TikTok perfected the algorithm that guesses what you’ll watch. It never built the one thing an algorithm can’t do: a friend who knows exactly what’s so you.
Try the prototype ↓Choose a product (TikTok, CapCut, or Instagram), find a real problem in its design, explain why it exists, and redesign at least one key point, shown as a working, high-fidelity, AI-coded prototype.
I chose TikTok. Not the safe choice. The people grading this use it every day and have strong opinions, but that’s exactly why a sharp insight lands harder here than anywhere else. The rule I set myself: go deep on one defensible problem, not wide on a feed redesign nobody can falsify.
The dominant creative act on TikTok isn’t posting. It’s the send: passing a video to a friend or group chat, “this is so you.”
It started as a personal observation. My friends and I don’t post. We share. We half-compete over who sends the best stuff, “let me see if you sent anything good.” If most users never create, they never get likes, never build an identity on the platform. So what if the thing they are great at, having taste and knowing their friends, became something the product actually saw and rewarded?
This isn’t a hunch anymore. It’s the biggest live shift in social product design, and it happened in the last year.
“That feed is dead. People stopped sharing personal moments to feed years ago… The primary way people share now is in DMs.” Adam Mosseri, Head of Instagram, December 2025
“Share” is a documented For You ranking signal (TikTok, 2020). Instagram’s Mosseri calls “sends per reach” one of its biggest ranking signals. The platforms already know the send is the highest-intent action. They just don’t design for it.
Friend-graph density is a classic retention lever. Facebook’s growth team famously found new users who reached 7 friends in 10 days were far likelier to stay. A denser friend graph is worth more lifetime attention than any single feed tweak.
Every time TikTok reached for “friends,” it built a new social surface, and each one struggled within a year. The lesson isn’t “friends don’t work.” It’s that new graphs tend to die, while enhancements to the existing share act tend to live.
True, a lot of sharing lives in iMessage and WhatsApp. So we don’t try to trap it; we make the sharing that does happen in-app dramatically more rewarding, so more of it comes home.
They built algorithmic shared feeds. Nobody rewards the human who deliberately picks “this is so you.” That’s the gap, covered just below.
Snapchat Streaks are linked to documented teen stress; TikTok shipped anti-doomscroll features in Nov 2025. So this design is positive-only: no dislikes, no streaks, no public rank.
It grows sends (a ranking signal) and friend-graph density (a retention driver), and produces a higher-quality human taste signal. Retention is the business case.
Trade-off I made on purpose: I dropped the “meme-god leaderboard” from my first idea. A public competition is higher-ceiling but fights both the graveyard and the wellbeing evidence. Recognition beats competition here.
Given a year of R&D each, the two biggest companies in social both reached for the algorithm again. Neither honored the deliberate human act of choosing a video for a specific person.
| Instagram Blend / TikTok Shared Feed | For You, From Friends | |
|---|---|---|
| Who curates | The algorithm blends two people’s interests | A real friend deliberately picks it for you |
| The user’s role | Passive co-consumer | Active curator with an identity |
| The signal | “What do these two have in common?” | “What does a friend know you’d love?” |
| Launched | Blend Apr 2025 · Shared Feed Dec 2025 | The layer they both left open |
They asked what the algorithm thinks two friends have in common. We ask what a friend knows about you that the algorithm never will.
TikTok’s inbox is a chat UI wrapped around a behavior that isn’t chatting. It’s mostly videos friends sent you, flattened into gray bubbles and forgotten. We split it into From Friends (a swipeable feed of everything friends hand-picked for you, tagged by sender) and Chats (the rare real conversations). One positive gesture closes the loop.
Lives in the inbox, where TikTok already put Shared Feed, as a “From Friends / Chats” toggle. No new nav tab (that’s the graveyard).
Vertical stays TikTok-native. Right is the one new positive gesture. There is no negative gesture, ever.
Credit is invisible by default. The system only ever sends good news: “Mario’s been vibing with your shares.”
A high-fidelity, AI-coded build of the core loop. Tap through the “what’s new” tour, then swipe up for the next clip, or right for “this is so me,” and watch the sender get credited.
The roughly half who never post get a creative identity as curators. Received shares stop dying. A reason to open the app that’s connection, not doomscroll.
“Send this to someone” finally has a destination. Content that resonates travels through friend networks, a discovery path the pure algorithm doesn’t offer smaller creators.
Grows sends (a ranking signal) and friend-graph density (a retention driver), and yields a higher-quality human taste signal to improve recommendations.
I deliberately chose recognition over competition, positive-only over dislikes, invisible-by-default over public scores, and capped data use at “better experience.” The engagement gains are designed to come from deeper friendships, the healthy kind of retention, not manufactured compulsion.
swipe_right_this_is_so_mevideo_share_sentcaption_added
share_watchedrecognition_notif_openedreciprocal_send
toggle_switchtour_completedbadge_display_opt_in
North star: sends per active user each week. Guardrail: by design, no negative-signal events exist to farm.
The four Phase 2 bullets above, made concrete. Every screen keeps the same guardrails: comparison-free, positive-only, opt-in by default.
Earned by real friends, opt-in to show.
Status without posting. Taste is the identity.
A crown you can lose, so it stays fun.
Every video you two ever traded.
This was built as a collaboration. I brought the product judgment and the calls; AI acted as a skeptical partner that researched, pressure-tested, and built, so I could move from a hunch to an evidenced, prototyped concept fast.
The core insight, the decision to reject a feed redesign, the call to drop the public leaderboard, the inbox placement, the “feed as default,” and every gesture and copy decision.
Five parallel research passes (the graveyard, dark-social data, pain points, competitor teardown), adversarial pressure-testing of my idea, and the coded high-fidelity prototype.
The most useful thing AI did wasn’t agree with me. It was pointing five loaded objections at my first idea until only the defensible version survived.