AI News, Nov 11: Anthropic Builds an Office Suite
The Big Story: Anthropic built an office suite a week before it prices its IPO
Anthropic spent Tuesday announcing what a model company announces when it wants your company’s work rather than just its API calls. Claude Sheets and Claude Whiteboards arrived in beta on paid plans: a spreadsheet Claude builds and teams keep editing, and a shared board it can turn into a doc or a deck. Claude Meetings, in beta on Team and Enterprise, joins Google Meet calls as a visible participant and starts on tasks while people are still talking. Claude Data, the renamed Dashboards product, went generally available, and interactive connectors now open full versions of apps like Adobe, Atlassian and HubSpot inside Claude itself.
The strategic claim sits in Anthropic’s own framing, quoted by The Register: “Inside Anthropic, non-technical teams are starting to get as much out of Claude as our engineers.” Claude already had Docs, Slides, Design and Motion. The new surfaces finish a product line pointed straight at Microsoft 365 and Google Workspace, built by a company whose money has come from developer tooling.
Hacker News put its skepticism on precedent: Claude Design launched to real enthusiasm and settled in as a fallback while serious work went back to Figma. The sharper point is that a model vendor hosting your analytics layer cannibalizes its own API customers, which is what a company does when core revenue growth slows.
Six rounds closed in a day for supervising agents, none for building one
Mercor raised $750 million at a $20 billion valuation, double its October 2025 mark, with Nvidia, Salesforce and Zoom participating and no lead named; it is moving from frontier labs to enterprises, scoring agents against each customer’s own standards. Daylight took $80 million on 40-fold revenue growth for detection and response run by agents its engineers tune. Lumbra, founded by a former CIA officer and a former special forces officer, raised $25 million for agent orchestration on classified networks. WhiteMatter raised $20 million for a guardian-agent layer policing other agents from validation through runtime. Plurai raised $13 million to build a simulated copy of a company’s systems and test agents against edge cases, a method brought over from self-driving cars. Scalestack took an undisclosed round for a gateway enforcing enterprise rules, spending limits included, on agent requests to CRMs and warehouses.
Nobody funded an agent this window. They funded what has to exist before you can trust one, and Cowbell started selling insurance against the gap, triggered by an “AI Incident”: model drift, agent manipulation, runaway usage charges, LLMjacking.
Today’s Top Stories
Google made Gemini 4 Carbon generally available at a quarter of Claude’s API price
Carbon, until now restricted to a cybersecurity-defender program, is live for paying developers through the Gemini API and for AI Ultra subscribers in Antigravity, the Gemini app and AI Mode in Search. Introductory pricing is $2 per million input tokens and $10 per million output through November 30, rising to $4 and $20. Claude Fable 5.2 lists at $8 and $40, GPT-6 Astra at $10 and $50. Antigravity also gets a /swarm preview that turns one prompt into a coordinated agent team. Hacker News took issue with distribution rather than capability: paying Gemini Pro subscribers report still being served 3.1 Pro.
Nearly every app ever built with Grok Build was readable by anyone
Security firm Claroc found more than 35,000 of 37,571 live Grok-built apps, nearly 95%, reachable without logging in, because the publish dialog preselected “Anyone with the link.” Twenty-four belonged to SpaceXSI and X employees, among them a sales territory map covering 594 accounts and an onboarding plan marked confidential. Tesla turned up with a service-center scorecard naming 22 technicians and their bonus status. Elsewhere: an ICU briefing with a patient’s lab values, and a timesheet app whose source carried 18 accounts with plaintext passwords. Claroc, which sells security products for AI-built apps, puts the number holding private data at 650 to 800.
OpenAI’s revenue grew about 50% in a quarter and the loss still widened
Internal figures reviewed by the Wall Street Journal put OpenAI’s Q3 revenue at about $10 billion, up roughly half from $6.7 billion in Q2, against 17% to 18% growth the quarter before. Enterprise sales doubled and are now most of the business. The adjusted operating loss still widened to $10.9 billion from $10.1 billion, though the margin improved to −109% from −150%. ChatGPT is at 1.2 billion weekly users, 100 million of them paying. President Greg Brockman called it “an incredible revenue turning point.” For scale, Anthropic booked $17 billion of revenue in the same quarter.
OpenAI turned ChatGPT into a claimable business directory
Businesses can now claim their ChatGPT profile, edit what shoppers see, and read 28 days of clicks and mentions in a Business Hub. Eligible US ecommerce and online-service companies can publish a Business Agent, a seller-run representative answering questions about its own products and policies, with ChatGPT labelling the handoff when a shopper crosses over. Verification runs through a business phone, email, domain or Shopify account, then a Persona identity check. It is free at launch, per a Digiday briefing with monetization VP Benji Shomair, who left charging open as an option. A Business Agent can already be the destination for a ChatGPT ad.
Quick Hits
- IPO: Anthropic’s roadshow opened on 825 million shares at $70 to $78, trading as ANT on Nasdaq, with pricing expected November 18; its seven co-founders are each selling about $250 million worth.
- Governance: Nvidia’s independent directors retained Wachtell, Lipton, Rosen & Katz, led by former acting US attorney Joon H. Kim, to run the Huang investigation; Morgan Stanley dropped the stock as a top pick after a 25% slide.
- Supply chain: TSMC’s board approved $500 million for Anthropic shares at the IPO price and $500 million for OpenAI preferred, reversing CEO C.C. Wei’s July answer on investing in customers: “So far, no.”
- Diplomacy: Treasury Secretary Scott Bessent said the US and China have stood up a dialogue mechanism on “super intelligence” to guard against shared risks, with no commitment to limit development.
- States: New York opened the Center for AI Responsibility and Research at Binghamton University, where 80-plus scientists will test models on the state’s $500 million Empire AI compute; it can only study what companies disclose publicly.
- Liability: An Australian review, ordered after OpenAI agents reached non-public government medical records, returned 18 recommendations including criminal penalties and a “failure to prevent” offence.
- Robotics: DeepMind’s Gemini Robotics 2.5 had Apptronik Apollo 2 humanoids kitting automotive parts at a 96% success rate in European factories, and shipped ASIMOV-Jailbreak, 1,500 open prompts for pushing robots into unsafe actions.
- Plumbing: Adobe released a local MCP server exposing Illustrator, Photoshop, After Effects and Premiere to ChatGPT, Claude and Gemini, with VP Deepa Subramaniam saying Adobe has “no need or desire” to limit what the endpoint exposes.
- Earnings: CoreWeave posted $3.7 billion of Q3 revenue, up 168%, and raised full-year guidance, sending the stock up about 15% premarket weeks after its credit spreads hit distressed levels.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
