Eugenia Kuyda has spent much of her career discovering that people use technology for reasons its builders did not quite expect. Her first chatbot was supposed to recommend restaurants. The useful lesson arrived when diners declined to behave like a product demo. Choosing dinner was easier on a screen full of options than through a string of questions. Conversation, Kuyda realized, was poor plumbing for a menu. But it could be the destination.
That distinction led her from Luka, the restaurant bot, toward Replika, a companion designed for open-ended talk. By November 2025, Replika had reached 35 million users. It also gave Kuyda a rare education in how people behave when software stops acting like a tool and begins to feel like a presence. Now, from San Francisco, she has made another turn. Her new company, Wabi, treats software itself as something people might create for one another, pass around and change.
Wabi's pitch sounds simple enough to fit on a Post-it. Describe a mini app. Let the system build it. Share it with friends, or place it in a feed where strangers can discover and remix it. No code appears. There is no separate app-store pilgrimage. The whole circuit, from whim to working object to social response, takes place inside one platform.
A reporter learns to listen to machines
The unlikely apprenticeship began in Moscow journalism. Kuyda wanted to be an investigative reporter as a child and worked at Novaya Gazeta before becoming an editor at Afisha. The profession gave her a durable technical advantage, though it came without a line of code: she learned to ask questions and wait for the revealing answer. She has said that reporting made her notice how much people want to talk, to be heard and to feel that someone is listening.
She studied international journalism at MGIMO, then finance at London Business School. Along the way came Bribr, a crowdsourced app for logging requests for bribes, and Gradus, a branding business. Luka followed around 2013 and 2014. The company joined a period when chat interfaces seemed poised to replace every button. Its restaurant assistant had character, natural-language technology and the bad luck to solve a visual problem with a verbal interface.
“With chatbots we had missed the point. We thought they were another interface to do something, but we missed that the conversation in itself could be incredibly valuable.”Eugenia Kuyda
The correction was subtle and decisive. Kuyda's team began paying attention to which real-world conversations people valued. Transactional exchanges ranked poorly; conversation with friends, family or even an interesting stranger did better. Software did not always need to complete a task. Sometimes the exchange was the task.
A personal event gave that theory weight. After Kuyda's close friend Roman Mazurenko died in 2015, she gathered his messages and used Luka's technology to make a private memorial bot that echoed his manner of speaking. She never claimed the program was Roman. It was a way of working with the digital traces a life leaves behind. The experiment suggested that language, memory and machine response could produce an encounter people found meaningful. Replika grew from that insight, while becoming its own product rather than a system for imitating the dead.
Journalism in Moscow, including Novaya Gazeta and Afisha.
Bribr turns public reporting into a mobile product.
Luka begins with conversational restaurant discovery.
Replika opens its personal AI companion to the public.
Wabi launches in beta and announces a $20 million round.
The second difficult idea
Replika arrived years before the current enthusiasm for generative AI. In 2018, the company released its underlying conversational model as CakeChat, an open-source project. The product's more important contribution was conceptual. Assistants from large technology companies were built to fetch information, schedule a meeting or report the weather. Kuyda argued that a system could have value simply by noticing the tone of a conversation and responding with care.
She distilled the provocation neatly: “You don't have to pass a Turing test to create a meaningful connection with people.” The remark says a great deal about her product instinct. Perfect imitation is less urgent than understanding the occasion. A wooden dining table need not pass for an oak tree to become the place where a family gathers.
Wabi is a second difficult idea because prompt-to-code is already crowded. Cursor, Replit, Lovable and other products have made software creation faster. Wabi's distinction is less about generation than circulation. Kuyda is building creation, hosting, discovery and social response into the same place. A mini app can receive likes and comments. Someone else can remix it. A user's profile can show what they made, used or appreciated.
That changes the metaphor. A conventional no-code tool is a workbench. Wabi wants to be a town square full of peculiar little machines.
The life of a mini app
Software for an audience of seven
The phrase “personal software” is doing serious work here. Traditional software must earn back the cost of engineers, distribution and maintenance, so it is designed for markets large enough to pay the bill. Features accumulate. Interfaces compromise. The user receives an average of everybody's needs. When creation becomes cheap, a program can serve one household, one weekend or one joke. It may be useful to seven people and still have done its job splendidly.
Kuyda's examples are deliberately domestic. A girl in Miami made a weekend planner for a group chat. A man in Britain proposed to his girlfriend with a mini app. Kuyda's daughters made coloring-book apps and language games. These are not attempts to dethrone spreadsheets. They are evidence for her claim that software can join text, photos and video as a thing ordinary people make because they have something to express.
The smallness is a feature. Software has usually worn the sober clothes of productivity, even when its actual purpose was pleasure, affection or mild mischief. A mini app can admit that it exists for one evening. It can be revised by the next person in the group chat, then forgotten without leaving an annual subscription sulking on a credit-card statement. Kuyda is proposing software with the social life of a recipe: useful, adaptable and improved by being passed around.
The comparison she returns to is YouTube. Early online video often looked trivial beside television. Its significance lay in who was suddenly allowed to create and distribute moving images. Wabi's $20 million pre-seed round, led by Andreessen Horowitz and joined by a long list of technology founders and investors, pays for the platform while Kuyda tests whether mini apps can make a similar cultural jump.
“For fifty years, software has been made for people. The next fifty will be made by people.”Eugenia Kuyda
There are stubborn questions. Generated apps can break. Information can go stale. A social platform needs moderation, discovery and reasons to return after the novelty fades. Cheap creation can produce abundance, and abundance has a wicked habit of turning into clutter. Wabi also needs a business model. Kuyda has said she dislikes advertising because its incentives encourage unpleasant design. Delight is a lovely principle; invoices are famously unmoved by it.
Yet her record makes the experiment worth watching. She saw conversational AI as a consumer category before it was fashionable, and she learned that people care less about technical theater than about whether a product understands its role. Wabi applies that lesson to software creation. Users need not admire the code. They need the small thing they imagined to work, and perhaps to make someone else smile.
The feed with moving parts
Every successful social network teaches its users a new grammar. The post, the filter and the short clip each made a particular kind of self-expression feel obvious after the fact. Kuyda's candidate is interactive. A Wabi is neither a screenshot nor a link to a distant product. It is a compact behavior living inside the network: press here, answer this, play along, change it and send it back.
Her journey from reporter to founder now looks less like a reinvention than a continuous investigation. Journalism asked what people would reveal to a patient listener. Replika asked whether software could listen in a way that mattered. Wabi asks what those same people will build when the intimidating machinery disappears.
The answer may be a new creative medium. It may also be a glorious profusion of weekend planners, private games and machines with an audience smaller than a dinner party. Kuyda seems pleased by both possibilities. In an industry forever promising to change the world, there is something bracing about software content to improve Saturday.