Cliff Weitzman: What I Learned from 100 of the World’s Top CEOs & Why Tokens Will Outspend Salaries

Don't even bother spending money on any platform that's not Meta until you reach $100,000 a month and spend on Meta.

We test almost a thousand AI-generated ads a day. Right now, we're getting to the point where soon we're going to spend more in tokens than we spend on actual salaries. If you don't spend a thousand credits a day, I'm disappointed in you. If you can really trust someone, you can afford to pay them significantly more. The world around us was built by people no smarter than you are. You could do things ready to go.

My pleasure. It's great to be here.

Yeah. So, I moved to the US when I was 13. I didn't speak English. And my parents knew about three schools: Harvard, MIT, and Stanford. And we had a college counselor in my high school, and she recommended applying to six schools. I was like, 'Okay.' So, I started filling out the applications. And quickly I realized that even for kids with perfect SAT scores and perfect grades, you have about a one in 10 chance of getting in. And I was like, 'Okay, it's a lottery. How do you win a lottery? You buy more tickets than everybody else.' And so, most kids in my school didn't need to do seventh period when they were a senior. But I had academic workshop, which is basically the class you go to if you're on the short bus. And so I went to academic workshop every single day because I just couldn't read very well. And I just made it my goal, my job, instead of leaving school, I'd go back to academic workshop and I would apply to colleges. And when I finished the normal application, I just kept going. So my main essay, instead of making three variations or five variations, I wrote 48 drafts. And then I had to make a second Common App account because I ended up applying not to 20 schools, which is the max, but 26. And in the end, I didn't get into many schools that I wanted, but I got into a lot of schools I didn't expect to get into. And so my overall principle in life is volume of work: more reps, more shots on goal, more free throws, and that's how you get on the bell curve unexpected outcomes.

Yes. So I am super dyslexic. First, third, fourth grade, could not figure out how to read, and I wanted to read fast, but I couldn't read fast. So we built a deep learning-based text-to-speech model in 2015 that now reads more than 10 million books' worth of words to people every year.

So you're a great guy.

Great guy.

Yeah. So, when I was 14 years old, freshman year of high school, I was very self-conscious about the fact that I was very bad at spelling. And you get a yearbook, and typically people will sign the back of the yearbook and write some nice message about you. And I didn't want to misspell stuff in everybody's yearbooks. So, I came up with a strategy. I would write the least number of words possible, and I would write the same thing in everyone's yearbooks: 'David, you are a great guy. Steve, you are a great guy. Bill, you are a great guy.' And then after the last day of school, we went to a sleepover at one of my friends' houses. He goes upstairs to show his mom's yearbook, and I hear him scream from the top of the stairs, 'WHAT THE HELL, CLIFF? I'M NOT GAY!' And I'm like, 'Oh my god, what have I done?' The crazy thing is, eight years later, I was graduating college, and I wrote a 30-page paper for one of my philosophy classes, and I submitted it before fully submitting it to a friend. They sent it back to me, and I had three spelling mistakes, and I cried because if you had told 14-year-old Cliff that he'd go to college, he wouldn't be sure about it. Definitely not expect himself to go to Brown, but for sure would never expect to be able to write a 30-page paper with three spelling mistakes. And the reason why I didn't have that many is because I can hear the spelling mistakes with Speechify.

Sure.

So the first, basically at every point in Speechify's history, I have led every department except for AI. And so design, engineering, QA, growth. And my rule is: I will read 100 books about the topic, listen to them with Speechify. I will identify 100 experts, and I'll go and have a conversation with them. So the first thing that I learned is if you send a good email, people will respond no matter how amazing they are. So I did Mike Krieger from Instagram. I did Ev Williams from Twitter. I did the founders of Plaid, of 23andMe, of Honey, of Grammarly, Vlad from Robinhood, and all those became Speechify investors very early on. But they would all respond to the emails. And the second rule was again volume of work. If the CEO didn't respond, I would message the CMO. And if the CMO didn't respond, I would message the head of growth. And if they didn't respond to my email, I would politely continue doing so until they either told me to stop, which is less than 1% of the time, or they agreed to go on chat. And I would also message them on Instagram. And I would also ask people for intros, just like be unlimited hungry to get the information. And the second thing I learned is once we would get on a Zoom call, we'd chat, and there was one thing that always succeeded, which is I have listened to 100 books a year, every year for the last 15 years. So I'd ask them, 'What's your favorite books?' And they'd tell me, and most likely I read at least one of them. So we'd connect on the book, and then I'd be like, 'Oh my gosh, I'm going to be in Denmark this Sunday. Do you want to hang out?' If they were in Denmark, and they'd be like, 'Oh, sure.' And then I'd book a flight to Denmark, and I'd show up. And so I did this with Anton at Lovable recently, and like it just, I'm still doing the same thing. And after I did this with the heads of growth, I did this with CTOs, and I did this with every type of person that I want to learn from. And then I'd go and I'd watch how they bought Facebook ads, or I'd watch how they made content. And from that, I figured out what would work. And so the first thing that I learned is growth is just an arbitrage game. You are competing with every single other person in the world who wants to get their product in front of users. And that means that your lever points are either the content that you make or how you distribute that content. When it comes to buying ads, there's always new tricks that you can come. At the time, adding polls on your Instagram story was really helpful. Right now, whitelisting ads is really helpful. Volume of work of like posting a ton of ads that you test is really, really important. There's different rules for SEO. There's different rules for how you structure your team as it comes to who is the creative director, who writes the scripts, how do you do the editing, etc. So every team does their own thing, but for every company there's a different rule. So one person I learned from was the founder of Blinkist. A small thing he said, like as an aside, that really helped me at the time was: don't even bother spending money on any platform that's not Meta until you reach $100,000 a month and spend on Meta. And then I stopped spending on other places. And then the thing I also realized is that a lot of the people who were senior in these companies were already rusty. They were not very strong individual practitioners of the craft. And so I would go down a couple of levels to find the person who's actually buying the ads, who's actually editing the content, who's actually seeing what the CTA is and what the things are. And I made sure that everybody in our company, no matter how senior they got, always were still not just a fat general sitting in the back, but a warrior who could take out their sword and actually hands-on keyboard buying the ads. And we had a situation where we hired a head of growth. 30 days in, nothing. 60 days in, nothing. And I was like, 'All right, man. What's going on?' And he's like, 'Well, we're trying to hire the people.' I was like, 'Okay, look. Well, okay. Tell me what you're doing for hiring.' 'Well, I gave the head of HR like the brief of what I'm looking for.' Bro, source the people yourself. Message the people yourself. Get on the calls yourself. You've got seven days. In those seven days, I need to see you making ads in CapCut. I need to see you editing ads in Figma or Photoshop, whatever tool you want. And I need to see you buying the ads. And if that doesn't happen, it's going to be very difficult to keep working together.

We look for three things when we hire people: fire in the belly for the product, high loyalty to the team, and the ability to learn fast and ship to production. And we have a principle even inside of engineering, which is you got to respond to messages within 60 seconds because you want to unblock people and just inform them if you're able to unblock them or not. If you can't, that's fine. They're going to go learn it on their own. And so, we have a rule: if you ship a PR to a repo that your team is not responsible for, it gets fast-tracked, and that team has to very quickly respond because I want to encourage people to learn how to contribute to other codebases. So, the person in the company who learns the fastest is my brother Tyler, who leads our AI team. And just the whole principle inside of our AI engineering team is just learn everything, and then you win.

So, companies go through cutting and bulking cycles like bodybuilders do. And there's a reason why cutting and bulking makes sense because you can't oscillate between the two in one week. You have to commit for like six months to doing one. And so for Speechify, for example, it took us four and a half years to find real product-market fit. The product stays exactly the same. The vision today is very much the same as it was in the very beginning. But in 2020, we started growing really fast. And that was around the time where install to trial, trial to paid, renewals all went up. People used the product a ton. And I tried to figure out how do I pour fuel on the flame? And the reason why companies succeed is you have a person who puts their beating heart into the product and their will to power. That's the thing that makes the things successful. And you can only do one of those things at a time. The second you split your focus, even fundraising, you can't fundraise and do sales at the same time. You got to do one. And so all I was thinking about at the time was growth, which is why it was a no-brainer to fly to all these places to meet people. And there was a point where I was actually very annoyed because I didn't succeed in having the growth outcomes that I wanted. And so I started logging my time in 30-minute increments. So every day I'd have 48 units. And I noticed that 30% of my time I spent helping other people. And 30% of my time I spent learning SEO. So I decided I'm not allowed to learn anything from that point on. I'm only allowed to execute. And like the second I saw myself reading an article, I would close the article and go do. And so the rule is you just have to go and you have to learn and you have to execute. So you have to have that maniacal focus on the one thing. And then we grew a lot. And now there was a focus on, okay, well, we got to get the margins to be good. And so that's a cutting year, and there you're focused on profits. And the conclusion that I've had across the almost decade I've spent working on Speechify is: any Harvard MBA can cut costs. If you're smart, you could do it. It takes a genius to grow revenue. And that's why investors value exponential growth of revenue even for companies that are not making a profit. But it's very difficult to do both at the same time. You're clicking the gas and clicking the brake at the same time. And so Speechify just finished four and a half years of running profitable. And now we're in a hyper-growth phase. And so it doesn't matter whether you're a company or you're an athlete or you're a social media creator. You have bulking years and you have cutting years. And you need to decide which year it is for you.

No. Not at all. So if you think about CAC, you have to consider, okay, what is my LTV and what's the first payment period? So, we like to think about first payment period. So, our product is $140, but if you're listening to this show, hit me up, [email protected]. I'll give you a discount. I'm never willing to burn CAC for no reason. The blended CAC can go up, not the direct CAC. What does that mean? I'm okay spending $500,000 a month making 20,000 pieces of creative because that is a creative investment. I'm going to learn a format that is new that will unlock additional growth later. I'm okay doing an experiment on AppLovin or a specific new product that Google dropped for YouTube videos. We're one of the 200 companies that's provisioned to test ads on OpenAI. So, we're spending a ton of money on OpenAI right now, and we're learning it faster than everybody else. So by the time it rolls out to everybody else, we'll already be really, really good at running ads on OpenAI. So you grow, you grow, you grow, you test new channels, you test new creative, and I'm very happy to burn an unlimited amount of money on excellent engineers. I'm not okay just like setting money on fire doing ads that are not attributable, right? Spending money on a billboard where I don't know if a user comes. I think about everything as a test, and if the test succeeds, then I do more. If the test fails, that's fine. The classic adage that you learn from CMOs. I learned this from Kelly at Netflix. They spent $1.7 billion on performance marketing in 2017. And you'll hear Brex also is really good at this. You know, 'I know I'm wasting half my spend. I just don't know which half.' And so that's okay. Do that, and then do the work to figure out which half you're spending. But don't just spend with increased CAC for the sake of growth. It always has to be an experiment you learn something from.

Yeah. Greg had Ladder. So, Ladder is a health and fitness company. Perfect example. He's the CEO of the company, but he buys the ads himself, and he found a lot of arbitrage plays that other people have not found. And so, you just got to go and figure out what those plays are.

Every single day. Every single day.

Whitelisting ads on Instagram works really well.

So, let's say you find a creator that is in your niche. For example, for us, maybe it's teachers, maybe it's someone with ADHD or dyslexia, maybe it's an AI creator. They make a video for you, and then you go, 'Hey, make me 20 videos. I don't need you to post them on your profile. I'll run them as ads myself.' And actually, if some of them are good, we can post them from your account, and we increase the spend. And so users might not see it as a normal organic video, but you get to test all of those videos, see which one performs, and then you have them post organically the one that actually is successful.

Not at all. Because you're marketing for different demographics. So for example, you'll find this interesting. We test almost a thousand AI-generated ads a day right now on top of the roughly 8,000 organic creatives we make with humans every single month. And one of the things we do is we reskin. So my sister Geffen leads our ads team, and we have a bunch of great ads with Geffen, and then we'll turn her into a 60-year-old lady, and we'll turn her into a 40-year-old man, and then we'll turn her into someone who's African-American or someone who's Asian, and we'll change the background from a coffee shop to a library to like a... I was going to say farm, but farm wouldn't convert for our users.

And so I know that for us, highest converting demographics would be people in their 30s and 40s in terms of paid, but we have like crazy usage in the 18 to 28-year-old market. But people who are 60 and above need reading glasses. So they use Speechify a ton, and now we're launching voice agents at symbolvoices.ai. And so that's a different demographic. And so all the initial ads for Speechify were me. And I would sit in my house, especially Saturday or Sunday, with my phone, and I'd record. I'd be like, 'As a medical student, I use Speechify to read while I'm working out. As a lawyer, I love working from home.' And, you know, you just need to read that contract and find that one gotcha. And Speechify can find that for you. You just need to listen asynchronously. And I'll be like, you know, 'I have dyslexia and blah blah.' I would just play all the characters, and it was my face. And it converted really, really well. But I got maxed out. And so when Geffen joined and we hired other people to the team, suddenly we found that actually women convert a lot better than men as like audiences. And so right now I would say we have more women who use the product than men by like a good margin at the moment. It's just because our team happens to be very woman-focused. And you just have to test everything on the video.

Yeah, because it's an arbitrage game. If you use the tool that everybody else uses, typically you're not going to win. So we just built our own platform. So there was a point several months ago where my girlfriend went to Disneyland. So I had four days free. And so in that time I went to school, Alex Hormozi, shout out. And I did a course for AI-generated video ads, and then I hired everybody from inside of the community that was really good. And so I built a team. I hired four new engineers outside of our core team. Like in one week I hired 10 people. And 24/7 all I was thinking about was AI-generated content. And my goal was to get to 1,300 ads that we tested every single day. And we just built tools for it. Initially I built it on Nand, and then all the Nand instances would time out. So then we built our own platform and our own website that does really, really good AI-generated stuff that allows you to reskin it, automatically post it to Meta and TikTok and YouTube and all the platforms, and it gives me all the data that I want. I'm obsessed with Manis, so I'd get like reporting for Manis every single day. And then after those four days, I would continue to monitor it, but I assigned a leader for that project, and that project has done really, really well. So we don't use any external tool. Obviously we'll use all the LLMs, but that changes on a weekly basis, right? One was really good, and then Kling was really good, and then Sora was really good, and now ChatGPT's new image model, and then a banana obviously. You just constantly test and iterate and iterate, and you find flows that work. So we just built our own.

That's wild.

Yeah. So, what you want to do is you post, you throw a lot of things on the wall. No matter how smart you are, you're not going to know ahead of time which thing is going to convert.

You don't. You have no idea. So, first you test. Once you test, you look for the following. In Meta, you can see the click-through rate on the ads, and you can see the CPA, cost per action, and you can learn what the CPM, cost per thousand impressions, are. So, if we see that the cost of acquiring a user and having it paid is less than a certain amount, we're like, 'Oh, very interesting. This is like a standard deviation better than everything else. Let me put it into the main campaign.' Now, it's going to compete against our best-performing ads ever. And we let it breathe. And if it does well, it gets more spend. And if it does less well, it gets less spend. And we cut it. And so you just, it's like evolution. You put it into the wild and you see if it performs. So every day you'll take the top 10. You'll take the top... Every day we upload 1,300 to our testing campaigns. We'll take of those the ones that win. It might be that nothing won that day, in which case nothing gets launched. It might be that a lot won that day, in which case they all get launched. And then you let them. It's like March Madness brackets. Like you just graduated to the next level of competition where you can get more spend. It's almost like duplicating the algorithm that TikTok and Instagram have anyway, where when you post a video, a thousand people see it, and then if it does well, 10,000 people see it. And if it does well, 100,000 people will see it. If it does well, 10 million people will see it. It's the same thing with performance marketing ads, except you're paying for every view.

A lot.

Wow.

So, when GPT-3 came out, actually even GPT-2, my youngest brother Eric has neural inflammation and psychosis, and he's had it for a long time. It causes OCD, causes paranoia. Me, my brother, and my sister got really good at using GPT-3 for medical stuff, and when Chat came out, we started using it even more. My main use for Manis is medical things. So I build on average a website or two a day. And so for example, I'll make a document of all the neurologists and syphilis or Lyme doctors in the world, and then I'll auto-message all of them. I'll upload all of Eric's medical records, and then I'll build an interactive graph of them. I might go and look at a company that's similar for us. So, Speechify is basically a B2C version of ElevenLabs. I'll go and find every article, every podcast, transcribe it, and map out their entire business, and like I'll understand how their business works. Or I'll go, 'Hey, we're selling it to education. Great.' I'll make a website with a map of all the United States, every single school district in the United States, who is the district administrator for that school, and make it into a website, and I'll send it to our B2B team. Why do I do it this way? Number one, it makes it easy for me to understand things. Number two, I can then plug that into another LLM, and it has access to the entire website that it can scrape. So, I'll then put that into Claude, or I'll put that into GPT, or I'll put that into Gemini. So, that makes it really, really easy. Manis makes it really easy to post automatically for you. It's really great at deep research. So, for example, my girlfriend and I went to a Passover Seder the other day, and she didn't know too much about Judaism. So, I wrote a 120-page document on the history of Judaism for her, and that way she didn't get bored at Seder because she could listen to that with Speechify, and like she just listened to it in 25 minutes in the car on the way there. And so I'm just constantly making things, and the thing is it's very easy to use from my phone. It's very easy to use from my computer. Like I use Claude Code all the time and dispatch from Claude as well. Manis is just like a little bit faster, a little bit easier to use. And so Jarvis, which is the Speechify B2C voice agent, is essentially like Manis but voice-first. And so that's the places where I like Manis more than I like, you know, Claude Code or Gemini.

Sure. So, rule of 100. I had gone and I had talked to all these people who were good at growth, and then I made a list of the top 100 creators, and I talked to them, and I'd live in their houses, and then I was making those ads in my apartment, and some of them were good, but I was stuck in a local maxima, and I wanted to break out to a global maxima. And my conclusion is the way that I would do that is I made a list of the top 100 best-performing ads in history. Every Super Bowl ad that was really funny, every ad that made me cry, every ad that won an award. I made a list of maybe 500 ads, and then I whittled them down to the top hundred, and then I read the script for each one of them, and I rewrote the script to be about Speechify. And then what I would do is I would up the ante on the ad. So the rule in videos typically is you win or lose by the first three seconds. Do you have good retention in the first three seconds? But the rule I developed is actually you win or lose by the first image of the video, like the first 1080 pixels. And so I wanted the first 1080 pixels to be as crazy as possible. So the way to represent Speechify is headphones. So I got bright red headphones. I put them on my head. I put a suit on, and then I went into a hot tub. And so the first shot is me with red headphones and a suit in a hot tub. At the time I had long hair. And I just threw behind me like the most random knickknacks, like weird photos and books and stuff like that. And I developed this character called Cruz Silver. Cruz Silver appears in many of our ads, and he's a version of me, and he's very gullible. He's kind of dull, but he has great confidence. And so the ad starts, 'Hi, I'm Cliff, founder of Speechify. What is Speechify? Well, you give us stuff you need to read, and we read it to you instantly. Yeah. Instantly. Are the voices any good? No. Our voices are [__] great.' And then it just goes into all these different scenes. And the video was really long. And so then I edited it, and I made like my college apps almost 50 variations of that video. And I tested all of them. And one did pretty good, and another did better, and 20 didn't perform at all, and one did amazing. And that YouTube ad basically launched the company into a whole new stratosphere and performed for like three years straight. That video got like hundreds of millions of impressions on Instagram alone. And then every year I remake that video. So again, I did the nacho commercial from the Super Bowl. I did the Isaiah Old Spice commercial. I literally got a horse for the ad, and just you just take it as far as you can.

Which one?

It's a complete random distribution of what works and what doesn't work.

I thought that the Old Spice commercial would absolutely annihilate. It got like 300,000 views. The one that I did not expect to work that worked amazing is I wrote a rap song about my experience with dyslexia and ADHD, and that thing made us like three mill... also like $3 million on a rap song.

Oh, of course that works amazing. So, formats are the thing that works, right? So, one of the things that works really well for MrBeast is ladders, right? $1 hotel, $100 hotel, $1,000 hotel, million-dollar hotel, competing where you have people age 1 to 100, like all competing against each other in the same thing, like 'I got trapped on an island,' blah blah blah. He does the same formats but changes the view and how it works. So, American founder versus British founder is a perfect format. And then you switch it to French versus Ghanaian, whatever you want. And then, of course, repurposing works. You take the same piece of content and you put it on TikTok, you put it on Instagram, you put it on YouTube, you change the hook. Especially in performance marketing, it doesn't matter how many times you post because you're not going to damage your brand by posting something that will not perform organically because it's behind the wall anyway. It only gets served. It lives in the shadows. So, just test 800 variations. Why not? Do you give a [__] about conversion? And I know that's...

So because we have TikTok and we get lots of views. Conversion back to core podcast and core product very little. Would you say then not worth it? Don't bother.

Fascinating.

Only do things that result in conversion. So we have more than 50 million people who use Speechify now, the organic B2C product that makes it really easy to read. And we like, if you don't download the app and you don't start a trial and you don't convert to paid, there's two options. The first one is I don't care. The second one is I actually do genuinely care that we solve dyslexia and ADHD. And so there's two billion people in the world that have either ADHD, low vision, autism, anxiety, concussion, second-language learners. Our philosophy is we roll out the red carpet for the user. We give them the best experience possible. If you can't pay, that's fine. You still have a great experience. And so I think about that as that's our form of charity. Like we acquire the user and they can use the product for free. Amazing. But that's not what's going to cover the ad spend. So the ad spend is covered by the people who pay. If you don't get people who actually convert on your product, there's no point in doing the work that you're doing.

It's all in one. Yeah. Yeah. It's not split, and there it's like there's sub-teams inside of growth. But they all report into the same people.

It's still word of mouth is like the biggest way of us getting people, right? It's someone telling their friend, 'Hey, I know you like podcasts and audiobooks, and I can see that you're reading at like a normal reading speed. Like don't you want to read 3x faster? Like use Speechify to make AI-generated audio books.' That's the biggest one. We crush it on Apple. So we were like, we won the design award last year. We got featured by them like 14 times last year. SEO is huge for us, 10%, 15% now. 15% of our users come organically from ChatGPT. Inside of performance marketing, Meta is still the best winner. Google is the most profitable. YouTube is pretty good, especially on desktop. AppLovin is interesting. TikTok is interesting, but they're not as big. And you have this great halo effect. So the more that you get the message out there in different ways, the more you sow the seeds that results in people talking about it and telling each other about it.

You just got to test like everything else.

Of course. Yeah. Massive. Massive. Massive. Because it knows so much about you. It's the same thing as Meta, right? We talked about this earlier. The reason why Meta ads are so good is because they know what you message on Instagram and what you message on WhatsApp. And as a result, they can target really, really well, so people convert. OpenAI knows everything about your history and what you're interested in inside of your psyche. And it's exactly the same way that Google has built such an amazing business because there's relatively high intent when you talk to it. And the ads convert. Now, it's extremely high CPMs. But if you find content that converts, the CPMs don't matter. And that's something that people really don't realize. Like it's okay to pay a high CPM as long as people convert and as long as you have attribution. So they just launched the SDK for tracking. Like that's really important.

Absolutely. Yeah. I think Meta is one of the most underrated stocks right now.

Number one, Mark Zuckerberg is an absolute savage. Like he will hire people if they're right, no matter what. Number two, I think that in a world where engineering is commoditized and design for the most part is commoditized, the most important thing becomes excellent QA. We can talk about that later, and customer acquisition. Google has pressure on it from all the LLMs, but Meta doesn't. The only real pressure on Meta now is TikTok. And TikTok has a lot of issues now that it's been separated, yada yada. So, and Zuck also has proven that he is the number one acquirer in the world. Like no one acquires as well as Meta. The engine, the leadership team is incredible. Like Clegg is amazing. Alex Schultz is amazing. And I think that also in AI world, data is everything. So they have more proprietary data than everybody else. They haven't figured out the model side yet, but I really do believe that they will. And I look at all the people who are vibe coding new products right now. Where do you think they're going to get their users?

No. No, it's not [__]. Everything is in taste as well. When I'm saying design, I mean the actual grind of sitting in Photoshop, sitting in Figma, sitting in Illustrator, sitting in Canva and pixel-perfect making things. So, we're already at the state that most of my leadership team is sending their emails from inside of IDE like Claude Code terminals. Definitely like design when we like want to launch a new product, we do not assign a designer in the beginning. Like we say just use the default Claude Code stuff, and then we'll bring designers after, and they'll make it a lot better. And we definitely do use design for user experience interviews and stuff like that. Like that matters a lot. Like design is a really important part of Speechify. But the hammer-and-chisel part of design is not as important, and therefore the thing that's left is taste. Like very few people have taste. I have taste in some things, but in other things I have bad taste. So, which is why we hire really great designers at Speechify to make up for my shortcomings. And so, no, taste really, really matters. Even if it's just telling you pick that one, not that one.

Hold. Dylan Field is a savage. Anything Dylan Field does, you should go for.

I do. I also think that Figma is a relatively anti-fragile company, just because it's a really good team. They're excellent at sales. They have really good taste. Figma itself is a great product because Dylan coded the original thing. I think it was in C++. It was like the first product that would render vector-based architecture in a browser. They can do the same thing. And today what happens is you have a world where everything is moving really fast. Do I bet on an old incumbent that's really bad at innovating, or do I bet on the Figma team? I would bet on the Figma team. So it might be the product that they end up with is not the product that they have today, but I think it's a great company. I think they actually benefited by being a publicly traded company, and I think that they'll continue to rise and do really great. Remember, Figma traded at 40x revenue, and now they're at 6x revenue. So, it's less so that it's not a good company. It's more so that the market is making a correction, and that's fine.

Claude Code. So, we were huge adopters of Cursor about two years ago. I think I got the Anthropic team plan in December. And it actually drives me crazy that not everybody has moved over to Claude Code. A lot of people will use Claude Code from inside the terminal in Cursor. I prefer if they use Claude Code inside of the Claude Code IDE. I think it's more powerful, but it's fine to use it inside a Cursor as well. But just most important is you have to use it to the max. And I'm always surprised like there's some limitations that you need to enable. It took people like three months to ask for some of the features to be enabled. I was like, why did it take you three months to ask me about this? Like I'm pushing people constantly to hit limits. And we're spending more in... we're getting to the point where soon we're going to spend more in tokens than we spend on actual salaries.

Like next year I expect we'll spend more in tokens than we'll spend on salaries.

It's atypical right now, but I don't think it'll be atypical in the long term.

Get there in three years? Yeah. You just need to be like a really strong early adopter.

No, across engineering. Across engineering. I would like to be spending more tokens on our growth team and creative. We're not there yet. We need a lot more testing mechanisms to be put in place for that to make sense.

What?

Yeah, it's very easy for you to change that already.

So I can tell you what we do. So remember like Speechify is a voice agents ecosystem with 50 million users and 1.1 million five-star reviews across platforms. We need to educate even our users on how to use AI. And so our first agent was a text-to-speech tool that reads to you. The second one was a speech-to-text tool that does dictation. The third one was a podcast generator where you can be in the podcast as well as two new hosts. Then we launched a teacher agent that will ask you questions about the document and then teach you back the content gap that you were missing. And now we just launched Jarvis, which is the Speechify for work. It like lets you do productivity. And then the enterprise, symbolvoices.ai. We're constantly educating our own users on how to use AI. And what we found is it takes about a hundred iterations on a feature before the feature gets adopted. But once it gets adopted, it gets adopted really well. So you need to do the same education for your B2C users as you need to do internally. Internally, I am constantly screaming from the rooftops that people have to use Claude Code. And I'm more extreme about it than it needs to be because I need to move people from all the way over here on the right to all the way over here on the left. And I'm okay if they meet me in the middle. So, I'm just extreme about it. I'm like, if you don't spend a thousand credits a day, I'm disappointed in you. Like, I need to see that happen. And so, send me screenshots, and I will get people on a Zoom call, screen share the entire engineering, and I'll show them how I use it. And they'll go, 'Tim, you're really good at this. Show us what you do. Tyler, you're amazing at this. Show us what you do. Simon, show us what you do.' And we're just constantly demoing. And you just got to make sure that people are aware of how much other people are using it. And then I'm like, if you are not using this tool right now, like it's like a chat in everyone's Slack. Comment explain why. It's going to be very difficult to continue working together if you don't use it and you don't explain to me why. So they have to explain to me why, and then I will call them and convince them to use it. And I'm like, you have 24 hours. Send me a Loom video with what you used AI to make.

So the personal side is medical stuff. We talked a little about my brother Eric, who has Lyme, neural inflammation, and psychosis. We did proteomics, mapped all of his genome, put it on Supabase, wrote an MCP for it, and then had it compare to every single article that had ever been published about anomalies in the genome. And we found four anomalies in how his genes expressed. And so that was like an amazing breakthrough for us. The second thing that we did is my dad has had prostate cancer on and off for the last eight years. There was a point where his PSA continued to go up, which meant it came back. We just didn't know where it was in the body. And so the next thing you need to do is what's called a PSMA scan. And the doctor refused to do it because he was like, 'Listen, the cancer is too small. What we need to do is sit and wait.' And I was like, 'That sounds like the worst advice ever. If we know he has cancer, waiting for another three years is a terrible thing to do.' And he's like, 'Well, we don't know where it is.' And so what I started to do is I went on Kaggle, which is the largest online competition for data scientists, and I started hiring people on Kaggle to help me diagnose my dad's cancer from the radiology reports. And it has a lot of open-source repositories on data for what's called time series of PSMA PET scans for cancer. And so I came up with four theories. The first one is...

called tipping theory, which is like you basically look at the level above and the level below where you think the cancer is, and you make the data really noisy and you see if you can figure out where it is. So we started making models for that, and then I basically took five different approaches inside of this research from talking to Gemini and talking to Grok and talking to PD. I realized that there was a machine that could help us. So in UCSF there's a thing called the China Basin that does a normal PSMA PET scan, and it will scan a voxel of 4 millimeters in diameter, but there's another machine called the U Explorer that can scan a voxel of 2 millimeters in diameter. And there were three of them in the United States, and one of them was in UC Davis. And so I messaged the guy who was responsible for that lab. He agreed to test my dad, and the doctor who I put a lot of pressure on to give my dad the exam had scheduled it for the end of January. In December, we got to get the scan in this higher quality machine, and we found it. It was in those left seminal vesicle. I would have never had the confidence to push on the doctor so hard to get the scan if I didn't have LLMs open during the conversation with the doctor. I definitely would not have found another machine, got my dad scanned, and then figured out. Right. So by January 7th, he had the left seminal vesicle excised. And so before he was even supposed to have the second scan, he had a surgery. It was good.

So three years into my dad having prostate cancer, we had a meeting with a doctor who was very renowned, and he was like, 'Yeah, like your dad's fine. We're just going to sit and monitor.' And I was like, 'Screw that.' Like that's the same thing as us being like a JV soccer team. No, I want to be coached like I'm Real Madrid or Barcelona. Like give me, like tell me exactly what to do. Do you want me to like wake up at midnight and like see the moon or put my feet on like whatever you want me to do, I will do that thing. Like I just want my dad to live longer. And the problem is not the doctors, the problem is the system where doctors have to treat statistics and they need to make sure they don't get sued. And so in your case, in my case, because we're educating ourselves and we're willing to take the risk ourselves, we can do things that result in better outcomes, but they're a little bit more risky, and that's fine. And so my contention is you should just learn as much as possible because the information is accessible to you, and then you are the quarterback of your own medical experience or that of your family members.

Yeah. So people talk a lot about IQ, intelligent quotient, or EQ, emotional quotient, but the most important one is AQ, adversity quotient. How good are you when things get really tough? And so I think about it, think of through a jeep that's going through the savannah. I don't want the jeep that goes the fastest. I want the jeep that's not going to have a flat tire. They're going to stick with it the longest. And so there's a heuristic I use for this, which is sometimes you look at an engineer and they're willing to grapple with a really hard problem for eight hours straight. Most people, if it's really hard, they quit after 30 minutes, but the things that really move the company are on the other side of five hours of grappling with a hard problem. And you kind of have to be obsessed to do that. And so the way that I find that is I look for projects that people actually did that they published to production. Let's say you went and you spent four years learning computer science, and we have a call together. My goal in that call is not to not hire you. Nothing will make me happier than hiring you. I just need to get you to give me the information that proves to me that I can underwrite hiring you. So I just, the whole point of that call is for me to find out how awesome you are. And so I want to hear about your side projects that you've built. I want you to screen share and show me the things that you've made. But if you haven't made them, you're just not willing to deal with a little bit annoying thing that requires you to register an Apple developer account and put an app on the App Store. The little annoying things that require you to set up a Cloudflare instance and like make sure that your website like sticks around three years after you published it. Like I need to see that those things happened. And then ideally I want to see something novel that you created. And you don't need to be a genius. You just need to be willing to do the thing. And so the last thing I'll say is I have a friend, Valentine Perez, who has a great line, which is a little bit of slope makes up for a lot of y-intercept. It doesn't matter where you start. It matters how fast you grow, and the growth rate has to do directly with how much pain you're willing to bear.

There's a great question from YC, which is, 'What is a non-computer science system you've hacked to your advantage?' It's a really, really good one because I'm looking for the story that shows me your ability to a) build a system, b) learn something, and then c) have an inhuman amount of desire and will and grit. So that you have clearly. So I'm then looking for the rest.

Yeah. So, no, there's a lot. The first one is any situation where you can discover dishonesty. So what I really don't like is when people give me a stat and they say, 'Oh, I raised the stat XYZ.' And the more questions you ask, the more you realize they changed the definition of what the denominator was. So like that's one thing that I really dislike. The second thing is people who talk too much. Like the amount of signal per sentence that I get needs to be high, and if it's not, I don't want you wasting the time with the rest of the people on the team. Like we're all about execution here. And so sometimes I'll meet someone who I think is really talented. They might even be really technical, but they just talk so much, and I'm like, you're just going to annoy everybody else and get us off track. I want someone who's hyper-focused on the actually one goal, the thing that really matters. And then the last thing is you don't want people who are too comfortable. You want people who will be loyal. And so I always ask people, what made you interested to interview with Speechify? And if the answer is bad, we shouldn't hire that person because there's no differentiating reason why they're here, not at another company, and they might leave.

Terrible answer. Never would hire that person. Not in a million years.

Because someone else can come offer you more money, and then you leave, and after we spent like a year coaching you up and making you amazing.

'I have a kid with dyslexia. I have a niece with ADHD. I use the product myself every single day.' Those are great answers. You know, I read the blog post by Tyler Whitesman on how he skipped four and a half years of math in high school and built 47 iPhone apps when he was there, and then he went and did his co-term at Stanford, and I really liked his explanation of how he shipped the first deep learning model that worked on the CPU of an iPhone. So I'd love to work with him.' That's a great answer. Like just do a little bit of research about the company. I like, I want to see a reason why people are there.

So Michael Dell has this great analogy, which is when you're running a company, it's like you're driving a car, and you want the car to fit the driver. So you want the company to fit the person who's running the company. If you have dyslexia and you're nine years old and you suck at the one thing you're supposed to be good at, which is reading, there's two options. The first one is you think the world is stacked against you and school sucks, and you go down this really dark path. Forty percent of people who are incarcerated have dyslexia, even though 17% of the population has it. The second option is you keep trying, and you try again, and you try again, you try again, and you develop this belief that it's okay to fail, and that even in things that seem impossible, if you try enough, you'll succeed. So 40% of billionaires also have dyslexia.

Forty percent of billionaires. Yeah. And it ends up being the case that you coach yourself that when it's okay to fail. And that's exactly what startups are. You're stacking the odds against you. You're doing something that feels impossible, but you're developing this AQ to never give up. If you have ADHD, people with ADHD, they're definitely not slow, and they're not lazy, and they're not stupid. Their mind just works differently. And actually, we find that for Speechify users, if your brain works at 600 words per minute, but you only read at 200 words per minute, of course you'll get distracted. If you start listening at 600 words per minute, you love that. You stay engaged. Running a company requires you to be responsible for and understand the engineering, the design, the quality assurance, the go-to-market. All these different things are stimulating you at the same time. Being ADHD is actually very good to be a CEO. The problem is how do you get from like zero to one to ten? Like the ten to one hundred, amazing place to be ADHD, but you got to go from the zero to one hundred where you're doing all the IC work yourself. We need to stay focused, and that's really difficult. And so it's great to be dyslexic or ADHD as the person who is running a large company, and it's not great as the individual person who needs to build the company.

Coach. So one thing that we do a lot in Speechify is coach people on how to use AI better, force them to use AI better. So we're still hiring people at a crazy rate. Last year, we had 178,000 people who applied to work at the open engineering positions at Speechify. We have an asynchronous technical challenge. 19,800 people finished that challenge. So we're very good at hiring good people. Still, we much more focus on coaching the people who are inside of the company. And this is something that has been my battle internally, actually convincing other leaders to do because they have a lot of things that they need to do. It almost feels like a waste of time to go and spend time teaching other people. But you have to do it. You have to have to have to do it. I have a friend, Austin Ray at Ramp, and like his entire job is teaching the organization how to use AI tools. And we almost had Austin join Speechify. It's one of my biggest mistakes that we didn't make that happen.

We hired him as a consultant. We worked with him for a while. I had super high conviction on him. I couldn't get to the level of conviction I needed from other people. And I made a huge mistake because I didn't get a like resounding 'we have to hire him' from everybody else. I didn't do it. I should have trusted my gut and like done everything to give the sexiest offer possible because he was also like obviously a very expensive hire. So there's like a cost to what I... we should have just given him that because it was clear he was such Speechify DNA.

I have the same philosophy about hiring. If you can really trust someone, you can afford to pay them significantly more. And it's actually just hire the person for more money, and if they don't work out, you part ways. Like it's fine. You should always be trigger happy to hire really good people if you see that they're really good, and just ignore the fact that they're expensive. Just make it work.