A clearer view of AI.
Submit for reviewSubscribe

WEEKLY NEWSLETTER

The AI Insider

A weekly round-up of the AI tools worth your time, and the hype to ignore.

Unsubscribe at any time using the link in each newsletter.

THE INDEPENDENT AI.

AI NEWS FOR BUILDERS · 15 STORIES

AI news, 12 November 2026

From the community

Jensen Huang resigns as Nvidia CEO; Colette Kress takes interim role

Nvidia announced on 12 November that Jensen Huang has resigned as president, chief executive and a board member with immediate effect. CFO Colette Kress becomes interim CEO while retaining her finance role. The board is searching for a permanent successor, considering internal and external candidates, and has appointed Harvey C. Jones as chair. Nvidia says its operations, customer commitments and product roadmap are expected to continue without interruption. This updates our earlier report about a possible leadership change.

Why it mattersFor AI buyers, the immediate question is continuity of chip supply and delivery. Nvidia says commitments remain unchanged; watch for concrete updates rather than assuming a leadership change alters availability or pricing.

Sourcesnvidianews.nvidia.com ↗

Google Antigravity's coding agent can now use a browser

Google is rolling out browser use in Antigravity, its agent coding app. The agent can research, write code and then check the result live in a browser, and it can run several browser tasks at once. You decide which sites it logs into, and any download waits for your approval.

Why it mattersAn agent that can open the page it just built and test it catches more of its own mistakes before you see them. The approval steps for logins and downloads give teams a simple way to keep it on a short leash.

Sourcesx.com post ↗

ChatGPT brings Finances to its desktop app, including on the free plan

OpenAI has added Finances to the ChatGPT desktop app. It helps you find ways to save, understand your spending and make sense of your investments, and you can build your own custom financial tools. It is available on the Free, Go, Plus and Pro plans after updating the app.

Why it mattersMoney questions are moving into general assistants rather than separate budgeting apps, and the custom-tools option means people can build their own trackers without code. Business plans are not on the list, so this is a consumer feature for now.

Sourcesx.com post ↗

Gemini comes to Apple CarPlay worldwide

Google is rolling out the Gemini app on Apple CarPlay worldwide to all iOS users, in every language Gemini Live supports. Drivers can tap the Gemini icon to talk hands-free: catch up on their schedule, talk through their inbox or brainstorm out loud.

Why it mattersThe drive becomes usable time: hands-free inbox and calendar catch-ups with an assistant that already knows your Google account. Claude added CarPlay voice in October, so the car is becoming a standard place for assistants.

Sourcesx.com post ↗

Gemini's managed agents can now run in your own sandbox

Google says developers building with Managed Agents on the Gemini API can now bring their own sandbox through self-hosted environments, so the agent's code runs on infrastructure they choose. Launch partners are Vercel, Modal, Daytona, Fly.io, Google Cloud, Cloudflare and E2B.

Why it mattersCompanies that want Google's hosted agents but need the agent's code to run on their own infrastructure can now do both, using sandbox providers many teams already pay for. That makes Gemini agents easier to approve for security-conscious teams.

Sourcesx.com post ↗

Google's A2UI, an open standard for agents that build screens, reaches 1.0

Google Cloud released A2UI 1.0, now marked stable, faster and easier to use. A2UI is an open protocol and framework that lets AI agents generate user interfaces, such as forms, cards and dashboards, that work across the web, Android, iOS, wearables, TVs and other screens.

Why it mattersA stable, open way for agents to show buttons, forms and dashboards means products can let an AI assistant build the right screen on the fly, on any device, without being tied to one vendor.

Sourcesx.com post ↗

Google Labs' Tabletop lets groups build their own tools together

Google Labs launched Tabletop, a shared space where a group such as friends, family or a club can create custom tools together. Everyone in the group can contribute at the same time, for example to plan a trip.

Why it mattersIt brings vibe-coding to groups rather than individuals: several people shape one shared app at the same time. It is an early Labs experiment, so treat it as a preview of where collaborative AI tools are heading.

Sourcesx.com post ↗

Meta's Astryx: an open-source design system built for people and AI agents

Astryx is Meta's open-source design system, built so people and AI coding agents work the same way. Meta says it grew inside the company over eight years and now powers 13,000+ apps. It ships 150+ accessible components, brand theming, dark mode, templates and a CLI. Meta says the API, docs and CLI are designed together, and that installing the CLI globally means the astryx command works in any folder and every coding agent on that computer knows Astryx. Official docs also describe a Model Context Protocol server that compatible AI tools can query for components, docs and examples.

It is in beta and needs React 19 or later. It is written in TypeScript and built on StyleX, but Meta says you can override styles with Tailwind, CSS modules or plain CSS, and a swizzle command ejects a component's source into your project. There are seven ready-made themes. The charts packages are published only as canary releases, with no stable version yet.

On 11 November 2026, Meta for Developers said Astryx had passed 24,000 GitHub stars and hit #1 on GitHub trending. The GitHub API showed about 25,600 stars when we checked on the morning of 12 November. Meta introduced it as open source in a blog post dated 18 June 2026. It is free under the MIT licence.

It suits builders, agencies and teams starting new React 19 projects who want one system for humans and coding agents. Existing apps on an older React version would need to upgrade first. We have not independently audited the component set. Astryx is now in the directory, with docs at astryx.atmeta.com and the repo at github.com.

Why it mattersA design system that coding agents can query, through a CLI and an MCP server, is more useful than a component library that only humans read. Astryx is free and open source, but it is still in beta and only supports React 19 or later.

Sourcesgithub.com ↗ · astryx.atmeta.com ↗ · astryx.atmeta.com ↗ · astryx.atmeta.com ↗ · astryx.atmeta.com ↗ · x.com post ↗

Adobe makes Creative Cloud agentic, and previews local MCP for third-party agents

Adobe said at MAX on 10 November 2026 that it is turning Creative Cloud tools into agentic workflows. The Adobe Creative Agent is now in Premiere, Illustrator, InDesign, After Effects and Frame.io. Adobe says AI Assistants powered by that agent are in beta in those flagship desktop apps; in Photoshop, the assistant is only in the Photoshop beta app.

The bigger change Adobe previewed is local MCP integrations. Third-party agents will be able to work directly in native Photoshop, Illustrator, Premiere and After Effects files on the user's computer, and can chain that work with tools such as Google Drive or Slack. Adobe says those integrations will be in private beta in early 2027, so this is a preview and is not available now.

Adobe Firefly is getting three new surfaces. Firefly Design (beta) puts raster, vector, motion, 3D and print on one canvas. Firefly Graph (beta) is for building repeatable workflows that combine AI models and Adobe's production tools. Firefly Builder (private beta) is a preview of design-to-code prototypes, such as a website, presentation or app, starting from natural language. Adobe says Firefly Design is included with Creative Cloud.

Adobe also named new features in the flagship apps, including Relight and Generative Restyle in Photoshop, and Color Mode and a Generative Media Tool in Premiere. Adobe says the Generative Media Tool can generate video, sound effects, soundscapes and music directly in the timeline. After Adobe's acquisition of Topaz Labs, Adobe says Topaz is introducing two new image enhancement models, Wonder 4 and Image HDR, for upscaling, denoising, deblurring, sharpening and turning standard images into HDR.

Adobe says the new Creative Cloud capabilities were generally available as of the 10 November announcement, alongside the betas and private betas above. The local MCP integrations remain a private-beta preview for early 2027.

Why it mattersAgencies and in-house creative teams already live in Creative Cloud. An assistant that can be directed inside Premiere or Illustrator, and later a hook for a third-party agent to work in the native file, is a meaningful workflow change for those teams — but the local MCP piece is a 2027 private-beta preview, not something you can use today.

Sourcesnews.adobe.com ↗

n8n's free personal agent Nod does your errands on its own computer

n8n Labs released Nod as a research preview on 10 November: a free, open-source app for Mac and Windows (or your own server) that gives an AI agent its own sandboxed computer and browser on your machine. It works in Gmail, Calendar, Drive, Slack and Notion and on websites with no integration, for jobs like downloading invoices from supplier portals, writing call briefs or cancelling a subscription, and it asks before it sends, pays, posts or deletes anything. You bring your own AI model (an API key, your ChatGPT plan or a local open model), and it shows what each task cost, with a monthly spending limit. n8n reports 98% task completion on its own 226 test errands; that's the company's figure. It stops when your laptop sleeps and can't make phone calls. Shared by @dr_cintas.

Why it mattersThe admin that eats a solo operator's week, like logging into portals and chasing paperwork, is exactly what it targets, and it costs nothing beyond model usage. It's a research preview, so start with one low-risk job and keep approvals on.

Sourcesx.com post ↗ · blog.n8n.io ↗ · github.com ↗

Anthropic's IPO is said to be several times oversubscribed

Bloomberg reported on 11 November that orders for Anthropic's shares are several times the 825 million on offer at $70 to $78, citing people familiar with the deal. With a $26 billion private placement to companies including Google, Nvidia, Microsoft and TSMC, Anthropic could raise up to about $90.4 billion in total. These are reported figures; Anthropic hasn't confirmed the demand. Flagged by @kimmonismus.

Why it mattersStrong demand means Anthropic will have far more money for computing capacity and new models, which matters if Claude is central to how you work. Nothing to do yet; the listing date still hasn't been announced.

Sourcesx.com post ↗ · news.bloomberglaw.com ↗

AlphaGenome 2: Google DeepMind's DNA model zooms in to individual cell types

Google DeepMind released AlphaGenome 2 on 11 November. It predicts how changes in DNA affect gene activity in hundreds of specific human and mouse cell types, rather than averaged tissue, after training on more than 100 million single-cell profiles across 68 tissues. DeepMind says it finds nearly twice as many variants that change gene activity as the first version; that's its own figure. It's available free for non-commercial research through the AlphaGenome API, in private preview for commercial research, and not approved for clinical use. @kimmonismus pointed out it got little attention among the week's model launches.

Why it mattersNot a tool for most businesses, but if you work in biotech, health or pharma, it's a research API worth putting in front of your scientists. For everyone else, it's a reminder that some of AI's biggest payoffs are coming from science, not chatbots.

Sourcesx.com post ↗ · deepmind.google ↗

portless can now give each code branch its own local address

@ctatedev added an opt-in setting to portless, his free tool that gives sites you're developing on your own computer readable addresses instead of port numbers. With the worktree prefix set to "branch", two branches such as feat/checkout and fix/checkout each get their own address instead of clashing on one.

Why it mattersIf several coding agents work on different branches at once, each one's preview gets its own address, so you can check them side by side without mixing them up.

Sourcesx.com post ↗

Describe your processes clearly and you've half-built the automation, says @mattpocockuk

@mattpocockuk says the part of his course that's blowing people's minds isn't the agent material but domain-driven design: agreeing one shared vocabulary for your business, using it with the agent, and renaming the code to match. In a follow-up he put it simply: if you can describe your business processes in clear, consistent language, you've already half-built the software to automate them.

Why it mattersBefore you buy an automation tool, write down how a process works in the words your team actually uses, and use the same words every time. That document is the best brief you can give an agent.

Sourcesx.com post ↗ · x.com post ↗

How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.