybe there's a recording and maybe there's not, so I change the way I interact with the world.

There are some people — and they tend to be of the older generation — who don't operate that way. Frankly, I think it's good for society, good for workplaces, good for avoiding harassment, if people operate under the assumption that maybe what they're saying is being recorded, and maybe they shouldn't say the thing they'd be afraid to have recorded. I think that might actually be good for society.

Yeah, and that's a fair criticism. You can say that of any privacy-invading or surveillance technology. But I think this is different. This is not about the state owning these recordings, not about the state persecuting you, not about Big Brother. This is about two people, one of whom wants better memory — they want more time with their kids, they don't want to sit in a meeting taking notes, which is a legitimate desire. And at the same time they're using a technology that does its best to respect the privacy of others. I think it's fair to say that people who want to live a better, more fulfilled, more present life shouldn't have to play second fiddle to somebody who's paranoid that an off-color joke gets used against them in the future. And again, with deepfakes, I think that risk goes away too.

Oh boy, do I have thoughts — and some of these will sound quite crazy. I'll start with the things Rewind does today that are, not by design, beneficial to people who are differently abled. One is ADHD. We were very surprised how many people, from almost the very first version of our product, said how Rewind gives them the ability to focus again — they can be productive members of meetings and society because they can go back, retrace their steps, and not be constantly lost. That was a huge surprise to me, and in hindsight it makes a lot of sense: the tool makes it really easy for them to be just as focused as somebody without ADHD.

There's also a whole class of people — I'm hearing impaired, I wear a hearing aid — where being able to capture and recall matters. My wife would be thrilled if I remembered, or frankly even heard, the thing she asked me to grab at the store right before I go to the market.

More broadly, a lot of medical professionals think about the normal human condition, and then diseases, or people who are sick or have conditions. But I very strongly believe that we live our lives with blinders on. We're not aware of the limitations of our own biology, and only when you have technology that can augment and overcome those limitations can you feel what it's like. That's why we say our vision is to give humans superpowers — because that's what it felt like to lose my hearing and gain it back again. I didn't realize until I tried the hearing aid how bad my hearing had gotten. The thing that was even more surprising than being able to hear again was the realization of how bad it had gotten.

I think that's true of memory too. It's true that we forget about 90% of what happens after a week, but most people don't think that or feel it, because they don't know what they don't know. But if you had a technology that gave you the feeling of remembering everything, of never missing a beat, of being present in meetings — those are the kinds of feelings that can transcend what are normally considered the limitations of a normal person's biology. That's something I'm even more excited about than helping people with ailments get to normal: taking normal and making it super-normal, so we all live a better life because of technology.

I strongly disagree — I think that's a bunch of baloney. First of all, we're not forcing you to recall these memories. We're giving you the option to go back to them, and having the ability to go back is far better than not, in almost every circumstance. And in situations where it's traumatic, you just delete the recording. We're not forcing you to keep it forever. So it's almost a knee-jerk reaction — maybe dystopian sci-fi from watching too many Black Mirror episodes — where you think you'll somehow be haunted by your memories. It's like saying, if you could have superhuman strength — well, I don't know, because you might open the can a little too hard and scratch yourself on the lid. I don't think there's much merit to it. If you want to forget, just delete it. We're not forcing you.

The majority of things you forget; a few of those things might be traumatic — just delete those from the product. And it's also not the kind of thing... it's not a camera you wear. It's the things you say and the things you hear, usually in a work setting. So, sorry, I'm not trying to be cavalier — it's just I've heard that, and I think it loses sight of the benefit you get, for the few cases where you might want to just delete what you captured in the past.

A couple of things I'd predict. One: I think we'd all be much better communicators. A lot of miscommunication comes from misremembering and selective remembering. So communication, commitment, agreement — having impeccable commitments to the people around you — would be dramatically better when you don't forget and don't misremember what was said. I think that would go a long way to resolving conflict, improving relationships, and improving organizations, because miscommunication is the core of so much conflict.

I also think that just because you have the ability to go back and remember doesn't mean you're somehow stymied in a way that prevents you from coming up with new ideas. In some ways, people say that as you get older you get more stuck in your ways. If you remembered your past truthfully, I don't think you'd be more stuck — you'd remember, "boy, I was pretty adaptable, I could change my mind earlier when I was younger; I do remember a time when I had an opinion and changed it with new evidence." Whereas often the criticism of older generations is that, even given new information, they don't change their ways. So I don't see it that way. We're not yet working on things like immortality, but I think better communication and better memory might have bigger impacts on society.

I'd describe the class of tools you mentioned — Evernote, Notion, Roam, Obsidian — as personal knowledge management, PKM. Evernote was a pioneer and built a following around this idea; in many ways it's the same vision we have of giving humans perfect memory and superpowers. Their logo is an elephant — an elephant has great memory. The difference between those tools and ours: note-taking requires you to actively predict, or have foresight about, what you might want in the future. The challenge is you don't have perfect foresight. You don't know in a meeting which note to take and which not to.

And there's a common misconception about note-taking. People think, "I'll take notes because it'll help me remember better." You will remember the thing you write down — but as you're writing it, the thing being said immediately after, you'll almost certainly forget. Your mind and body can't do both at once. So even by taking notes, by hand or digitally with a tool like Obsidian, you're still missing a lot of what's being said, because you have to actively take notes.

The fundamental difference is that note-taking is active and requires foresight; Rewind is passive and does not. It passively captures everything you see and say and hear, everything that appears on your screen — including your notes, if you have your note-taking app open. We do optical character recognition on it, turn it into words you can store in a database, and only if you need it later — which oftentimes is exactly the thing you didn't take notes on — you can go back and find it. So it's a fundamentally different approach to the same problem of remembering better, but without requiring you to be as active.

Absolutely — we're just getting started. It doesn't quite do as much as it could yet; there's a lot of cruft, and despite that it still works surprisingly well. But right now a focus of ours is to create and understand better the things we capture — not just capture it, but understand it. That's where large language models and embeddings and a lot of these new technologies make it really easy. We do think a knowledge graph is something we've explored and will continue to be helpful, in service of this greater power of giving you a personalized AI. But we're starting with the table stakes: capture everything, make it as robust as possible in how we capture it. And then, as you think about this pyramid of data to information to knowledge to wisdom, we want to move up the pyramid using the data we collect.

You got that great, by the way — you really did your homework. You got almost every major technology. The only thing you missed was automated speech recognition: taking a conversation like this and transcribing it. But everything else is the core way we stand on the shoulders of giants. With the M1 and M2 chips from Apple, it really is incredible what we built; even two years ago it wouldn't have been possible without that underlying hardware. In particular, the architecture is called a system on a chip, where things that would otherwise have been software running on your CPU have been completely offloaded to purpose-built hardware. We're in total debt and gratitude to the engineers of that product.

But to answer your question, the big breakthrough that I think will change our product — and the way people think about our product in terms of privacy, and every product like ours — is when the ability for machines to offer what GPT-4 currently offers, the best large language model from OpenAI, can be done entirely locally, using a local large language model. Today the best local models are Llama 2 and some more recent ones; the pace of innovation there is staggering. It's amazing how much better these are getting. So it's possible we're on the order of maybe months, maybe a year, away from models that are as good and efficient locally as GPT-4 is in the cloud.

The reason that's so powerful is that a product like Rewind can then work entirely locally. You can turn your Wi-Fi off. Everything we do — asking questions of anything you've seen, said, or heard — currently, everything we capture we store locally. We don't have access to it. If we're subpoenaed by the government, we can't hand anything over; it's all on your machine.

Correct — today that's true. None of the recordings ever leave. Sometimes when you do a search with GPT-4, we include some of that context, but just the text of it. And even that, if it's all done locally, creates a much simpler model for people to understand. By the way, we gave a demo of this a few weeks ago: Intel launched their new Meteor Lake chip, called Intel Core Ultra, and I was lucky enough to be on stage with the CEO of Intel at their developer conference, demoing a local large language model running on the Intel chip with the Wi-Fi off.

Well, first of all, that part of our product is optional. It's in the cloud, it's a nice-to-have — it's not the core functionality. Right now you can still browse back and forth in time, look at transcripts, and search those transcripts. And we're lucky that Sam Altman was our first investor, and we've had other co-founders of OpenAI invest as well, so that helps a little. But at the end of the day, what we're focused on is solving the problems for our customers and users. Today, OpenAI and GPT-4 are fantastic. My guess is it's going to be sooner than you think that you'll be able to do this all locally — which then means you can run it without an internet connection.

There are two ways to look at it. Going after a specific use case — described in tech jargon as an "ideal customer profile," a super-specific "who" — is kind of easy mode for startups. You have focus, clarity, alignment, decisiveness. Marketing is easier, product is easier. But you also have a much narrower market. Some tools do that, and they're fine with it. Others view themselves more as utilities, as a product that can help anyone — and it's harder, I'll admit it's readily harder. It's harder to decide what to prioritize; you get pulled in lots of different directions; the marketing message on your homepage now needs to appeal to a broader audience, and when you do that, you sometimes lose the edge you'd have if you had the perfect tagline for a doctor.

We've tried to come up with what we think is an ideal customer profile. We describe it as people who work at companies or organizations of 1 to 50 people, typically in the United States. This group retains really well for us. These are people who aren't at large companies where IT is going to be breathing down their necks. Right now the honest answer is we're getting pulled in so many directions, and it's hard to say no when we see a lot of market demand in lots of different places.

Yeah, it's definitely playing on hard mode, but it was by design. I'm lucky enough that Optimizely did well, so I could swing for the fences with this startup and do things a little harder that might have a bigger impact. Going after a broad market, and going to a consumer market — it's hard, but if you get it right, you can touch millions or billions of people much faster than you could with a traditional sales force knocking on doors. So part of it is the potential impact of our product.

I'd also be lying if I didn't say I really didn't love the enterprise B2B game. It's less about the quality of the product and more about the relationship — more about what I felt were the impure parts of building software. Steve Jobs has a great quote on this. Someone once asked him a similar question — why don't you go to the enterprise more heavily? He thought about it for a second and said something to the effect of: he loves his current job, because when you sell to consumers, you wake up in the morning, you go to work, and the buyer of your product is the user of your product. Oftentimes in enterprise the buyer is "confused" — that's his word, his disparaging way of saying they're solving for the wrong problems, their requirements aren't the actual things that matter to the user. I didn't want to be in a business building for confused buyers.

Our product has created so much more value in my life than the subscription cost. So it's playing on hard mode — but I'd much rather play on hard mode and win than play on easy mode and win.

I've evolved on this quite a bit in my career. When I started I was still very young. My company grew very quickly at Optimizely, and I was hiring people twice my age with twice the experience. I wasn't really able to deal with or understand conflict — I was conflict-averse.

So I had to rewire myself and overcome a lot of childhood trauma to really understand how to hold people accountable in a productive way. Once I learned those skills, I changed a lot as a leader. When I look at the theme of all the things I learned from my last company — 12 years, $120 million in annual revenue — all the things I regret most fall into one pattern: I wanted to do A, my gut said we should do A, somebody else — a great executive I'd hired, a board member, whatever — thought we should do B, and we ended up doing B, and it turned out poorly. There are plenty of times we did A and it turned out poorly, and I don't remember any of them.

I don't think you need to be a jerk, but you do need to be able to stand up for what you believe, and you need to be good at holding people accountable. There's a Machiavelli quote about needing a leader who's both loved and feared at the same time. So I think that's what you need — a combination of both.

I think this is very true of a lot of second-time founders. I certainly felt it a little before we launched Rewind the very first time, almost a year ago. I feel it less now because we've got traction. Going from zero to one, from no one knowing what you're doing — there are a lot of expectations if you had a prior successful company. First of all, the next time you launch something, somebody pays attention. It's not a lot of people, but sometimes when you launch as a first-time founder, nobody cares, nobody pays attention, no matter how good or bad it is.

And you hold yourself to higher expectations — I know I do. I'm proud of Optimizely; I think it's still going, I just spoke at their customer conference. Ten years ago, the company that acquired them renamed itself Optimizely, so I'm proud of that. But in my heart, I want Rewind to have 10 or 100 or 1,000 times the impact. So those internal expectations play a huge role too, and they make second-time founders a little gun-shy to talk about what they're working on.

There are a couple of really foundational things. First of all, I've made a lot of mistakes, and I have a lot of screw-ups — I should probably add them to my LinkedIn profile, they're just not there. But the bigger lesson, and the thing I attribute any success I've had to, is approaching life as a learn-it-all, not a know-it-all. Treating every opportunity as an opportunity to get better, never resting on my laurels, never thinking I'm good enough — knowing I could always be better. That feedback loop, and being willing to take it publicly — to go on podcasts and get ripped a new one —

those are things I think are important for me personally, and good for our business. I try to do it as a manager, as a leader, as a father, as a husband. A lot of people get blocked and don't realize how defensive they get. It's very hard for somebody — especially for me, I'm 40 years old.

That's very hard for people to do, and I've done it time and time again. I used to be a meat-eating carnivore, and I became a devout, almost prophesying vegetarian. I used to drink probably more than was healthy, and I don't drink anymore. I used to play video games, and I don't do that anymore. I've learned these things about myself, my body, and how I interact with my business. I used to make up for my lack of prioritization and focus by just working a lot of hours. Now I have three young kids — a five-year-old, a three-year-old, and a one-year-old — so I'm ruthless with my time. If I could take me as a CEO today and put me back ten years ago, I'd be 10x more productive in the hours I had, because I know how to focus on the things that matter and ignore the things that don't. So that would be my biggest lesson: be a learn-it-all, not a know-it-all.

It wasn't a switch, no. I've been very lucky to have interacted with people in the world who were models of this — I got to see what it looks like.

Being a learn-it-all is what people who are truly confident do. It's the people who are maybe a little insecure, who don't actually know what they're doing, who pretend they do. So in some ways, maybe that's the gift I've had: I've been lucky to have this confidence in the roles I've been in.

People put a facade in front of themselves that blocks them, and the people around them, from getting better. So I can't think of a particular time. I know there are certain people in my past who've been good models of this, and that's probably helped a lot.

My thinking on this has evolved. First of all, it was never by design. It was never "let's put something controversial out there because it'll go viral." It was an earnestness, a belief that what we're doing is good, and a recognition that it's different. Anytime you do anything different that changes people's perspective, they raise their eyebrows, they're cynics — so that was inevitable. But it wasn't manipulative. It was just an earnest, "I actually think this is a good idea, and it wasn't possible before." Just like airplanes sounded scary, and self-driving cars sound scary, this sounds scary — but eventually you realize, wow, it's not all that bad, and there's so much utility. "How can I live without it?"

So I recognized it from the beginning. At first it kind of hurt my feelings, people describing what we're doing as a dystopian sci-fi nightmare. Eventually I came to a level of acceptance when I realized that for every person quote-tweeting our post saying "this is a Black Mirror episode," half of their followers would actually be like, "hell no, this is actually super cool, I'm going to sign up." When I saw that, I came to peace with it — you can't make everyone happy, you're going to have haters. The good news is our haters are willing to spread the gospel of what we're doing, and half of their friends and followers actually think it's a good idea. So in hindsight, it actually helps us. If we were more middle-of-the-road, more vanilla, more familiar, people probably wouldn't talk about it as much. So I wouldn't say you should go out on a mission to create controversy — but if it comes, history shows time and time again that any new technology has that pattern. So I hope we follow suit, and build technology just as amazing and breakthrough as the phone.