Fily

Europe watching from the sidelines again.

We are ready for takeoff 🚀

Voice agents in the AI SDK at last 🙌 One catch: the realtime hook 'useRealtime' is React-only… Hoping Vue support lands soon.

Solid 💪

What if your tech frustrations turned into your best ideas? I was fed up with macOS native dictation, too limited… I submitted the idea to the new Codex 5.3, and minutes later a full native macOS app using SwiftUI and OpenAI Transcribe was born. Ask AI to build your ideas.

I was almost right lol. Let’s try Codex 5.3 and Opus 4.6 then ✨

All that’s missing now is Codex 5.3, and it’ll be perfect.

✨ AI still manages to shock me. My agent generated a custom (scraped) MCP on its own to create ultra-modern sites/landings using tools like AuraBuild, auto-push to GitHub, deploy on Vercel and show me a realtime portfolio of every page and site it creates. And that’s not all… Then it created a second MCP to launch a Claude Code sub-agent that automatically converts each HTML page into a full Nuxt 4 project, clean component structure, styling, animations, and all the best modules (Nuxt UI, icons, fonts… thanks to the nuxt-skills skill, which is a blessing | @hugorcd ). The result? In just a few minutes, I get one or several polished pages, perfectly converted to Nuxt and ready to use. The craziest part? It took less than 15 minutes to build all the MCPs, automate the whole pipeline… and at that point, I wasn’t even the one driving the idea anymore. Now I just send it a WhatsApp message when I have a landing page idea: it generates the design in real time, I validate it, and if I like it, it converts everything automatically. It’s simply incredible… Built on OpenClaw/Clawdbot (Claude Opus 4.5) with ~24 skills.

I installed Clawdbot yesterday on my old Mac Mini M1, connected to a small 10" screen. It looks like a tiny iMac, add a little robot in front and it genuinely feels like having a mini employee at home 🦀🤖 While using it, I went through several phases: slight disappointment, then amazement, and finally reality. With a few skills installed, it becomes surprisingly impressive. But what I’m really excited for is what comes next: what will Clawdbot look like with a potential Claude Opus 5 or 6 and far more advanced skills? ⚡️

So, which model is actually the best to use with Clawdbot?

Dear developers, vibe coders, AI users 👀 A slightly strange question at first glance, but I’m curious: If you had the ability to stop AI’s evolution today, freezing it at its current level forever, would you do it? Or would you let it keep moving forward at full speed? 🤔 Around me, the debate comes up a lot: excitement for some, concern for the future for others. I’d really love to hear your thoughts.

Really excited to see what’s coming for Codex, especially after how much I liked 5.2. Cybersecurity is a real challenge, but this direction makes a lot of sense. Curious to see how these tools will make software safer for everyone let’s see how it plays out.

Thanks @jetbrains, @OpenAIDevs The Codex integration is genuinely a joy to use. Such a powerful agent inside a clean UI… and with a limited free trial on top of that. 👏

Dear @jetbrains, @WebStormIDE, @kskrygan Thanks… however, I need to be honest about my experience: I have to admit I was very disappointed with the integration of Claude Agent in the AI Chat. For more than two weeks of intensive use, repeated crashes and unexpected stops kept happening. Each time: credits consumed for nothing, and the need to recreate a new chat just to resume where the agent left off. In total, I’ve lost over $60 with no value in return. I’m now using Junie, which I really appreciate for its stability, and I’m glad I can finally try Codex directly inside your IDEs. However, please: this issue can’t remain as it is. These frustrating losses are likely to drive users away, especially at a time when competition has never been stronger. I’m sharing this because I truly value your products and want the experience to improve. (even if I would have preferred not to lose money unnecessarily 🙂)