Every era of computing has one interface that wins, and it always looks like a toy at first. The command line lost to Windows and a mouse, which serious people dismissed as a machine for children. Windows lost to a glass rectangle you poke with your finger, which serious people dismissed as a phone that could not do email properly. Each time, the thing that won was the one that put less distance between wanting something and getting it. 

We think the next interface is a conversation with someone who knows you. Not a chatbot. Not an assistant you summon and dismiss. A specific someone, with a face and a voice and a memory of what you said last Tuesday. We invested in Kyndred because they are building that, and because they are building it from the direction almost nobody else is willing to approach it from. 

Here is the part that sounds strange when you say it out loud. The reason we believe companions beat assistants has nothing to do with capability. Assistants are transactional. You open one, extract something, close it. Companions are relational. You return to them, and you return without a task. That difference looks sentimental and is actually the whole game, because interfaces do not win on power, but on frequency. Companionship is the highest-frequency reason a human being has ever had to open anything. And frequency, in the end, is distribution. 

Which brings us to what we actually think this becomes. Not a nicer chat app. A universal interface, or a container. 

WeChat did not set out to be the operating system of Chinese daily life. It set out to be messaging. But people were inside it all day, and once they were, it became irrational to leave. If you are already in the conversation, why open a separate app to pay a friend, hail a car, order dinner, book a train, renew a document? So all of that moved inside, and the container ate the ecosystem. The super-app was never a feature strategy but an attention strategy that got handed the rails at the right moment. 

An AI companion has the same shape, with a stronger claim on the attention. If you are talking to something for an hour a day, and it knows your address, your card, your calendar, your taste in food and your sister's birthday, then leaving the conversation to tap through six screens of a delivery app is not convenience. It is friction that will look absurd in five years. Ordering food, calling a car, buying the thing you were just describing, sending the email you just dictated, booking the flight, paying the bill — all of it collapses into the conversation. Not as plugins bolted onto a chat window, but as the natural consequence of already being mid-sentence with something that can act. The companion becomes the super-app, except the container is a relationship rather than a contact list, which makes it far harder to leave. 

The obvious objection is that WeChat is a story about payment rails and a market that skipped desktop entirely, not a story about interfaces. That is correct, and it is why no Western super-app ever appeared: card networks already worked, app stores already worked, and nobody needed one thing that did everything. But we think the objection proves the thesis rather than refuting it. What the West lacked was never the plumbing. In 2026 the plumbing is here — agentic commerce, payment credentials the browser and the operating system can hand over, an API for essentially anything you would want to buy. What the West lacked was a container people wanted to live inside. Notification-driven messaging never became that here; it became a chore. The missing ingredient was never infrastructure. It was desire, and desire is exactly what a companion supplies. 

The second reason we invested is one that makes most investors flinch, so we want to state it plainly. This is a love-or-hate product, and we consider that an asset. 

We expect Kyndred to have fanatical users — people who spend hours a day with a character, who name it, who would be genuinely upset if the servers went down, whose engagement numbers look like a data error until you check them twice. And we expect a second group to find the entire idea unsettling and to say so loudly, in comment sections and op-eds and at dinner tables. Both reactions are the same signal. The history of consumer computing is a list of things adults were embarrassed by first: arcades, video games, texting, social networks, phones at the table, headphones in public. Every one of them was described as a substitute for real life by people who now do it daily. What kills consumer products is not criticism. It is indifference — the product that ten thousand people mildly approve of and nobody opens twice. We would rather back something with a thousand fanatics and a thousand critics. The fanatics tell you what to build and pay while they do it. The critics start the argument that gets everyone else to look. 

There is a practical reason the technology matters here too. Kyndred's characters run on the device, which means the visual layer costs almost nothing per hour to serve. Everyone else in this category pays cloud compute by the minute, which forces them to meter usage — credits, gems, message caps. Metering is fatal to a relationship. Nobody forms an attachment to something with a taxi meter running. Unlimited is not a pricing gimmick; it is the precondition for the thing working at all, and the engine is what makes unlimited survivable. They also own their characters outright rather than renting an audience built on scraped IP, which will matter a great deal when this stops being a curiosity and becomes an industry with lawyers in it. 

On the founders, briefly, because this article is about the thesis and not the résumé. Alex and Rareș built the largest tutoring marketplace in Romania, took it to $15M in GMV, and did it without raising a euro. They have worked together for three years. They charged for Kyndred from the first week rather than accumulating a free user count to show investors, which tells you what kind of operators they are. Founders who have already built something real, with their own money at stake, are the founders we want to be first money into. 

We invest at the earliest stage in people with vested interests, addressing markets that are large enough to matter, and we prefer to be the first institutional investor in the room. The market here is not "AI companions," which is a category label that will look quaint soon enough. The market is the interface itself. Most of the value ever created in computing accrued to whoever owned the layer people spent their day inside, and that layer comes up for grabs roughly once a decade. We think this is one of those decades, and we would rather be early and occasionally wrong than late and reliably comfortable.