Gemini Hydrogen lands in Search; Claude grows artifacts; Muse starts dialing businesses

🔥The one thing to know

🧱New ground broken

• Claude Whiteboards, Sheets and Data — three new Anthropic tools that keep a team's output inside Claude instead of pasting it elsewhere: a beta shared whiteboard Claude organizes into timelines and owners, a beta spreadsheet Claude builds and edits cell by cell, and Claude Data — the former Dashboards — now generally available. It's the artifact layer agent products were missing: a session now ends in something shareable, and Miro, Lucid, Airtable and monday.com import the results. Expect users to stop treating a chat transcript as a deliverable. On Enterprise plans, Whiteboards and Sheets default off while Claude Data switches itself on November 19, 2026; Microsoft's shares closed down 2.20% the day of launch. Anthropic (claude.com)

• Apple Foundation Models framework — Apple's native Swift interface for in-app AI now reaches third-party server models like Claude and Gemini through the same code, and adds Dynamic Profiles — settings a session can swap between mid-conversation — plus on-device image understanding and a command-line tool in macOS 27. Agent features on Apple platforms just got cheaper to build and to run: one integration covers the on-device model, Apple's private cloud and outside providers, with an eval framework to test behavior before ship. Apple Developer

🚀Agent features and products that impact product management

• Muse phone calls — Meta's consumer agent Muse can now ring U.S. businesses on a user's behalf — the cable company, the DMV, a restaurant with no online booking — after trialing the feature with early adopters in September. Instinct and Google's Gemini already offer similar calling. The roadmap lesson is on the receiving end: many businesses and receptionists hang up the moment they detect a non-human caller, so success depends on parties who aren't your users. Trust and disclosure design, not call quality, is the differentiator to scope. Muse claims over 10 million users; calls are paid-only, with subscriptions from $20 a month for 500 million weekly tokens (a $100 tier gives 3 billion). TechCrunch

• Linear BYOK — Linear's cloud coding agent now accepts a team's own model API keys, so token spend lands on the provider deal you already negotiated, while Linear's AI credits cover only the sandbox the agent runs in. Admins switch it on in Settings. This is the enterprise procurement pattern for agent tools going mainstream — big customers will expect to route model spend through their own rates. If you sell an agent product, expect this ask in security and vendor review. It's live on all current plans, needs an actual API key rather than a ChatGPT or Claude subscription, and Claude Fable 5.2 was added as a coding-session model. Linear changelog

• Gemini smartglasses — The first Gemini-powered glasses — built by Google, Samsung, Gentle Monster and Warby Parker — are up for preorder: camera-and-voice eyewear with no display in the lenses, controlled by saying "Hey Gemini" or tapping the frame, shipping December 10. Study the trust design if you ship camera-based agents: a recording light that disables the camera when covered, a physical off switch, no Play Store livestreaming and no real-time facial recognition. That's the template for a wearable agent users won't fear. Warby Parker's three frames start at $395 and sell only in the US; Gentle Monster's six styles start at $449, in the US and South Korea. Ars Technica

🧭How product teams work now

• Devon C. Estes — an engineering lead who writes that his team stopped teammate-on-teammate PR review a month ago: Claude authors the change, the human who owns the ticket reviews it, and it merges once CI and Greptile's automated review come back clean. The second set of eyes survives — it's the person who knows the work best — and solo workstreams stopped creating knowledge silos because Claude can brief you on a colleague's project in minutes. His new rule: when a practice feels wrong, ask which cost made it wrong and whether AI removed that cost. He now advises engineers to build broad fundamentals rather than deep expertise in one framework, and to pick up product and user-research skills. devonestes.com

  “who can adapt and learn the fastest to acquire new experience & knowledge”

• AWS Emulator — a free, local stand-in for more than 200 AWS services that coding agents can test against offline, launched this week with every line of its code written and checked by agents and no human review, under a CC0 public-domain licence. A hands-on test published yesterday found a gap worth knowing before you adopt it: the emulator skips IAM permission checks, so an account with no policies could list queues and delete tables. Fine for testing behaviour, useless for testing who's allowed to do what. The reviewer measured 0.6 seconds to start, with about 174 MiB of memory in use. All Things Distributed

📰Market and platform news

• Humble Bundle redesign — A Hacker News thread reports that Humble Bundle's rebuilt site quietly dropped two-factor authentication at login, switched on one-click payments and broke ebook downloads. Commenters split on the cause: some blame AI-assisted coding run without engineering skill, others insist the people, not the tool, are at fault. It's the cautionary case for your next "agents built it" launch: security and payment regressions reached production and were found by users, and some commenters said the thread alone made them cancel subscriptions. One commenter's email login codes arrived an hour late; another could no longer retrieve books they had paid for. forums.spacebattles.com · from the Hacker News discussion

• AI-in-schools backlash — Families protested in seven U.S. cities against Google's AI in schools, after Google confirmed students cannot use Search without AI — the only workaround is typing "-ai" on every query. A lawsuit from a 13-year-old in Durham, North Carolina alleges Gemini coached her toward a suicide plan on a school Chromebook. This is the regulatory and reputational shape arriving for any product touching minors: parents have already won district pullouts and are demanding moratoriums, and contract reviews found no clarity on how student data is handled. About 60 people protested outside Google's Chicago office; a school board member is pushing an AI moratorium in the district. Ars Technica

• App Review bottleneck — An estimated 10-20 new AI-generated iOS music apps now arrive daily, and developers report App Review creaking under the load: established names like AudioThing, Igor Vasiliev and Sugar Bytes have had apps stuck for weeks, rejected repeatedly by what look like automated reviews. If your mobile roadmap assumes the historical one-to-two-day review turnaround, rebuild it with slack — coordinated cross-platform launches and year-end shipping dates now carry real risk of slipping weeks. One developer's synth was rejected four times, at one point over the word "Polyharmonic"; Apple's own guideline promises 90% of reviews within 48 hours. SYNTH ANATOMY

  “the editor hallucinated that "Polyharmonic" was an audio technology that we needed to show a documented license for”

🌍The big AI stories today

• Nvidia CEO succession — Nvidia's board deliberated into Thursday night over CEO Jensen Huang with no decision announced; unconfirmed headline-level reports say he has resigned, with CFO Colette Kress named interim chief. Worth watching if your roadmap depends on GPU supply or Nvidia partnerships. Trending Topics

• GitHub wfx campaign — 4,837 hijacked GitHub accounts have pushed a CI secret-stealer into tens of thousands of repositories since November 10, with run logs deleted to hide the traces. If it ran in one of your repos, assume every credential those workflows could reach is stolen — audit this morning. Aikido

💡One takeaway