> Open Claude Code > Ask it to "draw the onboarding screen of my app" > Think it looks good > Look again > Was the hierarchy thought out, or did it just look nice? > Do the colors match, or did the model guess them? > It looks like a product, but doesn’t behave like one > You swear the problem is the prompt > Write a longer, more detailed prompt > It doesn’t fix anything > Something 99% of people generating UI with AI never stop to question > Install 7 free skills and run the workflow > The agent interrogates you before coding > Reads your project so it doesn’t invent colors that don’t exist > Generates tokens, breaks them into tasks, builds, and reviews itself > One command replaced months of screens that never became a product > The problem was never the prompt—it was the lack of process > You were writing the ending without having the script The step-by-step for all 7 skills is in the article below, from briefing to review

Ruben Casas, staff engineer at Postman, gave a talk on the future of AI-generated design, presented at AI Engineer Europe (Portuguese subtitles): 00:46 AI already designs screens better than I do, without ego 03:05 Why am I still stuck with static screens? Jarvis? 05:48 The 3 ways AI builds a screen 10:10 The generative mode—writes the entire screen in real time 13:57 Radio analogy to explain the future 14:57 You and the agent designing on the same screen This single talk covers more than most scattered "AI design" posts you’ve seen online. Save it, watch it today, then read the guide on the 7 skills in the article below.

This LLM Model is fire 🔥 I’m going to give it a review before free usage ends on @OpenRouter Wow @TencentHunyuan built a beast.

Rolling up to the exclusive Pac Heights founder, investor, and researcher house party in the OpenAI socks

IMO, GPT-5.6 is infinitely worse than Fable at broad, human-like “intelligence.” It will go out of its way to do stupid shit to achieve, or get stuck on, something it shouldn’t have tried in the first place, simply because it didn’t understand or ask what the OG intent was.

Built a completely hands free 4 Agent Codex workflow with Codex in 1 hour. Start app Maximus Codexium starts my current project on my left monitor, 4 Codex instances on my main monitor, and opens DC/TG to our project groups on the right monitor When I'm ready to work, I say "Agent 1" , or "Agent 3" An always on script activates the input chat @WisprFlow Hands Free mode activates I talk to my agent Wispr's built in "Press Enter" feature sends my prompt Definitely seeing some potential in making this a lot more interesting 😎 Sometimes it's the smaller projects for personal utility that are the most fun to create. Now I can talk all day What to add next? #buildinpublic Git it? #WisprFlow #ChatGPT

Grok 4.5 is sitting at #2 on the FrontierSWE leaderboard. Above Claude Opus 4.8. Above GPT-5.5. So yes, the coding model conversation just got a little more crowded at the top. Strong performance is one thing. Doing it with serious speed, better token efficiency, and lower cost is where it gets annoying for everyone else. Builders love a smart model. @SpaceXAI @Grok @XFreeze / Writer: Annette, Designer: Janné

you can literally build shit for free with free inference right now and it is more capable than what you would've had to spend $1000 for maybe 3 months ago that's how retarded it is right now