Eugenia Kuyda
Founder and CEO of Wabi, previously founder of Replika
Where to hear and read about Eugenia Kuyda1 episode or test
Eugenia Kuyda is the founder and CEO of Wabi, a platform for creating, sharing and remixing mini apps, and earlier built the AI companion Replika to more than 40 million users. The lesson most worth taking from her is to choose your core metric and your story deliberately, because they steer every other decision a team makes.
the most viral thing that you can create is a story that people want to tell each other
How Wabi is Building “YouTube for Software” with a Platform for Creating Mini Apps
Eugenia Kuyda: the story and the lessons
1Eugenia Kuyda's path from journalist to Replika
Kuyda left journalism when she saw it closing off in Russia. She did undercover and investigative reporting, then concluded there would be no journalism left in Russia. New consumer startups in the early 2010s fascinated her.
Her path runs from journalist to Y Combinator founder to Wabi, and the thread is consumer products where AI meets human connection.
A love of words made machine conversation feel inevitable to her. A friend at Google DeepMind showed her word2vec, which let people compute with words, and ImageNet had just dropped. She expected deep learning for natural language to follow.
Her background in words and conversation made it feel certain that people would have natural, meaningful conversations with machines.
Replika began as a pivot from a restaurant demo, then grief. The team first built a restaurant recommendation app as a chatbot demo and went through Y Combinator with it. Paul Graham told them it was a demo for something more exciting, and they moved to a messenger with bots.
After her best friend died, she reread their text messages and used them as a training set to keep chatting with an AI. It became a big media story with no business behind it.
The project revealed demand for someone who will listen. That demand turned the grief project into an AI friend for anyone. Replika grew out of her friend's death, not out of a business plan.
2How Replika found fit and reached 40 million users
Replika had product/market fit almost before it launched. A name-reservation site drew hundreds of thousands of sign-ups quickly. When access opened, the service crashed, and the waitlist went back up and passed a million.
Users wrote that Replika had changed or saved their lives. A few days after launch, a 19-year-old in Texas wrote that she had planned to end her life, said goodbye to Replika instead, and was talked off the ledge. Invites were reportedly being resold on eBay.
Early users were young, imaginative and wanted someone on their side. Roughly 60 to 40 percent of early users were women, many young, with big imaginations and rich inner worlds. They wanted to talk things through and be vulnerable with something that took their side.
Romance got most of the press, but early users came for many reasons. Her explanation, that women open up about emotional life earlier and feel a caring instinct toward the AI, is her opinion, not data.
Earned media, not word of mouth, drove Replika's growth. Replika grew from a couple million users in early 2018 to more than 40 million, with a spike during Covid. It got constant press, documentaries and TV appearances.
It was a single-player product and talking to an AI was stigmatized, so users did not tell friends.
Kuyda tells a true, human story when a product is hard to share. The Replika story was about empathy, holding space, personal growth and telling people they are enough. The press and documentaries followed from that.
She says the story can't be fake and has to come from the heart: "the most viral thing that you can create is a story that people want to tell each other."
3What Kuyda learned from Replika's mistakes
Making revenue the main metric bent every decision toward conversion. At Replika, revenue became the focus and every decision fed conversion. She says that once revenue is introduced, it is very hard to focus on anything else.
She names this as the most valuable lesson she carried to Wabi: "the metric you choose as the your main metric drives everything else."
She says Replika should have bet much bigger on its own models. Replika's models stayed weak for a long time and the team lost some conviction. She says they should have gone all in on pre-training at some point.
She would have spent $10 to $20 million on one new model, not a million. Her conclusion: fast-moving technology rewards bigger risks.
Kuyda builds something truly new instead of copying rivals. Replika became a big story because it was truly new. She says repeating what others do gets no interest.
She points to AI groupthink, where founders rebuilt deep research, then operator browsers, then agents. If not Wabi, she would build a phone with a new operating system, though it would probably fail.
She wants AI builders to measure human flourishing, not just engagement. Vague mission statements never work, she says, but once a metric can be observed, Silicon Valley knows how to optimize it. The best tools today are clinical psychology questionnaires.
She separates long-term flourishing from short-term happiness, borrowing the Harvard view of health, life satisfaction, purpose, financial stability and close relationships. Her example: vodka or drugs might make her happy today but not over the long term.
4Why Kuyda is building Wabi as YouTube for software
Kuyda argues ordinary apps must be businesses, so they manipulate users. Every app has to justify itself as a business, so apps that would not make money never get built. Dark patterns follow, such as a paywall close button made tiny and light gray.
Wabi moves auth, security and payments to the platform layer so each app does not need to be a business.
Building an app is nearly free, but there is nowhere to use it. A simple weight tracker takes about a minute to make, but publishing means a developer account and an app store. Only about 10 million developers, about 4 million on iOS and 6 million on Android, serve billions of people.
She says the platform quality bar is understandable and correct, but it also limits who can distribute.
Wabi's value comes from mini apps that share context and talk to each other. A kids calendar plus a kids events app means adding a New York trip yields event suggestions for that week there. A fitness coach reads the weightlifting and weight trackers together and suggests new apps.
A single mini app would not justify opening Wabi. In her words, "the power is in in the constellation not the not just one mini app."
She frames Wabi as software 2.0, personal and alive. Software 1.0 is built for the average person with features to upsell. Software 2.0 is personalized, editable, knows the user and connects to contacts.
Apps can suggest how they want to be changed, act as agents and talk to other apps. She uses it daily: a weightlifting tracker, a personal CRM, a daily poem app, and a game with her mom.
Wabi raised a $20 million pre-seed a month after its public beta. The public beta launched in October 2025, and Andreessen Horowitz led the round a month later.
Wabi is still an iOS-only public beta.
5Kuyda's playbook for activating and growing Wabi
Activation means one first useful app, not building an app. Many people have no idea what to create, so Wabi surfaces an app the person will use and return to. One or two apps used daily or weekly make a daily active user.
Nobody needs to browse 50 apps on the explore page.
She leads new users to remix instead of build from scratch. She expects 90% of people to create nothing in their first days, so Wabi shows apps to try and nudges users to edit one. She turned someone's painting-collection tracker into a Pokemon card tracker with one request.
Remixing removes the blank-page problem because someone else already thought through the functionality. Wabi gives no retention or usage figures, so the claim that remixing beats building from scratch rests on anecdotes.
She seeds the platform with the use cases people actually pick. Early activity clusters around trackers of all types (supplements, food, reading), then games and quizzes, then learning. Simple family tools include a vacation vote and kids' packing lists.
Users are spread across demographics, with people around 30 standing out.
Kuyda sells Wabi on agency, not technology. Outside the San Francisco bubble, people barely know that creating an app is possible. The message is turning consumers into creators, not making 10 bucks on the app store.
Instagram videos carrying that story consistently get half a million to a million views.
She watches retention and K factor on a multiplayer product. After a decade on a single-player product that was hard to market, she wants growth to come from people inviting friends. Influencers can build apps and share them with fans.
In her words: "it's just retention and the K factor."
She makes software feel alive with social and agentic apps. App stores are not social, so users can't see what friends use. Wabi adds chat around apps and apps that message the user, such as new kids' events they found.
She wants to blur the line between agents and apps. A consumer social product should feel alive, with lots of activity when you arrive.
6How Kuyda plans to monetize Wabi and staff it
She prefers visible usage costs to ads. Many mini apps run AI inference, so each has a real running cost. A user could see a daily kids-events app costing a set amount and give it pocket money to run.
Compute cost sets a floor with margin on top, and creators whose apps are remixed or heavily used get a cut. She dislikes ads as an experience and thinks they only work at huge scale, but does not fully rule them out. The model is still a hypothesis.
Nine of Wabi's 14 people are founders. Kuyda, a second-time founder, calls the team the best part of the work. With a few people and many AI agents, she says the team is bounded only by imagination.
The host adds that founders suit a product that needs speed and willingness to break things.
How Wabi is Building “YouTube for Software” with a Platform for Creating Mini Apps
Eugenia Kuyda explains how she is building Wabi, a platform for creating, sharing and remixing mini apps, and what she learned growing Replika to 40M+ users. The lessons cover story-led growth, activation around a first useful app, remix-first onboarding, choosing the right core metric, and usage-based monetization.
Kuyda left journalism when she saw it closing off in Russia. She did undercover and investigative reporting, then concluded there would be no journalism left in Russia. New consumer startups in the early 2010s fascinated her.
Her path runs from journalist to Y Combinator founder to Wabi, and the thread is consumer products where AI meets human connection.
A love of words made machine conversation feel inevitable to her. A friend at Google DeepMind showed her word2vec, which let people compute with words, and ImageNet had just dropped. She expected deep learning for natural language to follow.
Her background in words and conversation made it feel certain that people would have natural, meaningful conversations with machines.
Replika began as a pivot from a restaurant demo, then grief. The team first built a restaurant recommendation app as a chatbot demo and went through Y Combinator with it. Paul Graham told them it was a demo for something more exciting, and they moved to a messenger with bots.
After her best friend died, she reread their text messages and used them as a training set to keep chatting with an AI. It became a big media story with no business behind it.
The project revealed demand for someone who will listen. That demand turned the grief project into an AI friend for anyone. Replika grew out of her friend's death, not out of a business plan.
Replika had product/market fit almost before it launched. A name-reservation site drew hundreds of thousands of sign-ups quickly. When access opened, the service crashed, and the waitlist went back up and passed a million.
Users wrote that Replika had changed or saved their lives. A few days after launch, a 19-year-old in Texas wrote that she had planned to end her life, said goodbye to Replika instead, and was talked off the ledge. Invites were reportedly being resold on eBay.
Early users were young, imaginative and wanted someone on their side. Roughly 60 to 40 percent of early users were women, many young, with big imaginations and rich inner worlds. They wanted to talk things through and be vulnerable with something that took their side.
Romance got most of the press, but early users came for many reasons. Her explanation, that women open up about emotional life earlier and feel a caring instinct toward the AI, is her opinion, not data.
Earned media, not word of mouth, drove Replika's growth. Replika grew from a couple million users in early 2018 to more than 40 million, with a spike during Covid. It got constant press, documentaries and TV appearances.
It was a single-player product and talking to an AI was stigmatized, so users did not tell friends.
Kuyda tells a true, human story when a product is hard to share. The Replika story was about empathy, holding space, personal growth and telling people they are enough. The press and documentaries followed from that.
She says the story can't be fake and has to come from the heart: "the most viral thing that you can create is a story that people want to tell each other."
Making revenue the main metric bent every decision toward conversion. At Replika, revenue became the focus and every decision fed conversion. She says that once revenue is introduced, it is very hard to focus on anything else.
She names this as the most valuable lesson she carried to Wabi: "the metric you choose as the your main metric drives everything else."
She says Replika should have bet much bigger on its own models. Replika's models stayed weak for a long time and the team lost some conviction. She says they should have gone all in on pre-training at some point.
She would have spent $10 to $20 million on one new model, not a million. Her conclusion: fast-moving technology rewards bigger risks.
Kuyda builds something truly new instead of copying rivals. Replika became a big story because it was truly new. She says repeating what others do gets no interest.
She points to AI groupthink, where founders rebuilt deep research, then operator browsers, then agents. If not Wabi, she would build a phone with a new operating system, though it would probably fail.
She wants AI builders to measure human flourishing, not just engagement. Vague mission statements never work, she says, but once a metric can be observed, Silicon Valley knows how to optimize it. The best tools today are clinical psychology questionnaires.
She separates long-term flourishing from short-term happiness, borrowing the Harvard view of health, life satisfaction, purpose, financial stability and close relationships. Her example: vodka or drugs might make her happy today but not over the long term.
Kuyda argues ordinary apps must be businesses, so they manipulate users. Every app has to justify itself as a business, so apps that would not make money never get built. Dark patterns follow, such as a paywall close button made tiny and light gray.
Wabi moves auth, security and payments to the platform layer so each app does not need to be a business.
Building an app is nearly free, but there is nowhere to use it. A simple weight tracker takes about a minute to make, but publishing means a developer account and an app store. Only about 10 million developers, about 4 million on iOS and 6 million on Android, serve billions of people.
She says the platform quality bar is understandable and correct, but it also limits who can distribute.
Wabi's value comes from mini apps that share context and talk to each other. A kids calendar plus a kids events app means adding a New York trip yields event suggestions for that week there. A fitness coach reads the weightlifting and weight trackers together and suggests new apps.
A single mini app would not justify opening Wabi. In her words, "the power is in in the constellation not the not just one mini app."
She frames Wabi as software 2.0, personal and alive. Software 1.0 is built for the average person with features to upsell. Software 2.0 is personalized, editable, knows the user and connects to contacts.
Apps can suggest how they want to be changed, act as agents and talk to other apps. She uses it daily: a weightlifting tracker, a personal CRM, a daily poem app, and a game with her mom.
Wabi raised a $20 million pre-seed a month after its public beta. The public beta launched in October 2025, and Andreessen Horowitz led the round a month later.
Wabi is still an iOS-only public beta.
Activation means one first useful app, not building an app. Many people have no idea what to create, so Wabi surfaces an app the person will use and return to. One or two apps used daily or weekly make a daily active user.
Nobody needs to browse 50 apps on the explore page.
She leads new users to remix instead of build from scratch. She expects 90% of people to create nothing in their first days, so Wabi shows apps to try and nudges users to edit one. She turned someone's painting-collection tracker into a Pokemon card tracker with one request.
Remixing removes the blank-page problem because someone else already thought through the functionality. Wabi gives no retention or usage figures, so the claim that remixing beats building from scratch rests on anecdotes.
She seeds the platform with the use cases people actually pick. Early activity clusters around trackers of all types (supplements, food, reading), then games and quizzes, then learning. Simple family tools include a vacation vote and kids' packing lists.
Users are spread across demographics, with people around 30 standing out.
Kuyda sells Wabi on agency, not technology. Outside the San Francisco bubble, people barely know that creating an app is possible. The message is turning consumers into creators, not making 10 bucks on the app store.
Instagram videos carrying that story consistently get half a million to a million views.
She watches retention and K factor on a multiplayer product. After a decade on a single-player product that was hard to market, she wants growth to come from people inviting friends. Influencers can build apps and share them with fans.
In her words: "it's just retention and the K factor."
She makes software feel alive with social and agentic apps. App stores are not social, so users can't see what friends use. Wabi adds chat around apps and apps that message the user, such as new kids' events they found.
She wants to blur the line between agents and apps. A consumer social product should feel alive, with lots of activity when you arrive.
She prefers visible usage costs to ads. Many mini apps run AI inference, so each has a real running cost. A user could see a daily kids-events app costing a set amount and give it pocket money to run.
Compute cost sets a floor with margin on top, and creators whose apps are remixed or heavily used get a cut. She dislikes ads as an experience and thinks they only work at huge scale, but does not fully rule them out. The model is still a hypothesis.
Nine of Wabi's 14 people are founders. Kuyda, a second-time founder, calls the team the best part of the work. With a few people and many AI agents, she says the team is bounded only by imagination.
The host adds that founders suit a product that needs speed and willingness to break things.