Tanay Kothari
Cofounder and CEO of Wispr, maker of the Flow voice dictation app
Where to hear and read about Tanay Kothari1 episode or test
Tanay Kothari is cofounder and CEO of Wispr, maker of the Wispr Flow voice dictation app, and has chased a Jarvis-style voice computer since 2008. The thing most worth learning from him is to build the output people would have written themselves, so they trust it enough to send unread and spread it without paid marketing.
you speak differently than you write. And that's the key insight we went after.
How Wispr Flow Uses AI to Save Professionals Hours Every Day
Tanay Kothari: the story and the lessons
1Tanay Kothari's path to Wispr Flow
Kothari has wanted to build Jarvis since the first Iron Man movie in 2008. With a friend he built one of the world's first voice assistants, before Siri and Alexa. It reached 2.5 million users before Google shut it down because it disliked what the team was doing.
He has since built 60 to 70 products in AI personalization and voice systems. He believes the technology is only now ready to make devices feel like talking to a close friend.
He and cofounder Sahaj Garg met on day one of undergrad at Stanford. They roomed together for three years and agreed that life is short and should go to the most impactful work. Both were annoyed by people stuck staring at phone screens.
They wanted a bigger swing than one app, so Wispr began as a bet on how people interface with technology. Kothari says he knows Garg better than most people really should.
Seven agent products failed before Flow because none built a habit. Wispr's first seven products were agent-style action tools. Watching people work showed that typing dominates their time, about 5 hours a day, and Flow became the eighth product.
Kothari says starting with agents would have been a mistake. Dictation built the habit first, and actions can now be stacked on top.
Flow became the company's focus because the team could not stop using it. The first version drew strong traction inside the company, and everyone began using it for everything they did. That internal pull made them point the whole company at it.
The aim was a product an average person on the street could use.
Product market fit arrived in January with no launch behind it. Every day, 5 to 10 people posted unprompted about Flow on LinkedIn and Twitter, many of whom had never posted about a product before. Exponential growth began that month.
Kothari says that when people say a product changed their life, the 'would you be disappointed' question is unnecessary. Wispr has since raised a $30 million Series A.
2How Kothari designed Flow to write what people mean
Flow outputs what the user meant to write, not a transcript. Saying "let's meet at 5, actually no, let's do six" produces "let's meet at 6". Google, Apple and OpenAI optimize every metric around word-for-word accuracy, which suits recordings but not messages.
Kothari's mental model is a smart person sitting at your computer who writes what you tell them to write.
Flow shapes the same words to fit where they will be sent. A long, rambling list becomes bullets, an email is formatted like an email, and a text comes out lowercase with emojis. The test is whether people look at the result, think it is perfect and press send.
The next feature reads each thread and matches the user's tone there, since people talk differently in a public thread than in a DM with a close friend.
Kothari made Flow work invisibly inside every app. You hold a button, speak and let go, with no copy and paste, no integrations and no tab switching. It has worked in over 25,000 apps and sites.
Notion, Slack, Gmail, Outlook and a plain editor all handle bullets and pasting differently, so each quirk was fixed one by one. He calls it insanely hard to build but non-negotiable. Users who try another tool come back a day later because of one small frustration it never solved.
Flow shows text only at the end, never streaming. Reading words you said half a second ago is distracting and adds little value. Text that is edited after it streams, like a jittery Tesla display, makes people doubt the system.
Waiting also leaves room to make edits before the final output appears.
Wispr tracks trust with a home-made metric, zero edit rate. 81% of messages are sent with no edit, against 10 to 15% for Apple and Google dictation. Flow started near 40%.
Users press send about half a second after Flow writes because they stop reading the output. Kothari says each single point of improvement noticeably changes how happy users are.
3How Wispr ships and holds its quality bar
Features stay internal until half the company cannot live without them. Wispr keeps about 10 times more features inside than users see. The mobile app reached build 152 before it shipped externally.
The bar comes from how basic the product is: a keyboard that failed once in 100 presses would be thrown out the window.
Kothari pairs his speed with Sahaj Garg's polish. Kothari's habit is to ship an MVP after an all-nighter, and he had to adapt it. Garg leans toward focus and doing things right.
In Kothari's words: "I add in the hustle and move fast. He adds the polish culture and together it's polish with speed."
Growth ideas ship within one to five days, kept apart from product. A sensible idea gets no long debate: ship it, then improve it from what the team learns. Every growth initiative is labeled for awareness or for conversion.
This fast culture lives beside a slow, perfectionist product culture. Kothari admits it is hard because people fear mistakes and he counts himself a perfectionist.
He launched on Mac first, where audience and support were best. Silicon Valley, New York and LA are Mac-heavy, and macOS gave the most support for grabbing context. Windows took a few more months because Flow works at the operating system level.
Kothari assumed mobile would inherit desktop's fit and was wrong. The only thing shared with desktop was speaking and getting text back. Mobile had to find product market fit from zero and took about 6 months.
He calls it a rookie mistake. Android differs from iOS at the operating system level, so much will be rebuilt from scratch.
Wispr tests with the least tech-savvy users as the floor. About 30% of users are nontechnical, often people who set up parents and grandparents. The team sat with older users, which they call silvers, to find what was unintuitive.
His reasoning: if the product works for the lowest common denominator, it becomes intuitive for the average person.
4How Wispr grew Flow without paid marketing
Kothari read angry support emails during outages as proof of fit. When growth went exponential in January, systems broke and every slow hour filled the support inbox. Fixed problems brought replies like "I love you guys".
He concluded that making people feel things beats leaving them ambivalent.
Early adopters were founders, deep thinkers and developers. Founders work through communication, thoughtful people reason by putting thoughts into words, and developers who prompt AI avoid switching between speaking English and typing code. Cursor is one of Flow's biggest platforms.
A couple of users dictated one and a half million words into Cursor in six months. The top 20 apps account for only a third of usage.
Wispr gave creators free accounts instead of paying for sponsorships. Creators who accepted used Flow in every video, including one whose videos ran two and a half hours with Flow as the star. Commenters who asked what tool it was got answers from non-affiliates.
Because the creators are unpaid, the endorsement carries trust, and the company makes little content itself. That is why the growth team stays small.
He builds growth on other people describing the value, not the founder. After a hundred broken promises, people dismiss a founder pitching his own dictation tool. A user calling a dictation tool insane is a claim nobody has heard before, so it builds trust.
In his words: "It's not me. It's not Whisper saying things about itself. It's other people talking about the value prop."
Wispr wins individuals first, then upsells their whole company. Flow spreads easily inside companies, so one user can pull in the team. Superhuman, Substack and Mercury have moved over, along with hundreds of other companies including many law firms and financial institutions.
Flow was international from launch, and over half of users are outside the US. The first version worked across 100 languages. The 16-person team speaks 30 languages between them, which shaped the product's culture early.
Word of mouth did most of the work abroad.
How Wispr Flow Uses AI to Save Professionals Hours Every Day
Wispr Flow turned voice dictation into a zero-edit writing tool by outputting what users meant to write rather than a word-for-word transcript, then grew mostly through free influencer seeding. The result is an 81% zero edit rate, a 0.5 second send time after output, and use in over 25,000 apps, with a small growth team and a $30 million Series A.
Kothari has wanted to build Jarvis since the first Iron Man movie in 2008. With a friend he built one of the world's first voice assistants, before Siri and Alexa. It reached 2.5 million users before Google shut it down because it disliked what the team was doing.
He has since built 60 to 70 products in AI personalization and voice systems. He believes the technology is only now ready to make devices feel like talking to a close friend.
He and cofounder Sahaj Garg met on day one of undergrad at Stanford. They roomed together for three years and agreed that life is short and should go to the most impactful work. Both were annoyed by people stuck staring at phone screens.
They wanted a bigger swing than one app, so Wispr began as a bet on how people interface with technology. Kothari says he knows Garg better than most people really should.
Seven agent products failed before Flow because none built a habit. Wispr's first seven products were agent-style action tools. Watching people work showed that typing dominates their time, about 5 hours a day, and Flow became the eighth product.
Kothari says starting with agents would have been a mistake. Dictation built the habit first, and actions can now be stacked on top.
Flow became the company's focus because the team could not stop using it. The first version drew strong traction inside the company, and everyone began using it for everything they did. That internal pull made them point the whole company at it.
The aim was a product an average person on the street could use.
Product market fit arrived in January with no launch behind it. Every day, 5 to 10 people posted unprompted about Flow on LinkedIn and Twitter, many of whom had never posted about a product before. Exponential growth began that month.
Kothari says that when people say a product changed their life, the 'would you be disappointed' question is unnecessary. Wispr has since raised a $30 million Series A.
Flow outputs what the user meant to write, not a transcript. Saying "let's meet at 5, actually no, let's do six" produces "let's meet at 6". Google, Apple and OpenAI optimize every metric around word-for-word accuracy, which suits recordings but not messages.
Kothari's mental model is a smart person sitting at your computer who writes what you tell them to write.
Flow shapes the same words to fit where they will be sent. A long, rambling list becomes bullets, an email is formatted like an email, and a text comes out lowercase with emojis. The test is whether people look at the result, think it is perfect and press send.
The next feature reads each thread and matches the user's tone there, since people talk differently in a public thread than in a DM with a close friend.
Kothari made Flow work invisibly inside every app. You hold a button, speak and let go, with no copy and paste, no integrations and no tab switching. It has worked in over 25,000 apps and sites.
Notion, Slack, Gmail, Outlook and a plain editor all handle bullets and pasting differently, so each quirk was fixed one by one. He calls it insanely hard to build but non-negotiable. Users who try another tool come back a day later because of one small frustration it never solved.
Flow shows text only at the end, never streaming. Reading words you said half a second ago is distracting and adds little value. Text that is edited after it streams, like a jittery Tesla display, makes people doubt the system.
Waiting also leaves room to make edits before the final output appears.
Wispr tracks trust with a home-made metric, zero edit rate. 81% of messages are sent with no edit, against 10 to 15% for Apple and Google dictation. Flow started near 40%.
Users press send about half a second after Flow writes because they stop reading the output. Kothari says each single point of improvement noticeably changes how happy users are.
Features stay internal until half the company cannot live without them. Wispr keeps about 10 times more features inside than users see. The mobile app reached build 152 before it shipped externally.
The bar comes from how basic the product is: a keyboard that failed once in 100 presses would be thrown out the window.
Kothari pairs his speed with Sahaj Garg's polish. Kothari's habit is to ship an MVP after an all-nighter, and he had to adapt it. Garg leans toward focus and doing things right.
In Kothari's words: "I add in the hustle and move fast. He adds the polish culture and together it's polish with speed."
Growth ideas ship within one to five days, kept apart from product. A sensible idea gets no long debate: ship it, then improve it from what the team learns. Every growth initiative is labeled for awareness or for conversion.
This fast culture lives beside a slow, perfectionist product culture. Kothari admits it is hard because people fear mistakes and he counts himself a perfectionist.
He launched on Mac first, where audience and support were best. Silicon Valley, New York and LA are Mac-heavy, and macOS gave the most support for grabbing context. Windows took a few more months because Flow works at the operating system level.
Kothari assumed mobile would inherit desktop's fit and was wrong. The only thing shared with desktop was speaking and getting text back. Mobile had to find product market fit from zero and took about 6 months.
He calls it a rookie mistake. Android differs from iOS at the operating system level, so much will be rebuilt from scratch.
Wispr tests with the least tech-savvy users as the floor. About 30% of users are nontechnical, often people who set up parents and grandparents. The team sat with older users, which they call silvers, to find what was unintuitive.
His reasoning: if the product works for the lowest common denominator, it becomes intuitive for the average person.
Kothari read angry support emails during outages as proof of fit. When growth went exponential in January, systems broke and every slow hour filled the support inbox. Fixed problems brought replies like "I love you guys".
He concluded that making people feel things beats leaving them ambivalent.
Early adopters were founders, deep thinkers and developers. Founders work through communication, thoughtful people reason by putting thoughts into words, and developers who prompt AI avoid switching between speaking English and typing code. Cursor is one of Flow's biggest platforms.
A couple of users dictated one and a half million words into Cursor in six months. The top 20 apps account for only a third of usage.
Wispr gave creators free accounts instead of paying for sponsorships. Creators who accepted used Flow in every video, including one whose videos ran two and a half hours with Flow as the star. Commenters who asked what tool it was got answers from non-affiliates.
Because the creators are unpaid, the endorsement carries trust, and the company makes little content itself. That is why the growth team stays small.
He builds growth on other people describing the value, not the founder. After a hundred broken promises, people dismiss a founder pitching his own dictation tool. A user calling a dictation tool insane is a claim nobody has heard before, so it builds trust.
In his words: "It's not me. It's not Whisper saying things about itself. It's other people talking about the value prop."
Wispr wins individuals first, then upsells their whole company. Flow spreads easily inside companies, so one user can pull in the team. Superhuman, Substack and Mercury have moved over, along with hundreds of other companies including many law firms and financial institutions.
Flow was international from launch, and over half of users are outside the US. The first version worked across 100 languages. The 16-person team speaks 30 languages between them, which shaped the product's culture early.
Word of mouth did most of the work abroad.