Merge, exposure, rollout, and recovery are separate controls. A feature flag can limit exposure. OTA can fix some compatible JavaScript changes. A native failure may still need a new store release. React Native at Scale 2026 maps the boundary of each 👉 clstk.com/4hRgWa6

Join Callstack, @jetbrains & @expo for From Design to Shipped: A Modern App Development Meetup on October 22, 2026 at the JetBrains Amsterdam office. You’ll here from: 🔹 Michał Pierzchała (@thymikee): “From Design to Done in One Shot With a Cheap Model” 🔹 Craig Walker & Kasper Peulen (@KasperPeulen): “Agents Didn't Kill the UI Developer Workflow - They Changed It” 🔹 Cedric van Putten (@cedric_dev) from Expo Doors open at 5:45 PM CEST. Free entry, pizza & drinks after the talks. Sign up now: luma.com/i8wkxgyg

Meta just announced new ways to build for Meta VR devices, and React Native is part of that story. The new Meta VR Layout SDK brings spatial, multi-panel experiences to React Native, while the new Expo Launch integration makes the path from an Expo project to Meta's Release Channels and Store much simpler. We've been working with @Meta on React Native support for VR, and today we're updating our developer documentation alongside the announcements. More coming soon in our React Native reference app as we add examples for the new spatial APIs. Read more 👉 clstk.com/3VvaKMc

Open source has been part of our DNA since 2016. To give back to those who shared their RN experience for RN at Scale 2026, we donated $1.3K to hot-updater, lerna-lite, changesets, release-it, react-native-maps, and react-native-keychain. Thank you to everyone maintaining them 💜

We deliberately added an invalid commit reference to an agentic workflow. The Reviewer rejected it, the Developer corrected it, and the Reviewer verified the fix before human approval. A controlled recovery loop sent verified work forward. Read more about it👇

Teams can share plenty of code and still run iOS and Android as separate products. Where does new mobile work land by default, and who owns the native exceptions? React Native at Scale 2026 looks at how experienced teams make that call 👉 clstk.com/4hpke31

We’re excited to share this one: the Agent Conf talks are live on YouTube. A huge thank you to every speaker who helped us shape this conference. Watch Agent Conf on YouTube 👉clstk.com/4hIGiG7 Which talk are you starting with?

On the first day of @AgentConf we opened free access to our model Apex to all attendees. Apex achieves SOTA in React Native and Terra Max / Sol Medium in Artificial Intelligence Index. Enjoy unlimited tokens on us til the end of the weekend and show us what’ve shipped!

Callstack’s @LeSiOO and @artus9033 shared a behind-the-scenes look at how #Apex was created. Recording available soon.

Pop by the Callstack booth at @AgentConf for tech talks and good vibes.

Find us here 👇

We’re hosting @AgentConf tomorrow! Stop by the Callstack booth to talk to us. Our Apex team has deployed a self-hosted model to 200+ engineers. Ask them about open-weight models, or explore mobile agent workflows with agent-device via CLI or MCP. Nine months into fully agentic development, we’ve got plenty to share. Venue map with our booth marked in the replies.

Find us here 👇

5x to 10x WebGPU speedups. @huggingface is working on its own inference engine for browser AI. The approach uses an agentic system to write custom WebGPU kernels, merge operations, reduce memory reads, and optimize execution for specific devices. The work is still in an early phase. @nicodotdev joins @grabbou to explain what the team has tested and what this could change for Transformers.js. Watch the full episode of React Universe On Air 🎙️clstk.com/4yBBdpA

A React Native migration for a crypto trading platform gave coding agents project-specific skills, access to simulators and devices, and React DevTools context. This made engineering standards reusable across teams. Read the case study to learn more at clstk.com/4gVi4b2

An agentic SDLC routed verifiable tasks to autonomous agents and kept judgment-heavy work in an engineer-guided Codex loop. The result: under one month to production, 80% estimated agent-executed effort, and up to 3x estimated acceleration. Learn more at clstk.com/3V2Hts1

How much AI can you run in the browser today? @nicodotdev from @huggingface joins @grabbou to explain how Transformers.js brings pretrained models into JavaScript for text, audio, image, and agent workflows. The conversation covers ONNX Runtime, WebGPU, model downloads and browser caching, CPU fallbacks, performance across devices, local speech recognition, and background removal. They also discuss browser agents and Hugging Face's early WebGPU inference engine, where current experiments show 5x to 10x speedups. Chapters: 0:43 Welcome to React Universe On Air 1:04 Nico Martin's work at Hugging Face 2:04 What Transformers.js is 3:30 Browser AI beyond large language models 6:00 Bringing Python ideas into a JavaScript API 10:08 How the Transformers.js pipeline API works 14:23 Choosing compatible models and architectures 15:46 Why Transformers.js uses ONNX Runtime 19:01 CPU inference, WebGPU, and browser coverage 22:57 Model downloads, browser caching, and UX 28:28 Startup time and performance across devices 30:57 Why Transformers.js focuses on the browser 34:07 What developers misunderstand about browser AI 38:44 Building AI agents in the browser 43:19 Local fallbacks and hybrid AI 44:28 The browser AI roadmap 47:04 Hugging Face's custom WebGPU inference engine 50:57 Why speedups matter on constrained hardware 52:07 Nico's favorite browser AI use cases

Background removal and transcription run in the browser. @nicodotdev joins @grabbou to cover Transformers.js today, device limits, offline use, and which parts of an agent workflow can already run locally. 📺 YouTube premiere: piped.video/elAVSra9KCY 📆 Sept 7 at 2PM CEST

React Native can unify the codebase without unifying how mobile work is owned and delivered. Our new field guide, React Native at Scale 2026, combines 10 years of our experience with perspectives from 12 organizations. Use it to review your setup: clstk.com/4qUkWJS

We’ve been really happy to see the response to @margelo_com joining Callstack. 🥳 Over the past few days, we’ve read many thoughtful messages and reactions from across the React Native community. Thank you for all the kind words.

We wrote up how we embedded React Native inside a .NET MAUI app using a brownfield setup. The article covers packaging, lifecycle wiring, MAUI Shell navigation, and shared state across both runtimes. 👇

🎉 @margelo_com Joins Callstack to Advance High-Performance React Native Engineering Vienna, Austria. 1 September 2026. Callstack, a leader in React Native and cross-platform development, today announced that Margelo, the Vienna-based high-end engineering company behind Nitro Modules, VisionCamera and react-native-mmkv, is joining Callstack in a transaction valuing Margelo at more than EUR 20 million. Margelo builds high-performance technology for React Native. Nitro Modules (34M downloads) lets developers write fast, type-safe native modules in C++, Swift or Kotlin, and quickly gained widespread traction across the ecosystem. VisionCamera (40M downloads), Margelo's enterprise-grade open-source Camera framework and react-native-mmkv (67M downloads), a key-value storage library, power countless React Native apps today. Callstack has spent a decade helping large organizations adopt, scale and operate React Native, contributing to the framework and its open-source ecosystem along the way. Together, the two teams give clients the strongest React Native performance expertise in the market, backed by delivery proven in complex enterprise environments. Margelo's technology and Callstack's delivery experience will now be available to the same clients. The timing reflects a fundamental change in how software is built. AI is making implementation faster and cheaper, but architecture, performance, native integration, validation and release safety are becoming more important, not less. These are the fundamentals on which both companies have built their reputations and on which they will continue to build together. "Margelo has built some of the strongest performance technology in React Native. Callstack has spent a decade helping large organizations adopt and scale the framework. Bringing those strengths together gives us the team and reach to solve more of the hard problems that determine whether software succeeds in production," said Mike Grabowski, CEO and Founder of Callstack. "At Margelo, we’ve built high-performance technology that developers already rely on. We’re excited to partner with Callstack to bring it to more clients and scale its impact. As software development becomes increasingly agentic, we want that technology to provide the foundation teams need to build reliable software with agents," said Marc Rousavy, CEO and Founder of Margelo. Margelo's existing open-source libraries will remain open source. Marc will continue to lead their development and help shape the combined company's offering. The transaction follows Callstack's 2025 growth investment and supports its strategy to expand technology and open-source capabilities while growing internationally.

Registering a native module twice can crash a React Native app. Our Super App Showcase prevents that by keeping native code in the Host Shell. Mini Apps stay JavaScript-only and load when needed with Re.Pack and Module Federation. Read more about it: clstk.com/4gtAssJ

In our latest livestream, we covered Apex, Callstack’s specialized model for React Native, built for navigation, async state, animations, and cross-platform constraints. Catch up here: clstk.com/4zD3zRA Then join us on August 27 at 5 PM for our next livestream on graph engineering.

Native iOS and Android engineers are still needed in React Native apps. Most code can live in TypeScript, but teams still need to dive into platform capabilities for better UX. What changes is the focus: native engineers solve complex platform challenges, while RN devs stay productive shipping features for 2 platforms.

One request now starts the automated React Native Evals run. Airflow runs each solver/judge pair separately. MLflow keeps the run record and files together. Human review still decides what gets published. Here’s how we built it 👇

We keep migration knowledge in six versioned files: MIGRATION.md, SCREENS.tsv, STATE_AND_STORAGE.tsv, DEPENDENCIES.tsv, EVENTS.tsv, and PARITY_CHECKS.md. Teams review them in spreadsheets and Git. Agents use them for each task. Download the guide 👉 clstk.com/4quxRSF

Give agents two migration tasks: 1. Preserve native behavior 2. Refactor into idiomatic React Native A different agent reviews between passes. The team compares both apps on devices and repeats the checks after refactoring. Download the guide: clstk.com/3So6LzM

Last night in Cambridge, around 80 people joined us for Agentic AI in the Wild with @Qodo Dev Community, @ReactVisionXR, and Codex Community Boston. Talks covered code quality, local validation, spatial computing, and large codebase migration. Thanks to everyone who came!

On the podcast, @mrousavy breaks down a problem teams run into once camera work gets more advanced: HDR, 5K & 120 FPS may all be supported, but that does not mean those settings will work together. Full episode: clstk.com/4igYCI2

See you tonight in Cambridge for Agentic AI in the Wild with @Qodo Dev Community, @ReactVisionXR , and Codex Community Boston. Doors open at 5:45 PM. On the schedule: 🔶 Oliver Edis on "Prompt to Portal: How AI Makes Spatial Computing Accessible" 🔶 Nnenna Ndukwe on "How to Embed Code Quality into Your ADLC with Codex" 🔶 Matt Hargett on "Fuzz First, Farm Later: Local Validation Before Costly Cloud" 🔶 Rafael Mendiola on "Large Codebase Migration with Loop Engineering" Then a panel discussion, pizza, drinks, and time to talk. See you there!

Agent-driven QA becomes part of delivery when it changes a real merge decision. The webinar we held last month walks through the review surface, QA-agent setup, and a live pull-request demo. See the workflow, live demo, and Q&A in the recording: clstk.com/3TWcGfY

Boston, we’re coming. On Tuesday, August 18, we’re joining Qodo Dev Community and Codex Community for Agentic AI in the Wild, a meetup for engineering teams already working with AI agents in day-to-day workflows. We’ll talk about what it takes to make agents hold up in production codebases, CI/CD pipelines, and everyday development work. On the lineup: 🔷 Matt Hargett from Callstack on local validation before costly cloud runs 🔷 Oliver Edis from ReactVision on building a working AR scene with AI and React Native 🔷 Nnenna Ndukwe from Qodo on bringing code quality into the SDLC 🔷 Rafael Mendiola from okthink on large codebase migration with loop engineering Join us on August 18: luma.com/86l0iugn?tk=nQ7fcS

Brownfield keeps the current app live but adds a boundary between two runtimes. Greenfield moves the risk to cutover. A third option starts greenfield, tests representative flows, and keeps a fallback into the native hosts. Download the guide: clstk.com/3TSBOEp

Kewin (@Q1w1N) shares a view we probably all have at this point: nobody needs more vague AI talk. What we do need is a clearer way to judge where these tools are genuinely useful. In the next livestream, we take that question into React Native and ask whether a specialized model can outperform a frontier model on work that is narrow, technical, and framework-specific. Join us here on August 13 at 5 PM 👇piped.video/live/eiGcqvji4nk

A QA agent should return evidence that a reviewer can inspect, not only a pass or fail. This clip from our last webinar covers five rules for keeping those runs reliable. For a full walk-through of the PR-to-device workflow and a live behavior-review demo, watch the webinar recording: clstk.com/3THC09z

People are tired of AI talk for a reason. A lot of it stays vague, repetitive, and detached from the work. We’re giving our livestreams a sharper format to do something more useful with the topic: look at AI from a developer’s point of view, stay close to real workflows, and spend time on where it saves time, where it creates risk, and where teams still need to slow down and think. Here is what you can expect from now on 👇

Return evidence a reviewer can inspect: screenshots, visited screens, logs, and a concise PR report. Match model effort to the task: repeatable tests can use smaller models.

Make the run deterministic: boot the expected build, install it, and prepare the app state before the agent begins.

Give the QA agent one bounded mission, such as onboarding or checkout. Use the PR description and affected code to tell it where to look, not to explore the whole app.