
ceo and founder @callstackio, codex ambassador @openai, @reactjs foundation. i love cars 🏎️
Joined December 2012
- Tweets17.6K
- Following906
- Followers15.6K
- Likes28.1K
We believe in domain-specific models and domain-specific harnesses. At Callstack, we've been investing in both.
Excited to bring you the first piece of a broader family we're building. More to come!
Announcing Apex: agentic coding model for mobile and web, specialized in React.
We built Apex from a single idea: can an open-weight model deliver frontier-level performance at lower cost?
Yes. Apex is SOTA in React Native and Next.js.
Try it now: apex.callstack.com
My PR to fbjni was just merged!
It speeds up all calls to Java/Kotlin in react-native (and Nitro) by 14%! 🤯🚀
#MakeReactNativeFastAgain
github.com/facebookincubator…
Mike retweeted
find me there
Vercel AI Gateway
- Zero fees for BYOK
- Zero markup on list prices
- Zero payment fee with ACH
- Volume discounts available
[email protected]
Mike retweeted
Replying to @cramforce
and @apex__ai is there, a model specialized in React and React Native work 💪
Mike retweeted
Vercel AI Gateway
- Zero fees for BYOK
- Zero markup on list prices
- Zero payment fee with ACH
- Volume discounts available
[email protected]
Our model is getting better
This time ranked #2 in ReactBench's for fixing bugs in React code.
Pretty remarkable result. Can't wait to see where we get a year from now.
Try it out
Put @apex__ai on a React issue from your backlog 🔎 It currently ranks #2 in ReactBench’s Fixing React category: finding and fixing problems in existing code while preserving behavior. Try it in your repo: clstk.com/3VJmM4Z
agent device got faster again!
that 1M+ a month can't be wrong
so I got nerd sniped by @o_kwasniewski with an experimental renderer being faster than agent-device in e2e, so now v0.21.29 comes with:
→ 2× faster Android taps and verified fills
→ 18× faster immediate screenshots,
→ 18% quicker on a 42-test suite
enjoy
Mike retweeted
Checking an invitation panel means recreating the invitation every time.
With Appduct, an AI agent can call a function your team writes to trigger the panel with test data, wait until it’s ready, and inspect the UI.
See how: clstk.com/4smYoCo
We're waiting for release in the plugin store, but meanwhile you can try us by pulling our custom marketplace! github.com/callstackincubato…
Mike retweeted
a small company from Poland trained a model that's at the top of the React benchmark. everything is possible now, lol
Put @apex__ai on a React issue from your backlog 🔎 It currently ranks #2 in ReactBench’s Fixing React category: finding and fixing problems in existing code while preserving behavior. Try it in your repo: clstk.com/3VJmM4Z
Mike retweeted
What evidence would make you comfortable shipping an agent’s changes?
Join us in Amsterdam on Oct 22 for three talks with speakers from Callstack, @expo, @jetbrains, and Chromatic.
Sign up here 👉 luma.com/i8wkxgyg
Mike retweeted
Many proposed fixes, one accepted candidate. Carl Ross of @OpenAI joins @lukebfarrell and @lukasz_app to explain how agents explore code improvements and why discarding work is part of the process.
Why put 100 agents on code quality? Carl Ross of @OpenAI explains competing code changes, a grader that picks the best candidate, and agents that keep cleaning up as features ship. Watch the interview ⬇️
What Happens When AI Agents Have No Token Budget?
Mike retweeted
The printed report is out in the wild. Want a digital copy now, before you catch us in person? Download it at clstk.com/47MgFj7
React Native at Scale 2026 draws on conversations with 12 organizations and a decade of our React Native work.
Mike retweeted
I love having access to unlimited tokens (like @apex__ai for everyone at Callstack).
Opus 5.5 is king this month but there’s a lot of workload that I can just throw at a cheapest model that’s governed by a frontier one.
Mike retweeted
Turn mobile designs into working features with a lower-cost AI model.
@thymikee will explain the verification loops behind a rise from 60% to 98% of PRs merged on the first attempt in the workflow he’ll present.
Oct 22 · Amsterdam · Save your spot: clstk.com/3WynEcJ
Really looking forward to this one
I just decoupled Nitro from react-native.
The new package "𝚗𝚒𝚝𝚛𝚘-𝚌𝚘𝚛𝚎" exposes all portable native APIs, JS bindings and TypeScript types.
This makes it easier for other frameworks (like @LynxJS_org or @NativeScript) to adopt Nitro - all they need is JSI. Almost all JS Runtimes have JSI implementations: Hermes, V8, JSC, SpiderMonkey, Chakra, etc.
We also now test 𝚗𝚒𝚝𝚛𝚘-𝚌𝚘𝚛𝚎 standalone on Hermes, without react-native.
𝚛𝚎𝚊𝚌𝚝-𝚗𝚊𝚝𝚒𝚟𝚎-𝚗𝚒𝚝𝚛𝚘-𝚖𝚘𝚍𝚞𝚕𝚎𝚜 remains inside the repo and stays one of the official targets.
It contains react-native specific setup (like autolinking and .podspec), and react-native specific features like 𝙷𝚢𝚋𝚛𝚒𝚍𝚅𝚒𝚎𝚠 that was previously inside nitro.
As part of this, 𝚗𝚒𝚝𝚛𝚘𝚐𝚎𝚗 will receive a plugin system soon so we can fully lift out 𝙷𝚢𝚋𝚛𝚒𝚍𝚅𝚒𝚎𝚠 and other RN specific features.
This could set the foundation for third party language targets (like Rust) too.
Nitro V1 is 70% done and will be a breaking change for library authors.
Mike retweeted
Keep control of what ships when agents write your UI.
Craig Walker and @KasperPeulen explore how to steer and verify agent-generated UI with Storybook’s agent-native CLI and WebStorm’s design-to-browser workflows.
Save your spot at clstk.com/4BbFhz5
Mike retweeted
Agents and frameworks change. Do you rebuild the infrastructure each time? Nishant Gupta of @Meta explains a reusable control plane for agent fleets, plus why verifying outcomes still takes time.
What changes when you move beyond a single agent? Nishant Gupta of @Meta explains shared state, background ops agents, and a control plane for authentication and observability. Watch the interview ⬇️


















