The week AI stopped living inside apps and started living inside your entire workflow

If you only have five minutes, here’s your AI news this week in one line: Claude Code just got its own browser, Spotify decided you should talk to it like a person, OpenAI finally shipped a physical keyboard, and a wave of new open-weight models quietly rewrote the leaderboard. None of it is slowing down, and all of it matters if you use AI for work, content, or just doom-scrolling with better recommendations.

This roundup of AI news this week covers everything from Anthropic’s new browser feature to a state government hitting pause on data centers. Let’s get into it.

Claude Code’s New In-App Browser Is the Biggest Story in AI News This Week

Just days after OpenAI shipped a ChatGPT app packed with browser features, Anthropic rolled out something suspiciously similar: an in-app browser built directly into Claude Code on desktop.

Here’s how it works. The browser doesn’t live in the normal Claude chat interface — you need to be inside a Claude Code project. From there, hitting Cmd+Shift+B (or clicking the browser icon in the top right) opens a live browser pane on the right side of your screen, sitting right next to your code.

What makes this update stand out in this week’s AI news cycle:

An annotation feature lets you click an element on a live webpage — say, a button — and it automatically highlights the matching code on the left. Tell Claude to “change the text to Get on My List,” and it edits the code for you in real time.
The browser can be used as an autonomous research tool. Instead of paying for expensive API access to platforms like X (formerly Twitter), Claude can simply scroll and read the page like a human would, pulling information without touching a paid API.
It closes the gap between “AI that writes code” and “AI that can see and interact with the actual product,” which has been one of the biggest friction points in agentic coding tools all year.

This is arguably the most practical piece of AI news this week for developers and no-code builders alike, since it turns Claude Code into something closer to a full-stack assistant rather than just a code generator.

Google Is Turning Search Into an AI Agent

Google didn’t sit still either. This week’s AI news includes a quiet but important expansion of Google’s AI Mode inside Search, which can now connect directly to outside apps.

Ask AI Mode something like “make a grocery list and add it to a shopping cart,” and it will search for an app that can complete the task — in this case, Instacart — and prompt you to connect it. There’s no traditional “add apps” menu; the system decides what’s relevant based on your prompt and offers the connection on the fly.

Google also refreshed Google Vids, its slideshow-style video platform (not to be confused with Veo). The new Gemini Omni integration doesn’t just generate static slides anymore — it can animate them and let you edit videos conversationally, just by describing what you want changed. A new “import yourself” feature also lets you drop a personal avatar into auto-generated clips, like birthday videos, though it’s still fairly obvious you’re watching an AI-generated stand-in rather than a real recording.

Spotify Wants You to Talk to It — Literally

One of the more surprising entries in AI news this week comes from Spotify, which rolled out a conversational AI feature for Premium users. According to the company’s own announcement, the goal is to make it easier to “shape what you hear, understand what’s playing, and explore your listening just by asking.”

In practice, that means typing or speaking directly into the Home and Now Playing views to have a back-and-forth conversation about your music. You can ask what you were listening to two years ago, request a playlist based on that history, or ask for details about an artist — and Spotify’s AI answers in natural language while building the playlist live. Early testing shows it works surprisingly well for niche requests, correctly identifying listening patterns and building genre-accurate playlists on the spot.

The Open-Weight Model Race Just Got a New Leader

This week brought a flood of new open-weight releases, and the results reshuffle the pecking order in a meaningful way.

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first public open-weight model. It’s solid for an open-weight release but doesn’t top the charts against closed models.
GLM 5.2 remains the strongest openly available model overall, beating out newer competitors across reasoning and agentic coding benchmarks.
Bonsai 27B from Prism ML is a heavily compressed, roughly 4GB model built to run entirely on a phone. It’s fast — but not built for serious coding work.
The real headline is Kimi K3 from Moonshot AI. Once its weights are released later this month, benchmark data suggests it will be the most capable open-weight model available, reportedly outperforming GPT-5.5, Opus 4.8, and GLM 5.2 on several coding and terminal benchmarks — a genuinely big deal for anyone running self-hosted or fine-tuned models.

If you follow AI news this week for one reason, the open-weight race is it: the gap between “free to download” and “frontier-level” models is closing fast.

OpenAI Finally Ships Hardware — a Keyboard

After months of rumors about OpenAI hardware, the company’s first physical product has arrived, and it’s refreshingly modest: a keyboard.

The Codex Creator Micro, built in collaboration with Work Louder and sold through openai.com/supply, is a compact keypad tuned specifically for using Codex. It retails for $230. It’s not the futuristic AI companion device people were expecting, but it’s a first physical step — and reports suggest a screenless, Alexa-style smart speaker may be next, though that remains unconfirmed.

The hardware news lands alongside a separate legal story: Apple is suing OpenAI, alleging former Apple employees brought trade secrets and hardware concepts with them when they joined. OpenAI denies the claims, but it’s a battle worth watching as the AI hardware race heats up.

Rapid Fire: More AI News This Week

A handful of smaller updates rounded out the week:

Grok open-sourced its coding agent, Grok Build, and added scheduled “automations” — similar to the loop-style triggers already available in Claude and ChatGPT.
ChatGPT significantly improved in-app search, letting users filter across chats, projects, images, and documents from one sidebar.
Claude gained the ability to use 1Password credentials to log into websites on your behalf — without ever exposing the actual passwords to the model itself.
DoorDash launched a command-line interface, meaning AI agents can now search restaurants, compare deals, and check out entirely on their own.
Meta rolled back its controversial Instagram feature that let users generate AI images of tagged people, after backlash from creators and actors.
Siri’s newest version began rolling out with noticeably better AI features, including support on Apple Watch.
Anthropic launched Claude for Teachers, giving verified K-12 educators in the US free access to premium Claude features.

Data Centers Are Becoming a Political Flashpoint

Away from product launches, this week’s AI news included a notable policy shift: New York became the first state to enact a data center moratorium, blocking new environmental permits for facilities over 50 megawatts for up to a year. The stated goal is to give regulators time to address rising energy prices and environmental strain before more hyperscale data centers get built.

At the same time, Sunrun is testing the opposite approach — a pilot program that pays homeowners to host small, solar-powered compute units, effectively turning residential neighborhoods into distributed mini data centers that AI companies can tap into. It’s an early but interesting sign that the next phase of AI infrastructure might not be built in massive campuses at all.

The Bottom Line

Compared to the last few weeks of nonstop model launches and app overhauls, this stretch of AI news this week felt relatively quiet — and it was still packed with real, usable updates. A browser inside your coding tool, a chatty music app, a new open-weight leaderboard contender, and the first real OpenAI hardware product all landed in the same seven days. The pace isn’t slowing down anytime soon, so if this is the “quiet week,” buckle up for what comes next.