Tinder Social launched on July 21, 2016, and was discontinued by August 2017[1]. Hey VINA opened its doors the same year, lasted nine, and closed permanently in late 2024 after never raising more than $1.59 million in total funding[2]. Bumble BFF launched in 2016, was spun off into a standalone app in 2023, was rebranded again in September 2025, and as of Bumble Inc's Q3 2025 earnings filing, has not generated any revenue[3].
Three apps. Ten years. The same shape of failure.
All three started with the same hypothesis: take the swipe-and-match primitive that worked for dating, apply it to friendship, and watch it scale. None of them survived their own decade. I want to argue that the failure was not in the marketing, the user acquisition, or the feature set. It was structural. Each of these products was a dating app architecture asked to perform a non-dating function, and the architecture won every time.
You cannot moderate your way out of a structural mismatch.
The shape of the problem
When users of a friendship app start treating it like a dating app, the standard product response is policy. Add a clear statement of intent in the onboarding. Train classifiers to flag romantic language. Build a report button. Hire a moderation team. Suspend repeat offenders. This is the path almost every consumer social product has taken when usage drifts from designed intent.
It does not work, for three reasons.
First, romantic drift is not a discrete violation. Two people who meet on a platonic platform and develop romantic feelings have not broken any rule. Apps cannot moderate trajectory. They can only moderate explicit, identifiable harm, and the slow pull from "let's get coffee" to "let's get coffee, just us" is neither.
Second, the cost structure is wrong. Moderation scales linearly with user volume. Every new user is a new potential incident, a new ticket queue, a new edge case. Architecture scales at zero marginal cost. A design decision made once handles every user from then on. If you choose the wrong axis, you are building a business that gets more expensive to operate as it grows.
Third, moderation fights the user. People do not experience their behavior as a violation; they experience it as a natural use of the product. Telling them they did something wrong does not change the product. It just adds friction with the people you have already convinced to show up.
Moderation is downstream of design. By the time a violation reaches a moderator, the product has already failed.
What architecture means
Persuasive technology research has had a clean answer to this for over two decades. B.J. Fogg's behavior model, published out of the Stanford Behavior Design Lab he founded in 1998, describes behavior as the product of three factors: motivation, ability, and a trigger[4]. If you want to prevent a behavior, you can attack any of the three. Motivation is the hardest to move and the most expensive to maintain over time. Ability and trigger are structural. Change them once and the behavior changes for everyone, automatically, from then on.
Wikipedia keeps people honest not by lecturing them about honesty but by making reverts a single click. Calorie labels do not moderate eating; they restructure the decision surface. Bicycle helmet laws are policy. Bike lanes are architecture. The lanes win every time, because they change the default action, not the ethics of the actor.
A friendship app that wants to keep romantic drift out has the same choice. It can spend its operating budget on moderation, paying continuously to fight a behavior it has architecturally invited. Or it can change the product so romantic drift is the harder path and platonic activity is the easier one. The cost of the second approach is paid once, in design. The cost of the first is paid forever, in ops.
Four structural layers
HAOT is built around four architectural decisions. Each is non-negotiable at the product level. Each exists to make platonic intent the path of least resistance.
The first is that there is no 1:1 messaging before a shared activity. You cannot DM a stranger on HAOT. The interaction surface is an activity plan, not a person. If you want to talk to someone, you join the group chat for the specific event they are hosting or attending. Conversation arises from a shared real-world commitment, not from a profile pair.
This single decision dismantles the core mental model that dating apps train into their users. The dating-app primitive is two people in a private chat after a match. The HAOT primitive is three or more people coordinating a plan in the world. Everything else flows from this.
The second is group minimum. Every activity requires three confirmed attendees before it activates. You cannot be in a chat with only a host. The system enforces this structurally, not as a rule but as a precondition. Three is not a guideline. It is the floor at which the chat exists at all.
The third is that the discovery feed has no profile photos. What you see scrolling is the activity itself: time, place, host reputation, group size, what the plan is. Faces appear only after RSVP confirmation. This removes the appearance-based filtering that turns every social product into a dating product the moment you see the photo grid.
The fourth is real-time intent classification on chat. Stream Chat fires webhook events on every message in every activity chat. A lightweight classifier scans for the linguistic markers of romantic intent: compliments on physical attributes, requests for personal contact info before the activity has happened, suggestions to peel off into a private conversation. The classifier does not punish. It nudges. A short, contextual prompt reminds the user that the chat is for the activity, and that they are welcome to make a different plan after.
The point is not that these four decisions make romance impossible. Two people who meet at a HAOT climbing event and want to exchange numbers can do that. The system does not moderate the world. It moderates the surface. What it does make impossible is the slow architectural pull toward dating-app behavior that every previous friendship product has been unable to escape.
The counterarguments
The strongest counterargument is that we are just creating friction, and friction does not eliminate behavior, it displaces it. This is correct. We are not claiming that no romantic interaction will ever happen on HAOT. We are claiming that reducing a meaningful percentage of unwanted behavior at zero ongoing moderation cost is a better unit economic than reducing a higher percentage at continuous ops spend. Architecture compounds. Moderation degrades.
The second counterargument is that 1:1 messaging is a core social primitive and removing it costs us users. Yes. The users we lose are the users who would have used the product to date. We do not want them, and acquiring them would actively damage the experience for the users we do want. A friendship product whose addressable market is "people not looking to date right now" is a fundamentally different business than a friendship product whose addressable market is "everyone, but please be platonic." The first has structural defensibility. The second is a moderation cost center waiting to happen.
The users we lose are the users who would have used the product to date. We do not want them.
Why this matters
The WHO Commission on Social Connection published its flagship report on June 30, 2025, finding that approximately one in six people worldwide are affected by loneliness, and that social disconnection is linked to roughly 871,000 deaths annually, or around 100 deaths per hour[5]. The U.S. Surgeon General's 2023 advisory put the individual mortality impact at the level of smoking up to 15 cigarettes a day[6].
This is the category we are building in, and the consumer response so far has been wellness theater. Loneliness chatbots. Mindfulness subscriptions. Algorithmic feeds that promise to surface "people like you" but route you back into the same swipe loops that have not produced friendship at scale in the decade since they were first attempted.
The market does not need another dating-app architecture asked politely to behave like a friendship app. It needs a product built from the ground up around the unit that actually makes friends in the offline world: a real plan, at a real time, with a real group of people.
That is what we are building.
The next time someone asks why HAOT removed 1:1 messaging, or why there are no photos in the discovery feed, or why every activity needs three confirmed attendees before it goes live, the answer is the same one every time. We are not trying to convince our users to behave platonically. We are trying to make our product structurally incapable of becoming a dating app, no matter how many users come through it.
That is what anti-dating as architecture means. Everything else is policy.
Sources
- [1] Bustle, "Tinder Social Has Been Discontinued, So Here's What That Means For The Tinder App." August 2017. Link
- [2] Hey VINA, Instagram farewell post. Late 2024. Link
- [3] Bumble Inc, Q3 2025 Earnings Fact Sheet. November 2025. "As of September 30, 2025, BFF app has not generated any revenue and therefore is excluded from our key operating metrics." Link
- [4] B.J. Fogg, Persuasive Technology: Using Computers to Change What We Think and Do. Morgan Kaufmann, 2002. Stanford Behavior Design Lab. Link
- [5] WHO Commission on Social Connection, From Loneliness to Social Connection: Charting a Path to Healthier Societies. June 30, 2025. Link
- [6] U.S. Surgeon General Vivek Murthy, Our Epidemic of Loneliness and Isolation: The U.S. Surgeon General's Advisory on the Healing Effects of Social Connection and Community. May 2023. Link