One of the best examples so far of what can be done with Build. And it’s improving every day. It’s the worst version of Build there will ever be, as they say.
I created this game 100% on my phone with #Roblox Build.
What I loved is that I followed the usual process of game creation; first asking Build to create the different gameplay elements, iterating on them, and then assembling them.
AI as a tool to push creativity is amazing, and there is still a lot to discover about new workflow.
roblox.com/games/70649779571…
So exciting to be a part of this!
Announcing Build, a new mobile-first creation experience coming soon to Roblox.
With Build, we’re opening creation to more people while also rolling out powerful new tools for existing creators to help them test and iterate faster. Learn more: rblx.co/buildwithoutlimits
just open sourced xcode-studio-mcp... a unified MCP server that lets AI coding agents build, deploy, screenshot, and interact with iOS Simulator from one tool.
the gap: today you need XcodeBuildMCP for building + mobile-mcp or ios-simulator-mcp for UI interaction. two separate servers, context switching between them. nobody had combined them.
6 tools, ~750 lines of Swift, MIT licensed:
xcode_build (structured errors with file:line:col)
xcode_run (build + launch)
simulator_screenshot
simulator_tap / type / describe
works with Claude Code, Cursor, or any MCP client.
github.com/kevinswint/xcode-…
What is the most surprising thing about building real products with AI coding agents? It's not how astonishingly fast and capable they are. It's...
The point is that the skills that matter most today in AI-native development are the same ones that matter in traditional product development. Clear communication. Good product judgment. Systems thinking. Great documentation.
Wife: "oh, and do you remember talking about..."
Me: "Compacting our conversation so we can keep chatting...😵💫😵💫😵💫"
I replaced a 5-person dev team with myself + Claude Code and Codex.
The product shipped in two weeks. Not a prototype — a production platform live with real users right now.
Thread on what actually happened 🧵
The question companies should be asking isn't "how do we make our eng teams 20% more productive with AI?"
It's "what happens when a single senior product leader who deeply understands a domain can build the entire product themselves?"