The instinct for a social feature at a learning company is a shared path plus milestones. That is too thin, and it's a trap. It produces something that looks social and only works for people who already have the right friend in the right place.

V1 accepts a constraint, and it's worth being exact about what the constraint is. Course Buddies imports a relationship from outside the app and then does a lot of work, the shared start point, the review range, the second progression record, to manufacture the overlap two real-world friends almost never have by chance. The pairing rate is uncertain precisely because of that: the feature needs two people who know each other and are willing to line up.

Inside the app, that overlap already exists. People on the same course at the same position have shared progress and shared context by default, and there are many more of them than there are friend pairs who happen to align. The real-world social graph is the easy win. The in-app graph is the unlock.

So the direction after V1 is stranger to stranger, not friend to friend. Prove the quiz exchange on its own, on pairs who already overlap, then evolve it into a feature built on recommending or matching people within Duolingo's network rather than importing relationships from outside it. Leaderboards already does the matching half of this every week; what it doesn't do is put two matched people in the same material.

Matchmaking is its own problem. Safety, quality of match, what happens when a pair is mismatched, and whether a quiz from a stranger drives return the way a quiz from a friend does. Naming the direction is not the same as claiming it's solved. There is a cheaper middle rung to try first: friends of friends, or people encountered in a shared leaderboard cohort, some existing weak tie that avoids most of the safety surface a pure stranger would need.

Several things deferred from V1 only make sense once that turn is made. Suggesting a friend who hasn't started the course but learned something adjacent. Ranking pairs against pairs, which is a natural extension of a Duolingo strength and premature until pairs prove they hold. Groups of three or more, which need a different turn model, a different celebration model, and a different exit, and are a second feature rather than a scaled one.

The constraint V1 accepts is the thing V2 attacks. V1 exists to find out whether the exchange is worth building that for.