NAll news

All news

Friday1454 items

  1. The New StackProduct release
    AI score60

    Anthropic discloses Claude Duet's priority scoring and independent intent verification mechanisms

    Anthropic told The New Stack that Claude Duet is powered by Claude Opus 5.5, can monitor connected applications and handle tasks in the background, and that the company uses an internal scorer to evaluate offline whether its task priority judgments are correct. Anthropic said urgency itself is a subjective judgment, that building evaluations for urgency is imperfect but very helpful, and acknowledged that Duet may misread conversations or misjudge urgency. When performing actions, Duet has a separate set of automated checks extended from the Claude Code auto mode classifier, running independently of the model executing the task, to determine whether actions match user intent; official materials do not state how reliably these checks block out-of-bounds actions, and during the research preview users cannot view or take over each step of its operations in the cloud browser.

    #Product updates#Agents#Anthropic

  2. The New Stack
    AI score41

    The Five-Layer Architecture of AI Agent Sandboxes on Kubernetes

    Running AI agents on Kubernetes requires five layers of sandbox isolation: the isolation layer can use runc, gVisor, or Kata Containers, corresponding to shared kernel, restricted kernel interfaces, and full VM isolation respectively; the node layer requires configuring containerd, KVM, and the guest kernel for the chosen runtime; the access layer uses NetworkPolicy to deny outbound network access by default, issues short-lived credentials per task, and mounts data read-only; the request layer uses RuntimeClass to let agent Pods select the isolated runtime, and factors additional overhead such as the gVisor Sentry and Kata guest kernel into scheduling and quotas.

    #Tutorials / practice#Agents#English text is being prepared.

  3. The New Stack
    AI score48

    GitHub Executive Responds to New Competitor: The World Doesn't Need Another GitHub

    At the Universe '26 conference, GitHub COO Kyle Daigle responded to Entire, a GitHub alternative founded by former CEO Thomas Dohmke, saying, "I don't think the world needs another GitHub"—what developers need is for GitHub to evolve for the agent era. GitHub announced the addition of 4 million CPU cores and 160PB of storage, along with a comprehensive rebuild of its backend infrastructure; it currently serves more than 225 million developers. Dohmke responded that developers clearly need new options, and that Entire aims to make a new home for the community open source, extensible, and distributed across multiple regions worldwide.

    #Industry news#Agents#Open-source ecosystem

  4. The New StackModel releases
    AI score74

    Mistral Open-Sources Mistral Large 4 Weights, Leading GLM-5.3 by 1 Point on Coding Benchmark

    Mistral released the weights of its frontier open-source model Mistral Large 4 (ML4) on Hugging Face on Thursday; the model had only been available in preview three weeks earlier. Mistral calls ML4 "the world's best open-source agentic coding model," scoring 41% on Terminal-Bench 4.0, slightly above GLM-5.3's 40%, and says it uses an average of 47% fewer tokens per task than GLM-5.3; it scored 79% on the CyberGym E2E defensive cybersecurity test, compared with 29% for GLM-5.3. Mistral Vice President of Science Pierre Stock called ML4 merely a "stepping stone" and confirmed that smaller versions will be released in the coming months, one of which can run on a laptop.

    Same story · 15 reports

    #Model releases#Open-source ecosystem#English text is being prepared.

  5. The New Stack
    AI score45

    AWS's James Ward: MCP servers should tell agents what to do next

    James Ward, Principal Developer Advocate at AWS, believes that MCP servers should not just be "data-fetching" APIs, but should include guidance when returning data, telling the agent what it just did and what it can do next. He also pointed out that MCP has no trace ID by default; AWS's approach is to add a session ID parameter to tools, having the agent pass it in on every call, thereby linking individual calls into a complete chain. AWS opened MCP Server to everyone in May this year, and added Lambda diagnostics in September.

    #English text is being prepared.#English text is being prepared.#Agents

  6. The New StackProduct release
    AI score63

    Microsoft Opens Copilot Agent Harness Public Preview via Foundry

    Microsoft has opened the GitHub Copilot agent harness public preview through the Foundry Agent Service in Microsoft Foundry, offering built-in sandboxes, reusable skills, MCP integration, and sub-agent support; the same harness already supports Copilot Code, Copilot Studio, GitHub Copilot, CLI, and VS Code. When creating a prompt agent, this harness must be selected and cannot be changed afterward; CPU, memory, and idle duration are controlled by Microsoft, and bring-your-own-model is not supported. The service is free until December 1, 2026, after which it is billed at $0.0994 per vCPU hour and $0.0118 per GiB hour, with model inference and some tools billed separately.

    #Product updates#Agents#English text is being prepared.

  7. The New StackOpinion
    AI score56

    AWS Hydro Team: Using Compilers to Intercept Coding Errors Outside the Agent Trust Boundary

    Shadaj Laddad, research lead for the AWS Hydro team, believes that code review can no longer keep up with the speed at which agents write code, and advocates placing correctness checks outside the agent trust boundary, enforced by the compiler—for example, the Rust borrow checker catches race conditions before runtime, while Hydro catches multiple classes of defects in distributed systems at compile time. The team is researching how to turn recurring design patterns such as idempotency into general-purpose guardrails, while acknowledging that compilers have limitations: AI-generated Rust code can compile successfully yet still be wrong. Hydro and the same team's open-source agent runtime Infinity are both jointly developed by AWS and UC Berkeley and hosted at hydro.run.

    #Industry news#Agents#English text is being prepared.

  8. The New Stack
    AI score44

    Sam Altman once considered abandoning token-based pricing, but dropped the idea due to customer opposition

    OpenAI CEO Sam Altman revealed that the company had considered abandoning token-based billing for its next-generation models in favor of a model where developers submit a prompt and receive a single price directly, similar to AWS's billing model, but dropped the idea because customer feedback was extremely negative. Customers said they need to know what they are paying for, and would even rather accept inflated token counts and pay higher fees than give up itemized billing they can verify. Altman acknowledged that tokens are "a kind of fake unit" and not a good way to measure usage, but OpenAI still uses token-based pricing for now.

    #Industry news#OpenAI

  9. Minority
    AI score8

    The Wardrobe Always Lacks One More Shirt: A Practical Discussion of Decluttering and Style Rebuilding

    Minority author Hongling shares a wardrobe management method that goes from decluttering to style rebuilding: first organize all clothing by category, use a pyramid layering method to eliminate damaged, ill-fitting, and duplicate items, then give undecided clothing a 30-day cooling-off period. The author then allocates the wardrobe by scenario proportions—for example, work scenarios account for over 70%, leisure 15%–25%, and home only 5%–10%—and plans combinations of basic and stylized pieces accordingly.

    #English text is being prepared.

  10. sspai
    AI score22

    Morning Brief: Apple launches its first touchscreen MacBook Pro, Huawei unveils the Pura 90 Plus and other new products

    Apple has released its first 14-inch and 16-inch MacBook Pro models with touchscreen support, powered by M5 Pro and M5 Max chips, starting at $2,899 (22999 yuan in mainland China), on sale November 6. Huawei launched the Pura 90 Plus, the first device to feature the Kirin 9025 chip, along with the Enjoy 100 series; Microsoft introduced the MAI-Image-2.7-Preview model focused on image editing, ranking second on the AA-Image-Editing v2.1 leaderboard with 1193 Elo.

    #Industry news#OpenAI#Microsoft

  12. Shaoshupai
    AI score16

    Minority Editorial Team's New Gadgets: Cuktech Power Bank Card, HomePod mini 2, Sonos Beam Ultra, and Guyán Prepared Dishes

    In issue 254 of "New Gadgets," the Shaoshupai editorial team shared four products they recently purchased or received for testing. The HomePod mini 2's chip has been upgraded from the S5 to the S10, and Wi-Fi has been upgraded to Wi-Fi 6, but the difference in everyday listening is extremely small compared with the previous generation; under the 27.2 system, AirPlay latency is greatly improved. The Sonos Beam Ultra achieves 7.1.2 Dolby Atmos on its own, and the Sonos 27 system provides a native cloud MCP service that lets commonly used AI directly control speaker playback.

    #English text is being prepared.#English text is being prepared.

  13. Shaoshupai
    AI score33

    Paiday Morning Brief: XREAL AURA Goes on Sale, Tencent Partners with Visa, Tencent Releases Buddy One

    On October 28 local time in the US, XREAL opened overseas orders for the XREAL AURA, the world's first Android XR wired glasses, at a limited-time starting price of $1,099, with the first batch expected to ship in mid-November. Tencent and Visa signed a memorandum of cooperation, planning to deepen collaboration in areas such as cross-border payments, cross-border remittances, enterprise-grade AI, and AI agent commerce. Tencent also released Buddy One, a cross-application AI collaboration system that integrates products including WorkBuddy, Tencent Docs, and Tencent Meeting, with VIP membership at 700 yuan per month on a continuous subscription basis.

    #Industry news#Agents#Google

  14. sspai
    AI score12

    What the author of "Sedentary Health Guide" Ashui has installed: ChatGPT Plus, flomo, and Vibe Coding practices

    Ashui, a sports and health product manager, shared his hardware and software setup: the Apple trio (iPhone 15, Apple Watch S9, a 2,000-yuan M1 MacBook Air), a Yuexingtong X3 e-ink reader, and subscriptions to flomo Max, ChatGPT Plus, Readwise, and NetEase Cloud Music VIP. He uses ChatGPT to write Python scripts to read hardware drawings, sync TestFlight feedback to Feishu Bitable, and built an AI coach within a week through Vibe Coding. He believes that as the development barrier lowers, competition in fitness apps will intensify, and making money won't necessarily become easier.

    #English text is being prepared.#Tutorials / practice

  15. MediaNama
    AI score56

    Sarvam AI Signs MoU with India's Integrated Defence Staff to Develop AI Systems for the Three Armed Forces

    On October 27, Sarvam AI announced that it had signed a memorandum of understanding with India's Headquarters Integrated Defence Staff (HQ IDS) to establish a unified framework for AI projects for the Army, Navy and Air Force. Sarvam will design, develop and deploy AI applications for operations, training and decision support, with the three services and joint agencies sharing the same foundation models, agent infrastructure, and geospatial and perception systems. The work is handled by Chanakya, a firewall division focused on national security, and uses foundation models such as Sarvam 105B. The announcement did not disclose the signing date, signatories, contract value, funding and duration, or the specific project list, nor did it state whether any systems have already been deployed; neither HQ IDS nor the Ministry of Defence has independently confirmed it.

    #Industry news#Agents#English text is being prepared.

  16. MediaNama
    AI score30

    Info Edge Q2 Earnings: Net Profit Up 87% to Rs 650 Million, AI Products Annualized Revenue About Rs 240 Million

    Info Edge announced its Q2 FY2026 results, with net profit up 87% year-on-year to Rs 649.8 million and revenue up 12% to Rs 903.9 million. AI and data products for recruiters and job seekers already account for a high single-digit share of recruitment business revenue, with annualized revenue of about Rs 240 million calculated on a Q2 basis. The company also announced that founder Sanjeev Bikhchandani will take over as Executive Vice Chairman, Managing Director and Group CEO on December 3, 2026, succeeding the resigning Hitesh Oberoi.

    #Industry news

  17. MediaNama
    AI score27

    IndiaMART Q2 profit rises 10% to ₹90.7 crore, revenue ₹428.6 crore

    Indian B2B platform IndiaMART announced its results for the September 2026 quarter (Q2 FY27): net profit rose 10% year-on-year to ₹90.7 crore, and revenue grew 10% to ₹428.6 crore. The number of paid suppliers fell 2% year-on-year to 217,000, active buyers declined 7% to 40 million, and standalone business enquiries dropped 14% to 27 million. CEO Dinesh Agarwal said the decline in enquiries stems partly from Google AI Overviews, and from the company moving some content behind a login to block crawling by search engines and LLMs, and said he is willing to give up the current 10-20% of business enquiries to protect proprietary data.

    #Industry news#Search

  18. MediaNama
    AI score57

    Meesho says AI agents let one engineer do the work of a three-person team, with server costs rising to 2.3% of NMV

    Meesho said on its Q2 FY2026-27 earnings call that AI agents have become its default way of developing software: one engineer leading a group of agents can take on the work of what used to be a three-person team, and 65% of teams are driven by a single engineer. The company said its internal productivity score rose 171% from March to September 2026, code review approval time fell 41%, and the time for a model to enter online testing shrank from 17 days to 3 days. At the same time, AI has also pushed up costs: GPU capacity increased 150%, and server costs as a share of net merchandise value (NMV) rose from 1.9% to 2.3%. The company plans to migrate some workflows to cheaper or open-weight models. Meesho also made clear it will not add a platform fee at the shopping cart stage, even though it loses about 2.4 rupees per order.

    #Industry news#Agents#English text is being prepared.

  19. MediaNama
    AI score16

    CarTrade Q2 Earnings: Net Profit Up 3% to ₹6.6 Billion, Revenue Rises to ₹21.3 Billion

    India's auto classifieds platform CarTrade reported a net profit of ₹6.6 billion in Q2 of FY27, up 3% year-on-year, while revenue grew 10% to ₹21.31 billion, a record high. OLX India's revenue rose 26% year-on-year to ₹7.02 billion, with EBITDA up 36%. In the quarter, CarTrade launched an AI assistant for new car purchases and rolled out the VAYA Live intelligence platform for OEMs and dealers.

    #Industry news#Agents

  20. MediaNama
    AI score27

    Paytm expects UPI MDR to bring in Rs 70 crore per fiscal year

    Paytm Group CFO Madhur Deora said on the Q2 FY27 earnings call that after NPCI imposed a 0.4% MDR on P2M UPI transactions above Rs 2,000, based on the current GMV of Rs 28 trillion, this could bring incremental EBITDA of slightly over Rs 70 crore per fiscal year, potentially reaching Rs 90 crore in FY28. Paytm currently accounts for 8.1% of UPI transaction volume, trailing PhonePe's 45.6% and Google Pay's 32.2%; CEO Vijay Shekhar Sharma said the goal is to capture one-third of the market share.

    #Industry news#Agents