A platform no one is forced to use

The math that defines my job at YouTube: my platform team is a handful of engineers, and the standards we set (how navigation works, how logging works, which shared components you build on) govern how hundreds of engineers across the main app and the partner verticals write code every day. I have responsibility for all of it and authority over almost none of it. Nobody has to adopt our library. They can fork the shared components instead, and several teams already had.

You feel the cost of that in the architecture. YouTube’s navigation grew organically for years into a tangle of native screens, cross-platform UI, and webviews, stitched over a custom back stack the Android OS can’t see. That invisibility is the quiet killer: when the operating system has no standardized view into your internal navigation, you can’t adopt the platform features that ride on it, like predictive back gestures or consistent transitions, and every team that hit the wall solved it locally, re-implementing the same screens over and over. Fragmentation isn’t an aesthetic complaint. It is watch-time attribution lost to a brittle logging path and launches slipping because no two surfaces agree on what a ‘page’ is.

The instinct, when you own the standards, is to mandate them. I can’t, and I’ve come to think I shouldn’t even want to. The line I keep in front of me is the failure mode, not the goal: if the surfaces stop trusting the platform, they start building their own solutions, and once they do, you don’t have a platform, you have a library nobody depends on. The navigation API is where that trust either holds or doesn’t.

Authority gets you compliance until the next reorg. The only adoption that survives is the kind people chose because the thing was good.

So the work is mostly persuasion with proof underneath it. Interview the partner teams until you can name their pain in their own words. Run a summit, pull a real consensus out of the room, and write down the open questions you are explicitly not going to let block this year. Commit to a narrow scope with loud non-goals (this is a reusability framework verticals link into, not a plan to swallow them into one binary) because the fastest way to lose a skeptical partner is to confirm the land-grab they already feared. Then ship something small that demonstrably works, so the next conversation starts from evidence instead of a slide.

What I haven’t resolved is how patient this model is allowed to be. Adoption-by-trust is durable and it is slow, and slow has a price measured in the fragmentation that keeps compounding while you build consensus. There are mornings I’d trade some of the durability for a mandate, and I’m fairly sure that trade is wrong — that a platform forced on people quietly rots the moment the mandate lifts. I believe the patient version is the real one. I just can’t yet prove it moves fast enough to win.