1:02:53Did AI Kill React Native?
Monday, Sep 14 · 1 h 03 min · about 10 min to read
Theo-brain watched all 1 h 03 min
Shopify is rebuilding its mobile apps as separate native iPhone and Android apps, saying coding agents now make maintaining two apps a minor cost. Theo argues Shopify skipped React Native's real strengths, like over-the-air updates and Expo. He concedes agents are now good at native code, which shrinks the skill gap Meta's engineers gave React Native.
Skim itSkim it for Theo's own takes and porting tips. The notes cover the main points well enough to skip the rest.
Parts
Why Shopify's React Native bet mattered
Shopify's move to React Native, with apps that sometimes outperformed the native versions, was a huge moment for the framework.
some crazy wins. Not only was it performing better than you would have expected being React Native, it was actually often outperforming the native versions in particular with the Android app for their point of sale systems, which was so so cool. It ended up being a huge moment for React Native and
Shopify's case for going native
Theo reads Shopify's article: agents changed the cost math, and the Shop app was rebuilt and published in 12 weeks.
first to be migrated. Assisted by AI, the team was able to go from a proof of concept to a fully rebuilt built app published in the app store in just 12 weeks. They've written about the
Helix and agent tooling
Shopify found phone simulators too slow and manual for agents to test their own work, a pain Theo says is killing mobile.
Simulators are extremely slow and manual to work with with agents. It doesn't matter how good the model is if it can't test its work quickly, which is especially difficult for mobile. Yep, mobile is killing mobile so
What Shopify left out: updates and Expo
Theo argues over-the-air updates and Expo are React Native's real advantages, and Shopify barely used Expo.
is Expo. Expo is kind of the Nex.js/verell of the React Native world. They have built a lot of the tooling that is necessary to make React Native a good experience to work with. And they handle all the builds on their servers if you would like. They handle the OTAA and
React Native's honest downsides
Beginner mistakes, chaotic version upgrades and dependence on Meta. Speed is not on his list.
things like version upgrades. And I'm at the point now where when I have to do a big version bump on React Native or Expo, I delete the package JSON. I delete the package lock and I reinstall and then hope for the best. And when things don't work, I usually delete and
Shopify's quirks
Shopify never moved to React Native's faster new architecture, so its speed comparisons with native are less useful.
good stuff. And the team at Shopify was trying to decide, do they take all their native work they've done already for normal React Native from before the new architecture and port it to adopt this new architecture, or should they just go rewrite it natively? This means that the
Theo's own agent-built native ports
An agent ported his T3 Code app to native more than once, each build in under three hours, which surprised him.
Native. Every one of these builds took under 3 hours automated by the agent running itself in a loop with computer use. Went way better than I ever would have expected and both immediately felt
Living on the agent-maintained port
He uses the agent-built port every day and ships it to testers without reading its code.
don't even [ __ ] I've not read a line of the code for this app to be very clear. And it runs great. It's what I use for coding every day. It is a faithful port
Meta's native experts behind React Native
React Native was great because the best native developers Theo has ever worked with were on Meta's React Native team.
native devs I have ever met, I have ever worked with, I've ever collaborated with were on the React Native team because they loved Native so goddamn much that they wanted the worst JavaScript dev on
Wrap-up: learn once, use anywhere
React Native was never about writing one app for every platform. It let you use React knowledge on mobile.
that's true. None of that matters. It was never write once, run everywhere. It was learn once understand everywhere. Once you learned React, you could learn the platform and combine them. You
Tap any time to jump the video there.
Key takeaways
Shopify's stated reason: agents now do enough of the building, translating between platforms, testing and review that maintaining separate iPhone and Android apps is no longer a deciding cost, as it was in 2020.
What's changed is that agents can now do enough of the implementation, translation, testing, and review work that it's no longer a deciding factor like it was in 2020. React native apps
Theo's concession: React Native's real value was that Meta employs better native engineers than almost any team using it. Agents such as OpenAI's Codex are now good at native code too, which shrinks that skill gap. He would still miss React Native's update and list advantages.
question you have to ask yourself when you are writing an app is who has better native devs? Do you and your team have better native devs or does meta? And I would guess for pretty much everyone watching this video, Meta has hired and employs better native engineers than you do. And as such, you get a lot of
you a hint. It's codecs. Yeah, agents make the gap way way smaller. Previously, you basically needed to use React Native if your team wasn't like the type that go to WWDC or your app would be worse. Now you're just asking your agent anyways.
Theo says over-the-air updates (pushing new app code straight to phones without an App Store review) make it really hard not to use React Native. You can switch a one-day change on and off, or push a small fix, without Apple's long review cycle.
about React Native here. I would argue there are three letters that make it really hard to not use React Native. if you can justify it. Those letters are OTAA stands for overthe-air update. This
Native, you can push up the OTAA for it at the start of the day and you can push up the OTAA to get rid of it at the end. And if something goes wrong during, you can push up a new OTAA with that. And since the JavaScript bundle can be redownloaded and that JavaScript bundle tells Native how to behave differently, you can get around the whole long painful review cycle with small fixes,
The 'React Native apps are bloated' argument is a myth. Going native saved Shopify only about 1 MB. A default Expo app is around 20 to 30 MB, and real bloat comes from images, other media and translations.
size. They shaved a whopping 1 megabyte by moving to native. Wait, what? It's bigger? Yeah. Turns out the React Native runtime is not particularly large. A default Expo React Native app is in the like 20 to 30 meg range. I find that most bloated apps are more so bloated with things like a bunch of localization and translations as well as with just disgusting amounts of assets and media.
Meta has been shrinking its React and React Native teams. Theo says the quality coming out of React Native has plateaued and calls that a real cause for concern, though he doesn't think it's going away.
especially with agents. And the quality of what we get out of the React Native team has kind of plateaued because they're not maintaining it as actively as they did before. And that should be real cause for concern. I don't think
Two skills here are new to you.
Agent port using the existing app as the reference
- What it is
- Have a coding agent rebuild an app on another platform. It gets the original source code, a simulator (a virtual phone on the computer) and the working app to copy, then you correct it where it's off.
- Why Theo cares
- With a working reference, Theo finds the agent gets close to perfect, and when told what's wrong it almost always fixes it on the first try. He keeps his port current by having it pull in changes from the original app.
- How to try it
- Give your agent the existing app's source and a simulator, and ask it to rebuild one screen natively to match.
- When it gets something wrong, tell it exactly what differs. Theo says it usually fixes it on the first try.
- Periodically ask it to pull in changes from the main app and keep the port up to date, as Theo does.
Shopify experienced, that if you give an agent a simulator, some source code, and a clear existing implementation of how this thing should work, it usually can get a pretty close to perfect rendition. And if it fails and you tell it what it got wrong, it will fix it first try almost always. My honest guess here is
Any other changes on main that are worth pulling in? make sure we keep the app up to date with any improvements that we make to the React Native app in the wire protocol for how we actually get data to the client. An hour passes updated and
Expo for React Native New
- What it is
- A toolkit for React Native. One developer sets up a development version of the app, then others scan a QR code and start changing it. Theo says it also makes updates to users easier.
- Why Theo cares
- He says it takes React Native from 'kind of cool' to 'why would I build any other way'. Its simulator and web support let agents iterate without a native build, which he thinks beats Shopify's custom testing tool.
- How to try it
- Ask your agent whether your React Native app uses Expo, and what you'd gain from adopting it.
- Have agents test changes in Expo's web or simulator build instead of a full native build.
with React Native and Expo, you only need one dev to go through that process, and they can make a dev version of the app. And once you have that, you scan a QR code, you change out the React Native code, and now you're working on the app.
Expo is very helpful. Also, on the note of Expo, they have really good simulator support as well as web support, which means you can let your agents do a lot of the iteration without ever needing the native build in the first place. So,
Legend List for long scrolling lists New
- What it is
- A list library by developer Jay. It draws only the rows on screen and is written entirely in JavaScript, so it also runs on the web.
- Why Theo cares
- Theo says it worked out better for his team than Shopify's FlashList and leans less on native code. It's why T3 Code's scrolling is smooth in its desktop and web apps.
- How to try it
- If your app has long scrolling lists that stutter, ask your agent to evaluate switching to Legend List.
list for most projects in React Native. I will say that legend list has worked out better for us and has less native implications. Another thing we'll be
implementation could be ported to web because the whole thing is JavaScript. It is not a native library. All the other list libraries that made it way nicer to have long list in React Native are native code with a little bit of JS. His is JS only because he built the virtualizer in the JS layer so
Advice
If you want a good Android app, the fix isn't a framework. Make someone on your team use an Android phone daily. If you already have an Android user, have them use an iPhone for a while, then switch back and fix your app.
The solution is to make someone daily drive Android on your team. Actually, if you do have an Android person on your team already, you should make them daily drive an iPhone so they can know how a phone's supposed to work and then make them go back to Android and then they'll fix your app. I think I've said all I
When setting up a Mac for agents to build Apple apps, download Xcode (Apple's developer software) directly from Apple's download servers, not from the App Store. Theo says the App Store route is hell.
Xcode being Xcode. Make sure you download from the CDN directly. By the way, don't install from the app store. It will be hell. Really stupid that you
Hot takes
Shopify's Helix checkpoint system and its custom agent-testing tool are 'criminally overengineered', a sign of worshipping their existing codebase.
His reasoning: These are CRUD apps (basic apps for viewing and editing data). The part they made testable without the app's screens is just the data layer, and Expo already solves most of this.
the next one starts. I don't love this approach. This screams to me, we think our current implementation is more novel than it actually is. People worship their existing codebases a little too hard and this feels like OG codebase warship. like this is the only way they
My [ __ ] god, they'll overengineer anything. These are CRUD apps. The thing they are talking about making headless so they can test it and let the agents do it is the API and data layer. I'm sure that
The average React Native developer is worse than the average native developer.
His reasoning: Native is harder, so React Native becomes the default for weaker developers. That doesn't make the technology bad; winning tech always attracts them. But it means its pitfalls show up more.
you guys to know I mean it. The average React Native dev is worse than the average native dev. This is a fact. I will not be arguing. And this isn't because of one being better or worse than the other. This is because I I guess it's because native dev is harder and React Native is easier. Which means
Theo's honest guess is that Shopify's porting struggles came from leaning too heavily on Claude models and not enough on OpenAI's.
His reasoning: In his experience, the model he calls Soul ports an existing React Native codebase to native almost seamlessly.
almost always. My honest guess here is that they were using claude models too heavily and OpenAI models not enough because I have found that Soul's ability to navigate an existing React Native codebase and port things over in spirit to the like actual native build is almost seamless. The reason I so
Apple's own 'native' building blocks can make your app feel worse than a custom one.
His reasoning: SwiftUI's built-in navigation only lets you swipe back from the left quarter of the screen. Apple's own Settings app accepts a swipe from the middle. Users who swipe from the middle will blame your app, not Apple.
official Apple way to do navigation. The twist is that it only lets you use this left fourth for the back gesture and you cannot change the size of the gutter. If you use the native navigation stack in Swift UI, you cannot use the native gesture if you want to support touching anywhere other than that left fourth.
If I swipe from the left, it works fine. But ready for this? If you swipe from the middle, it does too. They're not using the native stack on [ __ ] iOS. the platform itself and the built-in apps aren't using the native navigation
How Theo thought about this
He checked Shopify's claim that agent ports produce unusable code against his own ports. He admitted his app is simpler than Shop, then scaled the result: if one part-time engineer with a loop and a model can do it, a dedicated team should manage.
not going to pretend my app is as complex as shop. But if I can do this as one engineer working on it part-time for fun with a loop and a model, then a company that has dedicated engineers on this should be able to do it fine. I'm
Theo read Tobi's reply: teams often ask him to make the unpopular call because they don't want to be the bad guy. Theo said that flipped him, noting he now tells his own team to let him send hard decisions for them.
They see what needs to happen, but they don't want to be the bad guy. He's happy to come in and make the call if he agrees with the premise. It saves enormous amounts of meetings and change management, etc. Sometimes it's jokingly referred to with as founder mode as a service here. Okay, that flipped me.
discovered this path for things, who has been telling their team more like, "Hey, if there's a hard decision to make here and you're scared to be the one to say it, just tell me what to send and I'll send it." This is that and I have
Sign in to mark what you've watched and keep notes.
Have a question about this one? Ask about this video →
Theo Daily is unofficial and not affiliated with Theo or t3.gg. Written by Claude Opus 5.5 from each video's transcript.