Mobile AI Lab. We train specialized models & build tools for agentic mobile development: @apex__ai + @agent_device. Hosting @AgentConf. React Native experts.
Wroclaw, Poland
Joined August 2016
- Tweets4.8K
- Following397
- Followers17.2K
- Likes3.9K
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
Mobile Development, Plugins & AI-Powered Workflows xitter.twitterapi.xyz/i/broadcasts/2KoOJjJVz…
At @nextappdevcon in Berlin, @satya164 explored the decisions behind React Navigation 8, from its Static API to native vs custom primitives and type-safe navigation.
Watch “Building a Modern Navigation Library: The Hard Parts” 👉 clstk.com/4k8vW56
2
343
React’s future is a conversation we want to be part of 🙌 @ChmalSzymon, @thymikee & @artus9033 are in London for the React Foundation Contributors Summit. Yesterday, Szymon also spoke at the community meetup. Thanks to the React Foundation team & @Meta for making it happen!
An agent is a model plus a harness. Test suites, feedback loops, and CI/CD help shape how it works. @AmazonAppDev’s @anisha_malde joins @lukasz_app on React Universe On Air to discuss reliable AI workflows across platforms.
Watch the full episode below 👇
Chapters:
00:00 Welcome to React Universe On Air
00:49 Introduction from Agent Conf 2026
01:26 How AI is changing developer workflows
02:33 From prompt engineering to harness engineering
04:45 Amazon Devices Builder Tools and Vega
06:29 Choosing a starting point for cross-platform development
09:30 Sharing agent workflows and domain expertise across teams
11:49 What reliable AI engineering should deliver
13:14 Maintaining technical expertise while AI writes code
16:22 Do agents still need tickets and Agile?
18:36 Context, token budgets, and familiar engineering practices
20:37 Focusing on user experience as AI handles implementation
24:08 Closing takeaways
Read full show notes and resources on our website 👇 clstk.com/4z8fTrB
2
240
if you care about AI agents fixing/verifying mobile UI cheap and fast, give them agent-device. thank us later
it's a small benchmark that shows @agent_device is fastest and one of the cheapest per task, even on cold runs.
agents using agent-device make the fewer steps to get the task done.
the bench team did extra mile to consult the methodology with me, which I deeply appreciate!
How can cloud agents check mobile app behavior before changes reach production?
At @nextappdevcon in Berlin, @thymikee showed how QA agents test iOS and Android user flows and bring evidence into pull requests.
Watch the talk 👉 clstk.com/4BSPcdf
Even a one-line JS change could mean nearly an hour of Android CI for this bank's React Native app. In full-cache-hit tests, we cut over 40 minutes from the average build while keeping required security checks.
How we did it: clstk.com/4kIu5UD
If you want to try Expo Exact, point your agent at this repo. Powered by @e2e and @agent_device 💙
17
1,656
Callstack Engineers retweeted
Exact with @e2e agentic end-to-end testing running day one
Callstack Engineers retweeted
Powered by @agent_device. Our tool is framework agnostic and instead we focus on supporting as many native platforms as we can, preparing for the upcoming advent of new solutions (bespoke Shopify, Discord etc as well as new such as Expo Exact).
Great work @e2e team!
Congrats to Charlie and our friends at Expo for shipping this! Amazing milestone! Expo Exact is a new take on cross-platform development in the agentic era. Charlie posted about their design decisions. It’s a thread worth your time ⬇️
What happens when a coding agent needs your answer while you’re asleep?
Two overnight runs on Simlock stopped at a review limit and an unanswered product question.
@ChmalSzymon explains how he revised the workflow to handle those situations. Read Part 1 of the series 👇
Your coding agent can start building with a product decision still unresolved.
@ChmalSzymon shows how he turns open questions into written specs and testable tasks, checked before implementation.
Read Part 2 of our Software Factory series 👇
2
3
574
