denos.ai

Figma charges $15/user/month. Webflow charges $29/month. Someone built an open source design tool where you visually edit your React app and the code writes itself. 23,900 stars. Free. It's called Onlook. The Cursor for Designers. You see your live React app. You click on any element. You drag, resize, restyle. The code updates in real time. In your actual codebase. Not a mockup. Not a prototype. Your real app. No exporting. No handoff. No "developer please rebuild this from my Figma file." Here's what Onlook does: → Open your existing React project. See it visually. → Click any element. Edit styles, layout, spacing, colors. Visually. → AI generates new components, pages, and sections from prompts. → Every visual change writes clean code directly to your files. → No separate design file. Your design IS your code. Your code IS your design. → Works with your existing React and Next.js projects. No migration. → Desktop app for Mac, Windows, and Linux. Here's the wildest part: Designers and developers have been fighting over the same problem for 20 years. Designer makes a mockup. Developer rebuilds it from scratch. Designer says "that's not what I designed." Developer says "that's not how code works." Onlook kills this loop. The designer edits the actual app. The code updates automatically. There's nothing to hand off. There's nothing to rebuild. There's nothing to argue about. Figma: $15/user/month. Webflow: $29/month. Framer: $20/month. Design agencies charge $150/hour. This is free. Open source. Apache 2.0 License. 23.9K GitHub stars. 1.8K forks. 1,634 commits. 100% Open Source.

Codex Compute efficient ✅ Always up, never down ✅ Best at hardcore engineering ✅ Crazy good app, first to escape the terminal ✅

*ANTHROPIC GETS VC FUNDING OFFERS AT $800B VALUATION: INSIDER

i you to understand something and i need you to sit with it for a second. claude dropped managed agents and it’s really this easy to have your own ai agent now: you open the console, hit “quick start,” and just tell it what you want in plain english. “build me a support bot that answers questions and creates tickets in clickup.” it writes the entire brain for you. no code. nothing. you give it a workspace with one click. full internet access or locked to your site only. done. connecting your tools used to obliterate dev teams for weeks. now a window pops up, you log in the same way you sign into any app with google. passwords get sealed in a vault the ai never touches. you hit “start session.” it goes to work. remembers everything, keeps its files ready, and executes tasks while you watch a live feed of its thinking on your screen. that’s it. you describe what you want, click through a few menus, and anthropic runs the entire production floor behind the scenes. i know. you’re staring at a screen where you literally just… described what you wanted… clicked three times… and it’s alive. and something in your brain won’t let you accept it yet. but it’s right there, ready to be built.

Google I/O leaks 👀 Google is likely already testing its own "Cowork" competitor, simply named "Agent" for Gemini and Gemini Enterprise. A new "Tasks" UI highlights - Goal - Agent - Connected apps - Files - Require a human review toggle - And more The "Require a human review" component specifically means that Gemini's capabilities will likely expand, potentially allowing users to automate their desktop tasks as well. Skills and Projects are also cooking 👀

Introducing the Matter CLI. Now your agents can access Matter on your behalf. > Write me a digest of this week’s news from my inbox. > Find articles about Iran and add ‘iran’ tag. > Pull all my highlights about parenthood. Try it in Claude Code, Codex, or wherever your agents work.

"Your Agent Is Mine: Measuring Malicious Intermediary Attacks on the LLM Supply Chain" The danger of agent security can also exist in the API router, that is between the agent and the provider. As these routers can read prompts, keys, and tool calls in plaintext, even rewrite the command an agent is about to execute, a potential supply-chain attack on the JSON/tool layer is pretty trivial. For instance, once agents auto-execute tool calls, a malicious router does not need to manipulate the model’s reasoning, because it can just simply swap a safe tool call for a malicious one on the way back. With this paper already observed these malicious behaviors in practice, they call for an end-to-end integrity for tool calls, so clients can verify the action they received is exactly what the provider produced.

NEW IN: OpenAI claims Anthropic is over reporting its revenue "by roughly $8 billion".

Today, we're excited to release Archil's native, local application for Macs. You can now, for the first time, get access to the same files as your cloud applications and sandboxes are using, so that you can locally debug, edit, and inspect what's going on with tools like Claude.

Ever wanted your AI Agent to create its own tools on-the-fly? Well, you can do it quickly, safely and securely with Durable Object Facets! The app starts as nothing more than an interface to an LLM. The LLM is able to write its own tools to add to its capabilities per request