A few years ago we tapped an NFC tag on a parking meter and a little card slid up: pay for two hours, Apple Pay, done. No App Store, no signup, no 80 MB download. Everyone on the team agreed it was the future. Then somebody asked how you'd find that same meter again tomorrow, and the room went quiet.

That silence is the whole story of App Clips and Android Instant Apps.

The half they got right

Getting rid of the install wall is a real engineering achievement. The OS fetches a small bundle, runs it in a constrained context, and lets you pay with credentials you already have. For a one-off task — a scooter, a parking meter, a cafe menu — that's genuinely better than a native app, and the teams who built it should be proud of it.

But the design assumes you arrive through a physical trigger or a link someone handed you. There's no browsable shelf, no search inside a host, no way for the merchant to reach you after the fact. Every session starts from zero. A product that can't be found again isn't a product with users; it's a product with demos.

DimensionApp Clips / Instant AppsMini-program inside a host
EntryNFC tag, QR, Safari or Maps linkSearch, share cards, QR, chat, profile
Coming backRe-scan, re-find the linkRecent list, follow, subscription message
Merchant presenceNothing persistentA storefront with menu, hours, bookings
DiscoveryYou must already know it existsThe host can surface what you don't know

What the host adds

Look at what a super app quietly contributes once it decides to host mini-programs. A catalog someone maintains. A search box that sees inside apps, not just their names. A share card that carries context into a group chat. An identity and payment rail the developer inherits instead of rebuilding. A notification the user opted into, so a booking can remind you an hour before.

None of those is a runtime feature. They're distribution features. That's why two products with equally good no-install technology end up with completely different retention curves.

We've seen teams treat this as a platform war and pick a side. It's more useful as a checklist: if your lightweight app has no shelf, no search, and no legitimate way to bring someone back, you've shipped the easy half.

The tradeoff nobody likes

A host that gives you discovery also gives you rules. Review queues, category restrictions, a cut of payments, and the constant risk that the host builds your category itself. An open catalog only works if someone does the unglamorous governance — spam control, quality bars, dispute handling. Open doesn't mean absent.

There's also a real cost in targeting more than one host: adapters, API differences, a lowest common denominator. We'd rather pay that than rebuild the growth loop from scratch in every market, but we won't pretend it's free.

Where this leaves us

If you're building a lightweight experience today, don't stop at "no install." Ask the less flattering question: what happens on day two? A mini-program that lives in a catalog with search and a reason to return is a different business from a clip that only works when someone is standing in front of a sticker.

The no-install trick stopped being the hard part years ago. Building against a runtime that can serve several hosts at once is how the same little service gets a shelf wherever it lands, instead of waiting for one platform to invent one.

Sources