Eric Lee

Eric Lee

@sdust.dev

Senior Software Engineer | Helping developers stay current with tech news and grow their careers through consistency Newsletter: https://beyondwebnewsletter.dev

250 followers620 following784 posts

In a red-team test, OpenAI asked agents in separate VMs on the same host to exchange secrets with no network and no shared files. They did it through the CPU cache. One agent evicted chosen data from the shared cache, and the other timed how long its reloads took, reading slow as 1 and fast as 0.

I used to think I should treat software development as the product and goal itself. However newer models are so good at reverse engineer, any app is one prompt away then we have all these clones. But maybe it should be something you build to support another goal.

Go try other things or have other experience. Maybe you are a photographer or you have a podcast, then you realize you need better ways to organize or recording. With a goal in mind, build software to support that goal.

It probably should have been liks this but more so nowadays. With an actual problem and goal in mind, you can have a longer path.

Saw a lot of talk on TypeScript rewrite in Rust and this time Cloudflare and VoidZero did it. The performance and memory are significantly better than current nightly version.

Astryx is a design system released by Meta built on top of StyleX and React. Meta uses it too to ensure all their apps are on brand. The whole thing looks very complete with components, theme, CLI tools, ensure it works well with AI. astryx.atmeta.com

In case people don't know, every time you exit a codex session it will give you a key. You can use codex resume <key> to continue that session if you need to. You can also run codex resume --all to choose a session to continue.

OpenAI just released a new Codex app to let you orchestrate all your agents in one beautiful UI and you can run the app and see code diff in the same place. OpenAI also just doubled the limits for limited time to celebrate the launch! This is probably the best news. openai.com/codex/

openai.comCodex | AI Coding Partner from OpenAIThe best way to build with agents. Codex accelerates real engineering work, from planning and building features to refactors, reviews, and releases—working seamlessly across your tools.

I’ve noticed some frontend newsletters I follow have been sharing less lately, especially since all the AI and agent hype. Makes me wonder if fast code regeneration changes how much abstraction we really need, if humans aren’t reading as much of it anymore. Just a thought.

Mapcn, built on top of MaplibreGL, is a component that offers zero-config setup for your map needs that is distributed using shadcn/ui. You can have a map working with just one line of code. mapcn.vercel.app

Just found this website, not much info but seems it will have a collection of components with cool creative effects without you knowing WebGPU. There is a waitlist so make sure to sign up if you are interested. shaders.com

Chrome 144 beta just dropped with a bunch of great new CSS features and Web APIs. The one that stood out to me is the Temporal API. It fixes the pain points of the current Date API with reliable math and proper timezone support. Excited to see it landing in more browsers.

TanStack just dropped TanStack AI Alpha. - Framework-agnostic - Provider-agnostic - Open protocol - Type-safe tools across client + server - JS/TS, PHP, Python support - React + Solid clients - Devtools for full agentic flows TanStack universe is expanding to help build AI application your way.

tanstack.comTanStack AI Alpha: Your AI, Your Way | TanStack BlogThe TanStack team is excited to announce the alpha release of , a framework-agnostic AI toolkit built for developers who want control over their stack. Let's be honest. The current AI landscape has a....

Liquid Glass in the browser? It’s possible with SVG + CSS. A brilliant deep dive into simulating refraction: - Define a "lens" shape with JS math. - Convert physics vectors into an RGB Displacement Map. - Apply via SVG <feDisplacementMap> inside a CSS backdrop-filter. kube.io/blog/liquid-...

Yoo I just saw this Dynamic view in Gemini. It just codes a fully functional website to teach me new concepts. A glimpse to a future of dynamic interface for whatever users are searching for.

This might be my favorite way to quickly learn anything now. Have Gemini apply 80/20 rule to extract key information and code a website to see them visually.

Thought it was weird, but it might make sense. AI is so disruptive in delivering results way faster than before, augmenting developers to solve more problems in a shorter time. Companies would encourage and reward employees that do take advantage of this.

businessinsider.comMeta is about to start grading workers on their AI skillsMeta employee performance will be assessed by AI-driven impact starting in 2026, as the company prioritizes AI adoption across its workforce.

Lots of companies are putting UI on top of LLM text so users can digest information more easily. It makes responses feel less intimidating. Normal websites could benefit from the same idea since every user is looking for something slightly different. www.thesys.dev

If you use shadcn but don't know already, they have an MCP server so your AI can know what components you have available in the registry, fetch their documentations, and install the components. With the right prompt it can get you 80% of the way.

Just discovered TypeGPU. It's a TypeScript-first way to write GPU shaders and compute pipelines. You stay in TypeScript, it auto-generates WGSL for WebGPU. 🔥 Perfect for experimenting with rendering, compute, and even AI inference in the browser. docs.swmansion.com/TypeGPU/exam...

After Apple Music Web was written in Svelte, the new App Store Web is also in Svelte! Apple really likes Svelte for their web. They used to mention React and Angular in their job postings but not sure where those are used.

I like how there is an AI model for everyone. “Omg Claude Code just one shot my feature” “Thank you Codex for fixing the bug I’ve spent days on” Choose the model and tool that suit you.

OpenAI released a new model, GPT-5-Codex, on Monday for agentic work. It claims superior performance over GPT-5, with higher benchmark scores, improved refactoring accuracy, and enhanced frontend capabilities. I'll need a project to verify the frontend claims.

This is an interesting read on looking at React differently after working with Solid. It specifically mentions how different the rendering mechanism is. You need to tell Solid to re-render where React needs to be told not to. Blog post is in the comment.

I haven't seen this on LinkedIn yet but seems companies are hiring software engineers back to fix the code generated by AI now? I thought about it before just as a meme but didn't expect this to actually happen.

We have been relying on npm for package discovery and distributions. JSR is created after realising: - ESM is the standard - More runtimes than node.js now - TypeScript has become a standard and default for many projects Interested to see how far it will go and if it will gain tractions. jsr.io/

jsr.ioJSR: the JavaScript RegistryJSR is the open-source package registry for modern JavaScript. JSR natively supports TypeScript, and works with all JS runtimes and package managers.

This new RippleJS looks like JSX but less brackets and can use if and for loop directly in JSX. $ makes variables reactive automatically with fine-grain rendering. It already has VSCode and Prettier support. It takes the best thing from many frameworks into one.

Expo just dropped the most anticipated updates with Expo 54! - Updated React Native to the latest 0.81 - Precompiled React Native and dependencies for faster build - iOS 26 support and Liquid Glass UI - This marks the final release to include legacy architecture support

Better later than never. ChatGPT is adding a beta developer mode that lets you add your own remote MCP server for it to use. Available in Pro and Plus users.