ChatGPT desktop's built-in browser now supports WebMCP, which lets a website expose its features as tools for AI agents. We build a WebMCP-enabled TODO app and inspect the arguments and results of the Site tools Codex calls from its execution history.

IndexedDB is an asynchronous, transactional database for storing structured data in the browser. Learn its fundamentals through the native API, then build a task management app with React and Dexie.js.

Forgetting to restore a spy in Vitest lets mock state leak into other tests. Since Vitest 3.2.0, assigning the return value of `vi.spyOn()` to a `using` declaration restores the original implementation automatically when the scope exits.

I thought AI-written articles were too generic and too wordy, so I only used AI to catch typos. Trying a workflow that hands drafting to AI, I found that reading the draft critically and adding my own experience preserves most of the learning.

Slack Code brings open, team-wide development into Slack. Instead of DMs and threads, you create a special channel called a code channel and work with a coding agent as a whole team. Here is what it was like to actually try it out.

Adding thousands of elements to the DOM at once drives up rendering and update costs. TanStack Virtual is a headless library that renders only the items needed for the current scroll position. This article covers virtualizing a list in React.

ARIA roles alone do not make a Tree View accessible. Learn to build a single-select Tree View in React with roving tabindex, arrow-key navigation, selection, and type-ahead, following the WAI-ARIA APG.

Chrome 151 introduces the `<usermedia>` element, a new HTML control designed to address limitations of `getUserMedia()`. This article explains the proposal and verifies its behavior with hands-on examples.

When you build with Web Components, name collisions happen if two libraries define custom elements with the same name. Scoped Custom Element Registries let you register different implementations under the same name in separate scopes on one page.

Generative AI chat UIs reveal answers a few characters at a time. Piping those updates into a live region makes screen readers read the response in fragments. I survey ChatGPT, Claude, and four chat UI libraries, then build an accessible sample.

`aria-actions` associates an element with separate interactive elements that provide actions for it. This article explains the problem it solves and demonstrates its use with secondary actions in an email list.

The Fetch API now includes textStream(), which reads response bodies as UTF-8 text. It replaces manually piping through TextDecoderStream with a single call. This article compares it with the previous approach.