Steve building the proper agent OS that can conjure apps from a single prompt

I really take this for granted now, but in less than 30 mins, we: - generated a realtime collaborative app - deployed it into an authenticated & secure env - used the software together - and then let AI agents use it seamlessly on their own!

Fun times with @pjausovec playing his generated Tic Tac Toe game on Prophet... and then having our own agents play a round for us!

so get smart about how you generate code... example token analysis of compilation output. model generates DSL / intermediate format, compiler turns it into actual runtime code. need to only generate 796 tokens to deploy 8,265 tokens. 10x token difference (!!)

To my knowledge, I think we have the best abstraction and utilization of @cloudflare’s Durable Objects ever. They become an invisible layer to some seriously impressive primitives.

Soon™️

things are getting weird

Much love to the @CloudNativeFdn Helm team using Extism to power v4's new plugin system. helm.sh/docs/overview 🎉 Congrats on the new release!

We've entered a completely new era of app & system extensibility... and I'm more excited than ever about an AI-driven stack we've been building... more soon In the meantime, we are narrowing down our focus on what's new, and sunsetting the old. As such, mcp . run and XTP will go offline this week! Thanks to all who helped shape these products. I think you're going to really like what is coming next.

5/ @dylibso introduces Turbo MCP for enterprise security. This self-hosted gateway enables regulated industries to connect apps to AI safely. x.com/dylibso/status/1981419…

Connect your apps to AI on your terms. Introducing Turbo MCP, our enterprise-friendly, self-hosted MCP security platform and connectivity gateway. If you're looking to bring AI productivity to your workplace, Turbo MCP was designed with partners in highly-regulated industries, ready to meet the most rigid security requirements. 👉turbomcp.ai

What if AI could extend your app’s capabilities? Steve Manuel (@nilslice), founder and CEO of @dylibso, joins @sjmaple on AI Native Dev to decode MCP and reveal how mcp.run lets developers safely scale AI beyond its original limits. He dives into how developers can extend AI beyond text, connect to diverse data sources, and design enterprise-grade workflows with confidence, all with MCP. Catch their full conversation now. Link in the comments.

The risk of letting AI do more than autocomplete? It can quickly spin out of control. On this episode of AI Native Dev, Steve Manuel (@nilslice), founder and CEO of @dylibso, unpacks MCP, the protocol that keeps AI extensions safe and predictable, and dives into mcp.run, his framework for tapping into shared MCP servers without losing control. With @sjmaple he shares: • why plugin-safe AI might be the most significant shift in developer tooling this decade • how mcp.run isolates compute to prevent AI-driven code from being unpredictable, letting developers experiment without risk • the practical and surprising ways web-connected agents expand what applications can do, from automating workflows to integrating new capabilities • why first-party code remains essential even in an AI-native world (00:00) Trailer (01:32) Deep Dive into MCP Architecture (04:50) Trust and Security in MCP Servers (21:11) Managed vs. Self-Hosted MCP Servers (26:48) Understanding Serverless and MCP Servers (29:10) Developing and Deploying on mcp.run (33:20) Security and Authentication in MCP (38:31) Future of MCP and AI Innovations

💪💪💪 Welcome, @shritesh!

Excited to be at #AIInfraConnect this week, showing off our new self-hosted, cloud-native enterprise MCP platform! It's not just devs, everyone is connecting everything to AI... if you want to streamline SSO IdP integration, OAuth, DCR, and tool security, find our booth with this epic backdrop 👀

🤔 Have you ever wanted a plugin system for your project that can support a bajillion different languages? @cmgriffing has you covered in his talk from SquiggleConf 2024 on @extism! Check it out! 🔽 links below

I wrote an article about implementing the #Chicory #Wasm compiler backend for Android Dalvik/ART bytecode, tricks to run ART in an unconventional way inside! blog.evacchi.dev/posts/2025/…

When we created Chicory i theorized this would be possible, but had no idea someone would eventually do it! nice work @evacchi

🫡 (one caveat, you can make many URLs, each for a “profile” which contains configured MCP servers, and provide them to a client with specific trusted tools!) ((that’s how we enable our SSO, to let you “log in” with your tools: docs.mcp.run/blog/2025/05/14…))