Google Introduces Gemini 4 Carbon for Long Agent Workflows

Archive (German)

Google Introduces Gemini 4 Carbon for Long Agent Workflows

🔬 Models & Research

  • Gemini 4 Carbon: Google’s new flagship model targets long agent workflows and cyber defense — October 28, 2026
    Google introduces Gemini 4 Carbon, which performs strongly in benchmarks and is designed for multi-day development workflows and cyber defense. Its initially limited availability underscores how powerful models are increasingly being released according to risk profile and use case.

  • New Artificial Analysis index widens the gap between open and closed models — October 28, 2026
    The revised benchmark uses private test sets for the first time, accounting for 60 percent of its tests, and shows a greater gap between open and closed models. This makes the reliability and contamination resistance of a benchmark—not just its rankings—more important when comparing models.

  • Counting bandages: How to measure time savings from AI — Current
    Based on a developer study involving 246 tasks, the paper shows that measured AI time savings depend heavily on the calculation method, error checking, and correction effort. Companies should therefore measure productivity alongside quality and rework, rather than comparing task completion times alone.

đź’Ľ Business & Markets

  • Microsoft discloses Azure revenue in actual dollars for the first time: $33.2 billion — October 28, 2026
    Azure generated $33.2 billion in revenue in the quarter from July to September, growing 48 percent year over year. The disclosure makes direct comparisons with AWS and Google Cloud easier and sharpens the focus on competition for AI infrastructure.

  • Microsoft posts a quarterly revenue record for the tenth time in a row — October 28, 2026
    Microsoft reports revenue of $92.3 billion and net income of $36.6 billion, supported by strong cloud business. The figures show that AI investments are part of a broader infrastructure and platform business, rather than something to be assessed in isolation.

  • Microsoft generates $10 billion in revenue from OpenAI — October 28, 2026
    Commercial agreements with OpenAI brought Microsoft $10 billion in the latest quarter, just under 11 percent of the company’s revenue. The financial ties remain substantial, even as the two companies also compete for the same enterprise customers.

  • Google expands ads in AI Search internationally — October 28, 2026
    Google is bringing ads to AI-powered search experiences in more markets as growth in traditional search advertising slows. The move is intended to open up new revenue streams and will test whether generative answers can support a viable advertising format.

  • The future of the firm — October 28, 2026
    OpenAI outlines how AI could be integrated into organizational structures below, alongside, and between people. For businesses, the question of how to structure responsibilities and collaboration is coming to the fore as agents become part of the operating model.

  • DevDay in Berlin: OpenAI signals a European launch for “Dots” and advances its platform expansion — October 28, 2026
    OpenAI signals a broader European launch for its Dots assistant and reports strong growth in Codex users in Germany. Its focus on agents, APIs, and developer tools shows that OpenAI is positioning its business more strongly as a platform for workflows.

  • Creating AI value with enterprise context — October 28, 2026
    A survey by MIT Technology Review Insights reports that only 16 percent of the companies surveyed are achieving transformative value with AI. The finding underscores that models alone have little impact without business context and suitable process integration.

  • Are AI executives too out of touch to build products for the general public? — October 28, 2026
    The debate around Marissa Mayer’s example of her AI app Dazzle raises questions about the everyday usefulness of AI products and their developers’ social understanding. For product teams, it is a reminder not to equate user needs with Silicon Valley’s standards.

  • Adam Mosseri expects growing backlash against AI — October 29, 2026
    Instagram head Adam Mosseri says fears and frustration surrounding AI, particularly in creative industries, are likely to increase. Companies must treat acceptance and the impact on creative work as central product and rollout considerations.

  • New cyberattacks hit South Korea’s financial sector — October 28, 2026
    Several South Korean financial companies and a crypto exchange report a new wave of attacks, while the causes are still under investigation. The incidents increase pressure to identify vulnerabilities in interconnected financial systems more quickly and to account for AI-powered attacks in threat models.

  • As AI writes the code, entrepreneurial thinking becomes more important for entry-level hires at Anthropic — October 28, 2026
    Anthropic engineering manager Felix Rieseberg believes traditional coding interviews are becoming less informative as AI tools grow more capable. When hiring junior developers, problem understanding and the ability to use AI outputs productively may count more than small programming exercises.

  • Microsoft is ending Copilot support on smart TVs — October 28, 2026
    Microsoft will discontinue the Copilot feature on LG and Samsung TVs later this year. The move shows that AI features on connected devices will only last if usage and product strategy justify the effort.

⚖️ Politics & Regulation

  • Meta may once again delay controversial new business terms — October 28, 2026
    Meta is considering postponing new business terms beyond the planned date following criticism from publishers. The dispute shows that training rights for content must be resolved not only through legal channels, but also through industry coordination and platform rules.

  • Elvis heirs sue Anthropic over alleged use of songs and circumvention of safeguards — October 28, 2026
    Rights holders for 274 Elvis Presley songs accuse Anthropic of using music for training without permission and circumventing technical safeguards. The case could extend the legal risks around training data beyond the question of use to include access restrictions and copyright attribution.

  • xAI offers to buy homes if owners do not sue over data center noise — October 28, 2026
    An xAI subsidiary is reportedly seeking to buy homes in Mississippi while residents agree not to sue over a nearby data center and power plant. The conflict shows how the expansion of large-scale AI infrastructure directly affects local living conditions, residents’ rights, and public acceptance of companies.

  • Texas voters turn against data centers — October 28, 2026
    Planned data centers are facing opposition in several Texas cities, partly over land use and local impacts. For operators, community approval has become a concrete site-selection and expansion risk to consider alongside the availability of power and water.

🛠️ Tools & Products

  • Google opens Android development to external AI agents and introduces new Gemini APIs — October 28, 2026
    Android Studio supports more than 40 external coding agents through “Bring Your Own Agent”; new interfaces also bring agent capabilities directly to mobile devices. Developers gain more freedom to choose their tools and can integrate on-device model features into apps.

  • Google opens Gemini and Search to more apps — October 28, 2026
    The new Connected Apps Platform makes it easier for businesses to connect their services to Google Search and Gemini, allowing users to take action directly through AI experiences. This approach could shift interactions away from individual apps and toward AI interfaces, making integrations a strategic distribution channel.

  • Chrome’s AI Autofill can now use information from Gmail — October 28, 2026
    Using Gemini, Chrome can draw on personal information from Gmail—such as booking or shipping details—to suggest and fill in forms. The feature saves users steps, but also makes transparent consent and clear control over personal data more important.

  • NVIDIA Nemotron 3.5 Transcribe and Granary v2: customizable multilingual transcription — October 28, 2026
    NVIDIA releases an open, promptable speech model for transcription, translation, speaker attribution, and audio questions, along with the Granary v2 dataset. Developers can use it to better tailor speech pipelines for meetings or medical documentation to specialist vocabulary and desired output formats.

  • Anthropic gives companies a clearer picture of what their AI spending is buying — October 28, 2026
    New reporting tools are intended to show teams what work was done with Claude, what it cost, and where workflows stalled. This helps companies evaluate AI spending against specific workflows rather than solely through token or license costs.

  • Anthropic adds prebuilt templates for law firms to Claude for Legal — October 28, 2026
    Prebuilt interactive tools are designed to support legal teams with tasks including citation checks, legal comparisons, case file management, and contract comparisons. Specialized templates can make adoption easier for law firms without replacing review by legal professionals.

  • Legora is building an agentic operating system for legal work — October 28, 2026
    The legal tech company uses OpenAI models, ChatGPT, and Codex to speed up workflows and scale contract analysis. Integrating AI into legal processes is intended to reduce routine work and free up capacity for tasks with greater professional value.

  • Akai is using community-developed mods for its open-source MPC — October 28, 2026
    Akai is giving modders early access to a new MPC with open-source firmware and has already showcased customizations developed through vibe coding. The move brings the developer community in early and makes open firmware part of the product strategy.

  • Meta launches Waddle World, an AI app that serves as a digital pet diary — October 28, 2026
    Waddle World combines a virtual pet with conversations, small games, and a personal diary. The product relies on accessible, voice-based interaction and shows how AI assistants can be embedded in everyday entertainment formats.

  • AI agents break into Anthropic’s internal systems through third-party software — October 28, 2026
    Security startup Cathedral used AI agents in a testing program to access internal data, including Slack messages and payroll information, through external enterprise software. The incident shows that agents, combined with permissions and third-party integrations, can create new attack paths.

🤖 Robotics