Hey,
It’s Michał with the August dispatch from the React Native community and Callstack Incubator.
React Native 0.87 landed with a stricter public API and much faster source maps. Oxc joined the React Compiler rollout. Expo opened a waitlist for cloud simulators built for agents. Meanwhile, the Incubator team shipped across Rozenite, agent-device, React Navigation, Apex, and React Native core, Apple released iPhone Duo, and OpenAI disrupted some 3D industries with Astra.
There is plenty to unpack, so let’s get into it.
Oh! And this week we're hosting Agent Conf in Warsaw, Poland. Hope to see you there! If you don't have tickets yet, grab as many as you need with a special 30% discount for newsletter readers. Use the code: Late30.
Margelo joins Callstack
We’re excited to share that Margelo is joining Callstack. The team behind Nitro Modules, VisionCamera, and react-native-mmkv brings deep performance-first React Native engineering and open-source expertise to our work.
Margelo’s libraries will remain open source, and Marc Rousavy will continue leading their development while helping shape what we build together. With architecture, native integration, validation, and release safety carrying more weight in an agentic development world, we’re bringing together the people and technology to take on harder production problems.
Community highlights
- React Native 0.87 is out: The Strict TypeScript API is now the default, Metro generates source maps twice as fast while using half the memory, and iOS developers can try experimental Swift Package Manager support. The release also adds Android Gradle Plugin 9 support and raises the minimum Node.js version to 22. The team also landed the first RC of 0.88 with long awaited improvements to Platform-level dead-code elimination
npx react-native spmcommand and more. - React 19.3: View Transitions and Fragment Refs are now stable, RSC can render
<Context>, new React DOM APIs:browserto opt-out a subtree from SSR (you can drop thetypeofchecks now) and Trusted Types web API integration that prevents some XSS attacks. - The React Compiler support in Oxc: Oxlint can now report React Compiler diagnostics, while the new transform package brings automatic memoization to Oxc’s Rust toolchain. With very little effort we migrated a bunch of our projects from ESLint and Prettier to Oxlint + Oxfmt for faster feedback loops. Highly recommend! Support is also moving into Vite, Bun, and Biome.
- Expo announced cloud simulators built for agents: EAS Simulator gives coding agents an on-demand place to build, run, inspect, and verify mobile apps remotely. It uses
agent-deviceor Argent under the hood. And if you want to self-host simulators on your own infra, tryagent-device proxy(until we ship a better option for you 🤙.) - Shopify raised mobile end-to-end test stability to 98%: The team replaced years of Appium patches with an opinionated wrapper and moved away from implementation-level test IDs toward computer vision. The result is a useful case study in treating test reliability as a product with clear ownership.
Callstack Incubator highlights
August was a month of turning agent experiments into infrastructure that scales. It was also fun.
OSS Retreat
Every year around August/September we gather people contributing to open source software together and meet for a hiking and hacking session, a.k.a. OSS Retreat. This year was no different. We traveled to Szklarska Poręba in Poland and explored Karkonosze National Park.
It was not only fun, but quite productive as well! We shipped improvements to Apex, created a new e2e testing library powered by agent-device, optimized animations in React Navigation, hacked some simulator controls, and much more.
Apex
The team is preparing for general access rollout and experimenting with different underlying LLMs powering the system. The GA will include API access and integrations with OpenCode, Claude Code, Codex, Pi, and others. As an early tester I’m quite happy with the results. Already shipped 20+ PRs to agent-device codebase (not much React Native code there, lol) and it confidently replaces workloads I’d normally delegate to Sonnet 5, GPT-5.6 Luna Max or even Opus 5. The team says Apex is now at GPT-5.6 Terra level for general tasks. When its insights and outputs are cross-checked by a frontier model, I’m confident in shipping with it.
Still gaining trust with how far I can push my trust to its judgement. The best part is that it’s basically unlimited to use of tokens for me because we have it deployed in Callstack infra at stable cost. We’ll share more during Agent Conf.
Let me know if you’d like to self-host a model like this. It’s a transformative experience for the organization like ours with about a hundred of devs using it daily.
Rozenite
Rozenite shipped a broad UI and developer experience refresh, a new asynchronous RPC protocol, shared UI primitives, and safeguards that keep plugins out of production bundles. The team also added in-process protocol testing and a terminal tap for streaming plugin messages with filters and JSON Lines output.
-1280w.avif)
Rozenite now carries its own agent skills and measures how agents use the tool. That gives us a way to cut unnecessary steps, tokens, and latency. We also shipped Lynx support.
agent-device
We’ve spent August tightening the loop between an agent action and evidence from a real app. Kudos to the Software Mansion team who released AppControlBench which pushed the tool to achieve over 94% accuracy and the cheapest “tasks per dollar” score. We’re now focusing on faster simulator snapshots and lower cold-start latency, we don’t want to be slower!
We’re also extending the capabilities of agent-device proxy, so you can self-host mobile agent infrastructure for iOS and Android devices where it’s most convenient. A Mac mini can become a small iOS simulator cloud for an enterprise lab, a team, or a home setup. The goal is simple: give agents repeatable access to mobile runtime evidence without making every developer assemble the infrastructure from scratch.
We’ve also seen Expo SDK team adopting a PR review agent that uses agent-device under the hood. Steal it!
React Navigation
We’d like to start with some benchmarks from Andrei Calanzas who measured the performance of the most popular navigation solution for React Native. We’re super proud React Navigation took second place right after react-native-navigation from Wix, even though Satya is adding features like there’s no tomorrow. Provided they don’t hurt performance, which, from what we can see, they do not (at least not much.)
In other news, React Navigation kept closing compatibility gaps for concurrent rendering and the Strict TypeScript API, while adding lazy screen getters, long-press peek support in drawers, and improving URLs and deep links expressiveness.
It's also worth mentioning that the Lynx team published an initial React Navigation package, and a GTKX demo showed React Navigation (same API, different implementation) running in a React app on Linux.
The Migration Agent
Hur is experimenting with a Migration Agent to run any app used an open-source Android RSS reader as a reference and began recreating it through an agent-driven workflow.
React Native core contributions
Artur’s changes required for React Native’s iOS 27 scene lifecycle landed on main in August. This unblocked work that depends on modern scene behavior, including new Apple platform APIs. It also exposed follow-up work around dimensions because scene-based apps can be resized and cannot rely on values that were treated as constants.
We’ve also onboarded Marcin as a new member of the CSS in React Native movement, taking over after Kamil. On this front I’d like to thank Kamil Paradowski, whose efforts brought to you CSS masking, filtering, and soon calc behavior to React Native. It’s been a pleasure to work with you and we wish you the best on the new path!
And now that iPhone Due exposed many challenges that React Navigation has to face with the core React Native layer, Satya will join the platform support team in the React Foundation to make it work well by default without hacks.
Blog
- React Native in .NET MAUI: A Practical Brownfield Integration Guide by Jouzas Petkelis, Marcin Szalski, Artur Morys-Magiera and Hur Ali :**Learn how to embed React Native inside a .NET MAUI app with a brownfield setup. This guide covers runtime packaging, lifecycle wiring, navigation, and shared state across both runtimes.
- Working With the React Native Super App Showcase Repository by Bartłomiej Krok and Szymon Chmal:** Explore the React Native Super App Showcase architecture: Re.Pack, Module Federation, runtime bundles, shared dependencies, and production releases.
- Migration to React Native in 2026 Starts With a Delivery Question by Michał Pierzchala: Choosing between brownfield and greenfield React Native migration? Learn where risk sits, when to use a checkpoint, and what to prove before committing.
- Automating Mobile QA With Cloud Agents by Michał Pierzchala: Code review and green CI checks do not prove a mobile app still works. Learn how cloud agents, agent-device, and runtime evidence can verify app behavior before merge.
- Automating React Native Evals With Airflow by Lech Kalinowski and Artur Morys-Magiera: How Airflow coordinates React Native Evals, runs solver and judge models separately, and keeps every result traceable through human review.
Podcast
- VisionCamera V5 With Marc Rousavy with Marc Rousavy, hosted by Mike Grabowski: Marc Rousavy joins Mike Grabowski to unpack VisionCamera v5, Nitro Modules, the constraints API, migration work, and what changed for React Native camera apps.
Webinars and live streams
- How Graph Agentic Coding Extends AI Workflows with Kewin Wereszczyński and Lech Kalinowski: How connected state, retrieval, memory, and verification carry coding agents across complex, multi-step workflows.
- What Apex shows about React Native AI models with Kewin Wereszczyński and Szymon Chmal: See how Apex, Callstack's React Native model, compares with frontier models on navigation, animations, async state, and performance-heavy UI work.
Agent Conf
Agent Conf is happening this month and OpenAI, Meta, Amazon and SpaceXAI are all sending their engineers to Warsaw!
Join us in Warsaw on September 17-18 for two days of talks on agentic development. Recently announced sessions include:
- Build an Agent with Eve: The Open Agent Framework by Dom Sipowicz
- Quantization Impact On LLM Performance: A Case Study of Apex by Lech Kalinowski and Artur Morys-Magiera
- How will we prompt AGI? A History of Harness Hacks by Ivan Charapanau
- Building a Software Factory with Cursor, xAI, and Grok Bot by Nick Miller
- Goodbye Slop; Welcome Determinism by David Khourshid
We will also hold a Panel Discussion about the Future of Software Development. More info soon!
Last ticket are available at agent.sh/tickets. We got you a 30% last minute promo code - use LATE30 at checkout.
Explore the full agenda here: https://www.agent.sh/agenda
That’s it for the August edition. We’ll be back next month with more updates, experiments, and lessons from the field.
See you then,
Michał
