AI Browsers

AI browser wars heat up as OpenAI launches Atlas

From Ars Technica:
"Back in 2008, Google launched the Chrome browser to help better integrate its industry-leading search engine into the web-browsing experience. Today, OpenAI announced the Atlas browser that it hopes will do something similar for its ChatGPT Large Language Model, answering the question 'What if I could chat with a browser?' as the OpenAI team put it."

OpenAI enters an increasingly crowded field of AI-native browsers. Perplexity soft-launched Comet in July and made it broadly free in early October. In September, Atlassian announced it would acquire The Browser Company (makers of Arc and the AI browser Dia), positioning Dia as its work browser. Microsoft has added Copilot Mode to Edge, and Google is rolling out Gemini in Chrome.

OpenAI’s Atlas includes an Agent Mode (preview for Plus, Pro, and Business users) that can navigate sites and complete multi-step tasks—e.g., adding recipe ingredients to Instacart or moving planning tasks across apps. It also offers optional browser memories for smarter suggestions and an Ask ChatGPT sidebar for any page you’re viewing. Atlas is available now on macOS, with Windows, iOS, and Android coming soon.

For Agent Mode, security concerns persist. OpenAI's Chief Information Security Officer (CISO), Dane Stuckey, acknowledged that prompt injection—where malicious instructions hidden in websites trick the AI into unintended actions—remains "an unsolved security problem." While OpenAI has implemented defenses, the effectiveness of these protections remains to be proven in real-world use.

Ars Technica | OpenAI looks for its “Google Chrome” moment with new Atlas web browser
⚡ Quick Read (5 minutes)

Simon Willison | Open AI's CISO on prompt injection risks for ChatGPT Atlas
⚡ Quick Read (5 minutes)

Frontier Models

Anthropic launches Claude Code for the web

Anthropic has released Claude Code on the web: a new way to delegate coding tasks directly from your browser, without needing to open your terminal. Coding tasks run on Anthropic-managed cloud infrastructure rather than locally.

From Anthropic:
"The web interface complements your existing Claude Code workflow. Running tasks in the cloud is especially effective for:

• Answering questions about how projects work and how repositories are mapped

• Bugfixes and routine, well-defined tasks

• Backend changes, where Claude Code can use test-driven development to verify change"

Claude Code on the web enables multiple coding tasks to be run in parallel across different projects, allowing users to check in on their progress from their browser or phone.

Anthropic is joining a growing trend of "asynchronous coding agents," which include tools like OpenAI's Codex Cloud and Google's Jules, that work autonomously rather than requiring constant input. Notable is Anthropic's emphasis on security: each task runs in an isolated sandbox with file system and network restrictions to prevent data leaks (where a compromised agent could send your data somewhere before you even realize what happened). Data leaks are a significant concern when AI coding tools work autonomously, without the user watching each step. Claude Code for the Web is currently in beta as a research preview for Claude Pro and Max users.

Anthropic | Claude Code on the web
⚡ Quick Read (2 minutes)

Simon Willison | Claude Code for web—a new asynchronous coding agent from Anthropic
⚡ Quick Read (5 minutes)

AI and Design

Figma acquires Weavy to move beyond one-and-done AI generation

Figma has acquired Weavy, bringing the Tel Aviv startup's node-based creative platform into its ecosystem as Figma Weave. Weavy's core innovation is unifying multiple AI models with professional editing tools, such as masking, relighting, and compositing, all on a single canvas. Think of it as a visual workflow builder where you can chain together prompts, AI generations, and edits: generating an image with one model, feeding it through image-to-image refinement with another, then using professional tools to adjust lighting or mask objects.

Founded just over a year ago and publicly launched only four months ago, Weavy has built its reputation around what it calls "Artistic Intelligence," prioritizing craft, control, and professional-grade workflows. The startup's team will join Figma, and the product will initially remain standalone before being integrated into the broader Figma platform.

Figma | Introducing Figma Weave: the next generation of AI-native creation at Figma
⚡ Quick Read (3 minutes)

Weavy | Weavy tutorial
Watch Time: 5 minutes

Designing for AI

AI does more than create new features; it solves old problems

Luke Wroblewski argues that AI's value isn't limited to chatbot sidebars: it enables fundamentally rethinking UX patterns we've accepted as inevitable for decades:

• Instead of empty states and tutorial flows, AI can generate working content for users to edit from day one.

• Rather than forcing people through complex search filters and taxonomies, AI can translate natural questions into multi-step queries.

• Instead of rigid web forms that require humans to format data for databases, AI can accept unstructured input, such as images or PDFs, and extract the structured data itself.

This post is a quick yet impactful read, featuring real-world design examples.

Luke Wroblewski | Tackling common UX hurdles with AI
⚡ Quick Read (2 minutes)

Frontier Models

Why Claude Code is the most underrated AI tool for non-coders

Despite the name, Claude Code isn’t just for developers. It’s a command-line AI agent that runs from your computer and can directly edit files, run commands, and automate workflows in your local environment. It can also connect to tools like Google Drive, Jira, or your filesystem with permissions. Think of it as "Claude Local" or "Claude Agent."

In his newsletter post, Lenny Rachitsky documents a wide range of practical use cases for non-technical professionals. People are using Claude Code to organize messy file systems, transcribe and synthesize customer call recordings, create hiring plans from internal documentation, improve image quality, and even build self-driving documentation systems. One user downloads meeting recordings into a folder and asks it to identify times they've avoided conflict. Another synthesizes voice notes from stroller walks into polished articles.

The post includes (easy) instructions for installing Claude Code on your computer; the first 18 real-world prompts are free; the full 50 example use cases are available to newsletter subscribers.

Lenny Rachitsky | Everyone should be using Claude Code more
☕ Medium Read (8 minutes) | 💡Bookmark for reference

That’s it for this week.

Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋

All the best, Heidi