Anthropic launched Claude Meetings, a live assistant that drafts plans and pulls data while a meeting is still happening. As participants talk, Claude Meetings can draft the plan being discussed, map a timeline on a shared whiteboard, answer questions in the meeting chat, and pull figures from connected work apps in real time. It is part of a broader office suite Anthropic announced yesterday, a notable expansion for a company that built its reputation on enterprise coding tools. The move puts Anthropic in direct competition with productivity software that has been the territory of established office platforms for decades. (TechCrunch)
Brief
Wednesday, November 11, 2026 · 5 min read
Google opens Gemini 4 Carbon to everyone with agent swarms
Good morning. Anthropic shipped a model, an office suite, and a meeting assistant in roughly 24 hours. Google shipped its strongest model and a healthcare product in the same window. Adobe held its Max conference and announced agents for everything. At some point the question stops being 'what launched today' and starts being 'what did not'.
Today's reading time is 5 minutes.
Listen · Snippt Daily0:00 / 3:41
INDUSTRY
For two weeks, Gemini 4 Carbon existed only for a narrow group of cybersecurity researchers while the rest of the market waited.
Driving the news: Google rolled out Gemini 4 Carbon to the general public yesterday after a two-week restricted release to cybersecurity defenders. The model ships with what Google calls an agent 'swarm' feature, where multiple AI agents coordinate on a task rather than a single model handling it end to end. Google says Carbon is its strongest model yet and positions it directly against the best systems from Anthropic and OpenAI.
- The cybersecurity-first rollout gave Google a two-week window to stress-test the model on adversarial tasks before general availability.
- The swarm feature is the headline capability: agents dividing work among themselves, not just one model with tool access.
Zoom in: Google has been trailing the narrative on frontier models for most of this year, with Anthropic and OpenAI taking most of the coverage around coding benchmarks and enterprise deals. Carbon is Google's answer to that, and the decision to seed it with cybersecurity defenders first is a deliberate positioning move: it frames the model as serious infrastructure rather than a consumer chatbot. The swarm architecture also echoes what several research teams have been publishing about multi-agent coordination, now arriving as a shipped product feature.
- Multi-agent coordination has been a research topic for over a year; Carbon is one of the first frontier releases to ship it as a named, documented feature.
- Google Cloud also launched Gemini for healthcare on the same day, suggesting a coordinated push across verticals rather than a single product announcement.
Why it matters: For anyone building on top of frontier models, Carbon's swarm feature changes what a single API call can orchestrate: instead of chaining agents yourself, the model handles the coordination layer. That removes a significant amount of scaffolding code from complex workflows. The cybersecurity framing also suggests Google is pitching Carbon at enterprise buyers who need to justify the model to a security team before deployment.
Bottom line: Google spent two weeks letting security researchers break Carbon before the rest of us got it, which is either unusual caution or unusually good marketing, and probably both.
The Verge ↗Get this every weekday
Two minutes, 7am ET. No fluff.
Also happening
Adobe went all-in on AI agents at Max, adding MCP support and a Figma-competing prototype builder to Firefly. At its Max conference in Miami, Adobe announced that AI agents can now automate creative workflows across Photoshop, Premiere, and Firefly, with users able to connect their preferred AI assistant via MCP, the open protocol for tool access. A new prototype builder in Firefly takes direct aim at the UI design space. Adobe also announced a deeper integration with Google, allowing PDF editing inside Google Drive and Adobe AI access through Gemini. (TechCrunch)
OpenAI added business profiles to ChatGPT, letting companies surface themselves in AI-generated answers. OpenAI is building out ChatGPT as a consumer discovery engine, with business profiles that let companies control how they appear when users ask relevant questions. The feature makes ChatGPT structurally more like a search product with a commercial layer. Singapore Airlines also published a case study the same day showing its Kris chatbot doubled its customer satisfaction score after integrating OpenAI models. (TechCrunch)
The indie game community fractured over AI after IGF jurors said they would not vote for any game that used generative AI. Several Independent Games Festival jurors published a statement refusing to vote for games that used generative AI in any capacity. Within hours, the developer of acclaimed indie title Big Walk disclosed it had used AI tools sparingly in the final stretch of development. The timing put the studio in an uncomfortable position: the disclosure came after the juror statement, not before. The episode illustrates how quickly a blanket policy collides with the reality of how games are actually made. (The Verge)
The thread
Agents move from feature to product
Three separate stories yesterday, Google's Carbon swarm feature, Anthropic's Claude Meetings, and Adobe's agent layer across its creative suite, all shipped multi-agent or agentic coordination as a named, usable product feature rather than a research demo. The practical effect for builders is that orchestration logic that previously required custom scaffolding is now something you configure, not something you write. When three major platforms ship the same architectural pattern in the same 24-hour window, the scaffolding you built last month starts looking like something the platform will absorb.
From the publisher
Snippt
Image, video and voice tools in one place
On our radar
- Zyphra released ZONOS3-Mini, an Apache 2.0 text-to-speech model with 800M total parameters and 100M active ones that clones voices, allows word-by-word delivery control, and runs in real time in the browser.
- MBZUAI's Institute of Foundation Models released PAN 2, an open Apache 2.0 world model pairing an LLM planner with a 29B diffusion transformer that can simulate up to 20 seconds of action outcomes as video.
- Google Cloud launched Gemini for Healthcare, a product aimed at unifying clinical data and automating administrative workflows to reduce documentation burden on clinicians.
- Cyanogen co-founder is launching Kora, a free alternative to Windows built around AI models and agents, targeting users whose primary computing activity runs through AI tools.
- A new arXiv paper tested 14 model variants across two families to ask whether LLMs actually represent psychological constructs internally or just produce outputs that look like psychometric structure - the answer, per the cross-persona paradigm used, is more complicated than either camp would like.
- Researchers published EarthRecord, a benchmark for measuring how well multimodal agents can turn four decades of satellite imagery into reliable land-change analysis, a task existing benchmarks had not properly measured.
AI worth acting on, in your inbox
Get AI Weekly every Friday: the AI news, tools and tactics that matter for marketers, sellers and creators. Two minutes, with audio.
You can change these choices or unsubscribe at any time. Privacy Policy