Full Tutorial: How to Use Voice AI to Work and Code 10x Faster | Tanay Kothari

The average employee spends about 5 hours every single day typing. That's the part of the day that we first go in and capture. You can speak much larger and contextual prompts than you can type. What Whisper does is it tags all of the files that you want and also gets variable names right. So we actually don't look at other software companies for inspiration because most software companies are terrible at building habits. What I look at are games. The key metric that we have in our team is not ARR, it's not number of users that we track. The key thing that we track is how good of a job are we doing to build a habit of voice.

Wait, but you actually got the brain-to-text working? Like you don't have to even talk?

You do not have to even talk. You want me to show you a little sneak peek of it?

All right, welcome everyone. My guest today is Tanay Kothari, co-founder and CEO of Whisperflow. You know, Whisperflow is genuinely one of my favorite apps. I probably use it as much, if not more, than ChatGPT just because it lets me talk to AI. To get work done. And I'm really excited to get him to show us his advanced tips to get the most out of Whisperflow and also how he built such a great product. So welcome, sir.

Thanks a lot for having me, Peter. It's been great getting to know you over the last few weeks and super excited to do this with you.

Yeah, it's a, it's a great product, man. But, but I think you probably know a few more advanced tricks than I do. So maybe, maybe you can show our audience here how Whisperflow works.

Yeah, of course. So, TL;DR, Whisper, simple product, runs in the background, and it basically lets you use voice across anywhere on your computer. And the biggest thing that makes it different is it actually gets what you say and it fixes mistakes for you and structures things out, which no other voice tool does until today. So what I can do is I can actually share my screen and show a few examples of how power users of Whisper use it day in, day out. And I'll show a few different examples that highlight some of the best things about Whisper. So here's one where I'm writing emails. Really helpful because I have to do this about 100 times a day. Hey Peter, there's 3 things Whisper does really well. It's insanely fast, it's insanely accurate, and it gets names right. And what you see is it knows I'm in an email, so it structures things perfectly for me. It puts things in a bullet list. Small things like putting the colon here as well and knowing how to spell Peter's name because it grabs the context as well from what you're doing.

Yeah, this is great because normally I have to, like with other voice tools, I have to like manually edit it to make it, you know, in bullets, take some words and stuff. But here I can just send. So it's great.

Yeah, exactly. And the other good thing is you can also change your mind and it understands that as well. So for example, hey, wanna do our podcast at 5 o'clock? Actually, let's do 5:30. And it fixes that for me even if I change my mind on the go. So you can just always trust Whisper to get things right for you.

That's awesome, yeah. And let's talk about some other stuff. Like let's talk about the snippet. Yeah. Yeah.

So this is a new thing we released which has become a massive hit. Where you can have something like this where this is something, context about Whisper that I have to drop into every single prompt that I do, right? Now I can just say Whisper intro to do that. And so if I'm in Lovable making a new product, I'd be like, Whisper intro, can you make a referrals dashboard please? And this just lets me do that immediately in my flow without having to copy paste this from my prompt library wherever. And we're actually gonna launch a new thing that has the best prompts across the world that everybody in our community contributes. So you don't have access to just your prompts, but the best of the best.

Yeah, this saves so much typing. Yeah, this is great. I need to use this feature.

Yeah, we need to get you on it, Peter. Yeah. You know, the other thing when we see developers use Whisper a lot for is it essentially becomes the primary thing that they use when they are coding. And so Cursor is one of the products I love, use all the time. And similarly with Cloud Code, Windsurf, this works across the board. We had a lot of our developers using Whisper for all the prompt writing that they were doing, because you can speak much larger and contextual prompts than you can type. But we added some more fun things into it. For example, hey, can you look at auth.ts, constants.ts, and media manager ts and fix the bugs across all of them? And what Whisper does is it tags all of the files that you want and also gets variable names, right? Yeah, this is great.

I use cursor to vibe code a lot and, uh, yeah, you can't really feel the vibes if you're just typing all the time. So like the real vibe code is you just got to talk to it in like a dark room. Yeah. Yeah.

That makes a huge difference. And, and the next biggest thing that we see, you know, is there's a lot of companies that are starting to adopt Whisper Flow across the entire org. right? But they have an open office plan. So how they do it is with the whispering mode where you can actually speak insanely quietly and Whisper still picks it up.

You know, like, um, when I'm at work, I often have to go into a conference room, man. Like, because it's like, I just want to talk to AI. Like, I don't want to talk to my coworkers. I want to go talk to AI. Yeah. So it's actually good that, uh, yeah, I mean, it's in the name. You gotta support the Whisper functionality.

Mm-hmm. 'Cause one of the main reasons why we wanted to kind of build in the spaces, I think voice overall has so much potential, but it still hasn't been something that has, that the people use day in, day out for everything that they're doing. And we wanted to change that. We wanted to make voice a lot more accessible. And that requires a lot of smaller problems that need to get solved. For example, it being insanely fast. For example, it being actually something that you trust, not making mistakes, having it be something you can use when you're around other people, when you're sleeping in bed beside your fiancée. It should just be something that you can, you can continue using. And so there's a lot of problems that come out from there, both on the machine learning side, signal processing, the UI of it, uh, that we need to go out and solve.

Yeah, you know, my, my 7-year-old doesn't even know how to type yet, but like she knows how to use WhisperFlow and, you know, like other voice input tools to like talk to AI and get stuff done. So I think the next generation is not even gonna— like, why even type? Just use your voice. Wait, that's incredible. Yeah. So let's do a little bit of a look back, right? So like right now the company is pretty successful. I'm sure a lot of people, including myself, are using it every day. But you went through quite the journey. Like I see your post on LinkedIn. And in fact, there was a pivot where you went from a hardware company to WhisperFlow is now, right? Can you kind of talk about that a little bit?

Of course. So the mission of the company was always the same, is I wanted to change how people interact with technology and make it just as effortless as talking to a close friend. So growing up, whenever I saw anybody on their phone or computer, it just felt so effortful and so mechanical and not how I want technology to feel like at all. And so 2021, GPT-3 just came out. Me and my co-founder, Sahaj, are looking at that and we're like, people are going to be speaking to their machines in the next few years. Well, when that happens, what happens when you're around other people? You want privacy, you don't want to disturb them. Well, you need a way to still use voice then. And so we need to solve the problem of silent speech. How do you talk to a computer without making any sounds? And so what we built was a wearable device. It looked like a little Bluetooth earpiece. That you could put on and it would go from thoughts to text. Wow. And we worked on it for 3 years, 40 PhDs, machine learning, signal processing, electrical, mechanical, neuroscience, and we actually got this thing to work. Wow. Yeah, and so we started with that and then when it started to work, and this we're talking, this is not too far back, this is like May or June of 2024. Mm-hmm. We, uh, connected with ChatGPT, Siri, Alexa, and it all sucked. And so we realized we had to build our own system that goes from your rambly thoughts into something that is structured and ready to send. And we unsurprisingly call that Flow OS. Now at that time it used to only run on that hardware device, but I wanted to send it to some friends, so I wrapped it in a desktop app. Those 2 friends became 5, became 100, became 1,000, and that product started to explode. And that was when in August of 2024, we made the hard decision of focusing the company entirely on building Flow. Because the thing we realized is before people need a hardware for voice, they need to build the habit of voice. And we had hoped somebody would solve that problem. but nobody did. And so this was the first thing that we needed to build if we want to see, if you want to build the kind of world that we want to see.

Wow, wait, wait, but, uh, you actually got the brain-to-text working? Like you don't have to even talk? You just could do brain-to-text?

You do not have to even talk. Uh, you want me to show you a little sneak peek of it?

Yeah, I mean, that, that sounds like a game changer. That sounds like Neuralink or, you know, I mean, that, that is a game changer, right?

Yeah. It is like Neuralink except it doesn't need brain surgery. So I don't have it live on me right now, but what I do have is one of the videos that we recorded with Sehaj in his kitchen. And this is even before we started the company, right? This is the first prototype that we made of what this would look like. And at that time we got it to do about, 10 words that it could detect that you were thinking about. Later on, we expanded it more so it could do full speech, sound exactly like you. But, but I want to show you a sneak peek of, of how it started. So, so this is July 2021. This is many, many years ago. And here you go.

Wow, dude. So he's not whispering, he's just sitting there?

Nope, nope, he's just sitting there. And by the way, audio is on, there's just no audio in the video.

Okay, I mean, maybe Whisperflow would take you to $10 million or $100 million, and then this product— you should keep working on this product, man. This product sounds pretty amazing.

You should bring it back. It would be absolutely incredible. If you're in the office sometime, I'll show you the actual hardware devices that we built, because those were fully functional and we could use them with Flow. And that's how the initial version of Flow worked.

Wow. Okay. So, but, but you probably had to like, like go on some of the PhDs and kind of like focus the company more on the voice tool, right?

Almost every single one of them. We had to go from 40 people to about 5 people.

Wow. Okay. I don't know, man. Like I feel like if you have this thing, do you try to sell to people or like why didn't people want to buy it? It just wasn't ready yet for people?

It feels like one of those things where if you ask somebody like, hey, would you want to have something that can read your mind and just do things, right? Everybody's going to be like, yes, of course. Like, do you want a flying car? Like, yes, of course. But the thing is people actually don't. Unless there's a very strong use case, it's hard to get that individual to adopt. Because the fact is there are flying cars, uh, they're like single-person, like, eVTOLs that, yeah, you can buy. But I don't think most people want flying cars bad enough, as cool as they sound. And, and similarly, I don't think people wanted silent speech device in a world where they're like, no, I can write my emails just fine. What am I gonna need this for? And people are much more resistant to change their workflows. And this was a time where like a year ago, if you told somebody like, hey, I'm using voice to do everything, people wouldn't believe you. 'Cause voice sucked. Now it makes a lot more sense. And now you look at it and you're like, hey, why does this not exist? And so I think that is, we were ahead of the time when we did build this. So that is likely a play in this in the next few years that's gonna come out. Which may or may not be us, which actually gets this out to market and has real silent speech working for the average person.

Yeah, just let me covertly install it on my wife so I know what she's thinking without her knowing it.

I tried this problem for you and me. It doesn't solve that. It doesn't solve that.

Are you trying to install it on your wife? Is that what you're saying?

I tried. I tried. I knew nothing better. After that than I did before.

Got it, got it. Okay, got it. Okay, all right. Let's leave our wives out of the interview.

I'm going to get scolded for this, man. What are you doing?

Yeah, okay, well, cut this part out. No, no. Yeah. Okay, cool. All right, man. Let's go back to Whisperflow. This episode is brought to you by Bolt. Bolt recently launched a major upgrade to become the fastest tool for product teams to prototype new features. Now you can use Bolt's browser interface with the best AI agents like Claude Code to build working prototypes with real infrastructure like databases, authentication, and hosting, all managed for you automatically. Now Bolt projects can also sync with GitHub for version control, and the whole product team can collaborate in a shared workspace without having to switch to other tools. Check out Bolt V2 at bolt.fyi/peter-yang to get started. Now back to the episode. I think some people on Twitter are like, oh, you know, like Whisperflow is just like a wrapper on like some OpenAI stuff, right? Like, you know, this is just voice, like what's the big deal? But like there's like a lot of stuff that went behind this. In fact, you have a team of, you have a team of like 100 people now, right? Or something. It's like a pretty decent company.

Uh, no, team is small. Team is 25 people. Okay. Okay. I don't know. Care a lot more about quality than quantity, but we, what we do have is some incredible machine learning PhDs who make Whisper what it is. And I think right now Whisper is the best model in the world for voice across 80 different languages if you benchmark it on latency and accuracy.

So is the magic in the model or like the stuff around it? Or maybe give us a, like, give me like an explain like 5 version of how Whisper Flow works.

The magic comes from and insane attention to detail. Hmm. The magic comes from the fact that we know that, hey, if somebody's rambling a list, if you put it in bullets, people are gonna be delighted. When you do that, if you also put a colon, it shows that like little attention to detail, people will like it. If they're talking to 3 different Bryans and Whisper got the right Bryan's name every single time, no voice tool has ever done that. Most LLMs also fail at doing that. And that is going to make people feel special. And there's a lot of these small problems and each of them are hard, right? Most of them don't have a machine learning solution. Most of them aren't like, hey, we're just going to fine-tune this model and it's going to work. It doesn't. And that is, I think, the hard part of it, which is there are dozens of companies that have built speech-to-text or what I would call transcription models where it takes every word that you say and it tries to put it down word for word, which is great. But there is no company apart from Whisper that has built a dictation model where it goes from what you speak to what you wanted to have written, which is not the exact word for word. It's different. And so I would say this is a completely different category of models and product that we're building right now that hasn't existed before.

Yeah. With some of the voice tools I used before Whisperflow, I always had to just copy the whole transcript into like ChatGPT or something and be like, hey, can you clean this up? Can you like make this like bullets and like clean up all the random words I said? Yeah. But I think Whisperflow is the one where like I don't have to go through that process. I can just send. Yeah. So maybe there's like a prompt or something in the background working or like, there's probably a lot more than that, right?

So I wish it was a prompt that would make our life so much easier. Yeah. But no, there's a lot more that goes in there. Cause the fact of the matter is the way you write is very different than the way I write. And even within that, the way you write your emails versus Slack versus text is gonna be, that's gonna have variations across it. And so what matters a lot is how do you make the user happy in every single scenario? Which is the heart of what we're building is not just a voice dictation product, but it's actually the product that is able to learn context both short-term and long-term, is able to build a model of the person so that it can make you happy every single time, and the feeling you get is this product just gets me. And so that contextual engine is the hardest part of the technical stack to build and do well. And also one of the things that I think very few companies have been able to crack.

Yeah, I wasn't even aware of that, dude. Like, so the model actually is learning about how I talk and like getting— give me better output over time. Yeah, that's awesome. Yeah, um, because yeah, because I was going to ask you about like, um, product modes and like, you know, retention, right? Like, like, um, like right now it does a great job of just like changing my voice into text that's, you know, actually well-formatted and correct. But like, where do you plan to evolve this product moving forward?

Like, you know, you know, on the, one of the great things we've seen is our retention numbers are likely the highest of any AI product that's out there. —both in terms of paying users, usage overall. And that all comes from the fact that the key metric that we have in our team is not ARR. It's not number of users that we track. The key thing that we track is how good of a job are we building to, or are we doing to build a habit of voice for people? Because if we nail that, that gets you the word of mouth. That gets every single person to be posting on Twitter and talking about the product, which gets us our incredible growth rates. And so when I think about where we, we go next is, uh, my mental model for that is say there's 8 hours a day that you spend on your phone and computer. I want as much of that as possible to happen through Whisper. Mm-hmm. Which is why we started off with typing, right? Typing is our biggest time sink. The average employee spends about 5 hours every single day typing. That's the part of the day that we first go in and capture. And actually save the person 3, like 2 hours a day just on typing alone. Then the next thing that we do is like a series of actions that are the next most frequent things that this person does. And so Whisper in its limit starts to feel a lot more like Jarvis, where you speak and it writes for you, you speak and it does things for you. And finally, because it knows so much about you over every single application, it helps you proactively. Got it.

Yeah, I'm just thinking about like what else I spend a lot of time on, like I guess meetings, copy and pasting stuff over and over, you know, trying to understand, trying to read a bunch of documents, like, you know, just trying to get people to summarize. Yeah, you can probably help on all that stuff.

Yeah. And, and it's not like we're not going to build like, oh, booking an Uber into Whisper. It's a very niche, small use case, honestly not that valuable. But hey, like if that is something that Peter, you're spending 10 times in a day doing, maybe an hour a day, every day, I want to know what that is. Cause I want to help take that hour in your life and give that back to you and let Whisper take care of that for you.

Yeah, man. I mean, as a PM is like, you know, it's like writing Google Docs and like, yeah, like, you know, doing all that stuff. Yeah. If you can take that away from me, use my voice, that'll be amazing.

Yeah, I wanna, I wanna do more. I wanna, I think as a, as a PM, maybe the best thing you do is take in a lot of information, absorb it, then have ideas on, hey, what do you wanna do next? And that is what likely your deep work time looks like. Then you're spending a lot of time maybe putting the document in the right format. Maybe it's all the time you're spending like on Slack doing back and forth with your team. Every time you open Slack, you are getting distracted by the other 50 unread messages that you have. And so the question that inside Whisper and in our user research calls we're asking people is like, what are your time sinks? Where would you rather not spend time? What gives you joy? What does deep work look like for you? And how do we scale the deep work and reduce the grunt work? And Whisper is essentially the core platform that's, hey, they're with you 24/7. Actually has the opportunity to do that.

You know, that is how AI is good for society, man. Like getting rid of all the groundwork. No one wants to do the work that no one wants to do.

Yeah. No, that's my core thing. I wanted to augment people, not replace them. And you augment people by taking over the, the like garbage work they don't want to do. Yeah. That's the biggest value I feel whenever I build any tool for me that does that.

Got it. Yeah. You want people to just be in the flow state ideally. Yes. Yeah. That's why it's called Whisper Flow. That is where the name Flow comes from. Got it. Okay. All right, man. Well, how about, how about like building companies? Like, do you have any product or company principles when it comes to, you know, building a product? I guess one of them is like obsession about details. Yeah. Uh, you know, what, what have you learned on this journey? Yeah.

Peter, that is so much. We can talk about it for the next, next couple of days nonstop, but I'll talk about a few different things, uh, that really help. Let's start about things with the product first, right? The key thing that we realized is when we are building software, right? Whisperflow as a software is supposed to build habits. So we actually don't look at other software companies for inspiration because most software companies are terrible at building habits. What I look at are games because there are games that people come back home to and play every single day that they get hooked on. That becomes a part of their personality. Games are the best habit builders of anything that's out there. And so I get a lot of my product philosophy from looking at how games are built. And I bet that's part of what you do at Roblox as well, 'cause I guess you are, you guys are building games. And so where that goes to is a lot of key insights there, which is one, take the example of product education. I think about our product as, okay, there's a game, there's a number of interactions, and these key components that we want to educate people, these are the rules of the game. Press and hold the function key, text shows up, here's the other things you can do. These are all the tools in your toolkit that you can use. And what games do really well, they don't jump you in and teach you all the controls at once. No, no, you slowly level up as you get better at one thing, then it teaches you the next thing, and then you have time to master it, then it teaches you the next thing. And if you look at Mario, Mario does that in a phenomenal way. So I'm much more inspired than Mario while building Whisperflow than any other voice or AI tool that is.

Yeah. They don't have a lot of onboarding flows. It's just like playing the game is like the onboarding.

That's kind of, yeah. Right. Playing the game is the onboarding and that is what makes it magical. Mm-hmm. Yeah.

Yeah. Okay. Uh, playing the game and then, um, what else? Give, gimme, gimme like a few, few more. Yeah.

Another one I would say is, again, when I think about brand building, right? Most startups build a brand that is just their product. You're like, okay, this company does this one job, great. But there are some companies whose brand transcends the products and it essentially becomes a core identity. Take the example of Nike, right? Nike sells a commodity and the thing they do is in all of their ads, they celebrate athletes. And so when you think of Nike, when you go in a Nike store, you're not thinking about shoes, you're thinking about your dreams and aspirations, and that's what you associate the brand with. And so again, when we think about the Whisper brand or what job the marketing team is doing, I want to look at Nike, I want to look at Gucci and Chanel and LV, and I want to see what they're doing to create this brand that everybody aspires to get to. Got it.

I, I feel like a lot of it is like maybe tying the brand to the emotion, do you feel? Or like, you know, Nike is kind of like, you know, overcoming adversity, you know, just, just do it kind of thing. And yeah, maybe you guys is like, yeah, maybe it goes back to like the flow state versus like the grunt, grunt work stuff.

Yeah, it all is like, what is the feeling? Like, like when you heard me say like Like Peter, I want to give you back hours of deep work in your day. The feeling you got is the feeling I want people to get when they, when they think of Whisper. Got it.

Got it. Yeah, that, that makes a lot of sense. How about on the company building side? Is it just about keeping the company as small as much as possible or what's the company building side? Yeah.

Actually, no, I look at a different metric. The thing I look at is output per employee, right? So what, what happens is when you're a 5-person company or a startup, you're moving really fast. Then when you get to 25 people, you aren't 5 times as fast. Most companies are maybe twice as fast as 25 people. And then you have this like sublinear curve, 1,000-person companies and 40 times as effective as a 25-person company and, and so on. So the thing I care about is how do I keep the output per employee metric as high as possible? 'Cause if we can make a 25-person company 5 times as effective as a 5-person company, you're winning. 'Cause then you can scale that. And there's a couple of ways to do that. It's like first, like we started, 'cause the problem was about a couple of months ago, we were in the state where we were just twice as effective when you were 5 people and that was driving me nuts. And so we restructured how we were building the company, running everything. Went down to a couple of things. Number one is communication overhead. When your team starts to get bigger and bigger, meetings start to have more people because apparently more opinions make things better, uh, which is absolute lie. Uh, it makes it a little better, but what you're losing is all the work that they could be doing in that time. Because when you ask somebody for a 5-minute opinion on something, they're then thinking about that for another hour every day. And that is wasted productivity. So, okay, make communication loops a lot, lot smaller. Number 2 is decision-making. A lot of people start to have a lot of opinions on things, which is, you know, great. You care about the product, you care about the company, but also that slows down decision-making. I want decisions to be made like this. And so what we do instead is for every single thing that we're doing, that is one very clear decision maker. Everybody else's opinion is an opinion. They can take that or not. They don't need anybody else's sign-off. And this also means we need to hire people who I can trust to make decisions and not even have to loop me in. Um, but what that does is we in our 25-person company right now are running about 20 different projects in parallel, each with independent decision makers. Usually 1 to 2 people who are working on any given thing at a time. And we have pretty much upped our velocity by 2 to 2.5x over the last couple of months by starting to put these things in practice.

And you measure output in terms of like, uh, yeah, like number of features shipped or like, you know, kind of like that kind of stuff, right?

Or, or yeah. Yeah. Number of features shipped, number of experiments run, number of deals closed, just all of that. Like with every team it differs what, what I care about. Yeah, um, but you just feel it in the office.

Yeah, I think that's really important, man. Like, um, I, I think there's a difference between getting a lot of people's opinion on stuff, which you probably should do just to like, you know, get diverse opinions, versus a lot of people trying to make a decision on stuff. Like, like if anything you make requires like 10 checkboxes, then you're screwed. Screwed. Yep. You know, so empowering people was like really important. Exactly. Yeah, exactly. What about using AI in the office? Like, I imagine everyone has a microphone.

Everybody actually has a microphone. Yeah. Yeah. We have these gooseneck mics that like come up like this and people are using AI with that all day because we are in an open office situation. Okay. I would say at this point, I, it's funny. I don't even think of like, are you using AI tools or not? It's more of a question of like, it's a given that everybody's using it. It's again, and the thing that the team shares among each other is what are the best ways they found to use it? For example, I just got our marketing team onto Claude Code. Why? Because I can get them to have the code base on their computer, Claude Code running there, and then they can ask themselves what a feature does, or if they want a new feature built to be marketed, they can ask Claude Code how hard is it gonna be? And so on, and it has access to the codebase so it can answer so many of their questions about the product. That gives them a lot more power. Our product team, for example, uh, whenever they want to talk about a new feature, it's not just a PRD. The PRD is joined with a lovable prototype that they can just like show, like, this is what I'm talking about. And then the engineering team knows exactly what to build. And then, for example, like whenever we're kind of doing a brainstorming session people have to show up to the meeting with having done the basic ChatGPT work beforehand so that, okay, you've got the core thing set up on the way. Now the hard thing that we have to do is of these 20 ideas that ChatGPT said could be good, what are the 3 that we're doing? And so we would spend the meeting talking about that. Then it's having as a brainstorming meeting because the coming up with ideas part, ChatGPT does well. Actually making decisions based on what you know about your company and business, that you can do a lot better.

Yeah, yeah, because it's way faster to just consult AI and make your— just, just like use AI to make your thing better before you actually go meet with a human. Like, that, that's kind of the TRDR, you know?

Yes, that is now like an unset expectation within the company. You just have to do that.

Got it, makes sense. Um, and, uh, let me ask you a few, like, just two more questions, man. So, um, yeah. I feel like there's like a lot of hype, man, in the AI space. And yeah, there's a lot of companies talking about AR and agents and all this stuff. And I feel like a lot of this stuff is kind of bullshit, man. Like, I don't know, do you think we're at a point where you have a bunch of AI agents running all over the place doing everything for you? Maybe in coding, that's true. But like, what do you think?

Yeah, I think Agents is honestly the dumbest term I've heard. We're probably gonna get a lot of shit for it, but I cannot take anybody seriously when they start using the word agents. No, but I do think that is the end goal, but I think a lot of people in the industry are approaching agents wrong. You're thinking of it as more so building capability, like, hey, we built a thing that can actually connect to your Slack and your this thing and do things for you. Like, no, that is, that is step zero. That is not even something to be proud of right now. The thing you gotta be proud of is, is there one task that your agent can do reliably? Just gimme one thing, one thing that you can do that I can blindly trust you with. And that is the thing to be proud of. 'Cause that is a bar that most, almost, I rarely have seen any product meet that bar. Got it. And so that is, I think, what the most successful companies in the space are gonna do. 'Cause most AI agents are at the level of mediocre interns at this point. Yeah. Actually building reliable agents is way, way harder than that. Needs a really good understanding of the problem you're solving, not trying to boil the ocean. There's a number of startups that I advise on this, and I think once people have the right essentially loss function that they're optimizing for, they have the right problem and like framing of it. Hmm. It's solvable. All problems are solvable. Uh, but generally with the, with the AI space, the other question that you were asking, is it, is it hyped up with the ARR numbers and all? Yeah. The thing I really want to see with the companies is what does retention look like? Because right now in the world, there's a lot of AI tourists. Yeah. They just try something and churn. Yeah. They just try something. They may even pay you money, but 2 months later, you're not having that user. So what happens is like, yes, companies get to $100 million of ARR very quickly, but then the problem is of the $100 million they have today, they're just going to have $60 million left the next month. So they have to spend all the marketing dollars to recover the loss, the churn revenue back, and then grow on top of it. That I think is the big problem to solve. There's another problem where like some of these companies are not unique and economic profitable, especially all the white coding tools. That I think overall I'm less worried about, 'cause model costs, all of that is gonna go down over time. They'll do optimization. So that is less existential of a problem as retention, I would say is.

Yeah, I mean, it's like Product Building 101, right? That you need to have good retention. Yep. A lot of the investors these days are just like, oh, like how fast can you grow AR? Like, I don't know what's going on, man. They don't care about retention. It's just like, grow AR.

They, they are, they are. And we, we've been in an interesting position where ARR has been doing well. It's growing 50% month over month. Can't complain. But I, in my last board meeting with our investors, I told them we are no longer having ARR as a north star metric. They're going to track it. We're going to look at it, but that is not the thing that every person in the company is going to know back of hand. The thing that we are gonna be the best at is we're gonna track how our habit building looks like, which is all the numbers around retention, engaged users, not just if somebody signs in, up, signs up or not. And, and all of that. 'Cause I think once you have that, then your product starts to look a lot more like Spotify and Duolingo. Yeah. And those are the companies when I think about habit building, I would rather model Whisper off of.

Yeah, even just having that goal, man, probably would differentiate you against a lot of the other AI companies because now your people are focused on the right thing.

I think so. I think this is the right thing to be focused on. Got it. And the revenue is going to come once you start adding more value to people. Yeah, that makes sense.

Yeah. Um, okay, cool. All right, man, so let's talk about, um, how can people follow in your footsteps, man? Like, because I, I was looking at your background, you did competitive coding before, uh, you even went to the point of stealing your mom's laptop to code at night. Like, where, where does this motivation come from, man? Like, how do people— how can people have the same drive, man? How can people, like, you know, make something of their life?

You know, I, I don't think passion just, just comes. I think the, the key thing that always drove me was curiosity. I wanted to— I saw somebody with an app and I was like, I want to build that. I wonder how you build an app. And then I go went and taught myself. And then the more I did it, one, the more I was enjoying it, and two, the better I was at it. And often competency precedes passion. And so once you just see yourself killing it at something, you just want to do that again and again. That's what it was for me. That's what it was like. I would rather not sleep and code because one, I can build stuff out of thin air and it's magical. And I can also make money off of it, which was great as a child, right? But, uh, but, but two was like, it essentially became my craft. It's what I, it's what I dreamt of. It's what I did day in, day out. It's what I talked about. It's what I obsessed over. And you know, it's like, not saying everybody should be a founder, everybody should be a software engineer, but I think everybody should have something that you're madly, deeply obsessed about that you want to devote your life to that is bigger than yourself. 'Cause I think it just generally makes it so much more worthwhile.

So even though you were not very good at it in the beginning, you just kind of like wanted to get better at it. Is that kind of— Yeah. The gist of things? Yeah. Yeah.

I was, I was a shit programmer as a 9-year-old, but a year later I was less shit. Got it.

Got it. Got it. Yeah. And I feel like having that, um, self-drive to learn whatever the hell you're interested in is actually really important, man. Like, especially in the US, dude, because the education system here is not great. And I'm like, I'm trying to teach my kids to actually just like learn whatever the hell they want to learn about. Like don't, don't rely on a teacher. Yeah.

No, loving to learn. I think like when I have kids, that's the, the one thing I want them to have.

Yeah. Just give them all the AI tools and let them go explore whatever the hell. So, um, where can people find Whisperflow and where can people find you?

So you can find Whisperflow at our website, which is w-i-s-p-r-flow.com. Flow.ai, and I will pull up the website for people to see. You can get started with the product on Mac, Windows, and iPhone. And if you have any questions or more, feel free to just reach out. I'm available on Twitter and LinkedIn, but really excited for a lot more people to switch over from their keyboards. Because I think there's a whole world out there for people to experience.

Awesome, man. Yeah, we had such a great chat. I lost track of time. So yeah, that's good.

Me too. Me too. This is fun.

All right, dude. Thanks for your time. Yeah.