<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Новости Радио-Т</title>
    <link>https://news.radio-t.com</link>
    <description>новости информационных технологий</description>
    <managingEditor>umputun@gmail.com (Umputun)</managingEditor>
    <pubDate>Sat, 21 Nov 2026 03:20:50 -0600</pubDate>
    <item>
      <title>Gemini 4 Neon</title>
      <link>https://news.radio-t.com/post/gemini-4-neon</link>
      <description>Gemini 4 Neon launches across Google AI products, offering a one-million-token context window and pricing of $3 per million input tokens and $15 per million output tokens. It targets software engineering, enterprise, scientific and cybersecurity work, with strong benchmark results, lower task costs than competing frontier models, and safeguards against misuse, misalignment and prompt injection.</description>
      <guid>https://news.radio-t.com/post/gemini-4-neon</guid>
      <pubDate>Sat, 21 Nov 2026 07:49:14 +0000</pubDate>
    </item>
    <item>
      <title>We promoted developers into architects without informing them</title>
      <link>https://news.radio-t.com/post/we-promoted-developers-into-architects-without-informing-them</link>
      <description>AI-generated logging made a Google Forms add-on’s welcome screen take 15 seconds by making multiple API calls to retrieve user details. AI follows requirements without weighing context or challenging costly choices, so developers must now define architecture, design data models, set constraints for AI, and review its code. They should balance performance, security, and business needs, and push back on requirements when needed.</description>
      <guid>https://news.radio-t.com/post/we-promoted-developers-into-architects-without-informing-them</guid>
      <pubDate>Fri, 20 Nov 2026 19:41:34 +0000</pubDate>
    </item>
    <item>
      <title>Meet the New Docker: Built to Run on Trust | Docker</title>
      <link>https://news.radio-t.com/post/meet-the-new-docker-built-to-run-on-trust-docker</link>
      <description>Docker has launched a new brand focused on trust to address the evolving challenges posed by AI agents. While maintaining its legacy of openness, simplicity, and container technology, Docker is shifting to provide enhanced security, isolation, and governance for autonomous agents. This evolution includes a refreshed brand identity featuring an updated Moby whale logo, symbolizing Docker&#39;s growth beyond containers into the agentic era. Docker invites users to explore its new offerings like Docker Sandboxes and AI Governance to experience trusted agent management.</description>
      <guid>https://news.radio-t.com/post/meet-the-new-docker-built-to-run-on-trust-docker</guid>
      <pubDate>Fri, 20 Nov 2026 05:23:50 +0000</pubDate>
    </item>
    <item>
      <title>Java4Shell Flaw (CVE-2026-114938) Enables Pre-Auth RCE in Jira, Jenkins, Keycloak and More</title>
      <link>https://news.radio-t.com/post/java4shell-flaw-cve-2026-114938-enables-pre-auth-rce-in-jira-jenkins-keycloak-and-more</link>
      <description>Java4Shell (CVE-2026-114938) could enable unauthenticated code execution in six applications on affected AVX-512 Java servers.</description>
      <guid>https://news.radio-t.com/post/java4shell-flaw-cve-2026-114938-enables-pre-auth-rce-in-jira-jenkins-keycloak-and-more</guid>
      <pubDate>Wed, 18 Nov 2026 10:55:27 +0000</pubDate>
    </item>
    <item>
      <title>Microsoft says its new MAI Code 2 does 96% of Claude Opus 5.5’s work for 5% of the price</title>
      <link>https://news.radio-t.com/post/microsoft-says-its-new-mai-code-2-does-96-of-claude-opus-5-5-s-work-for-5-of-the-price</link>
      <description>Microsoft’s MAI Code 2, Cyber 2 Flash and Image 2.7 launched at Ignite, claiming near-OpenAI and Anthropic scores at a fraction of the price.</description>
      <guid>https://news.radio-t.com/post/microsoft-says-its-new-mai-code-2-does-96-of-claude-opus-5-5-s-work-for-5-of-the-price</guid>
      <pubDate>Tue, 17 Nov 2026 21:32:41 +0000</pubDate>
    </item>
    <item>
      <title>Introducing GPT-6.2 Astra and Sol | OpenAI</title>
      <link>https://news.radio-t.com/post/introducing-gpt-6-2-astra-and-sol-openai</link>
      <description>OpenAI introduces GPT-6.2 Astra and Sol, advanced AI models designed for professional and everyday tasks. GPT-6.2 Astra excels in complex workflows, scientific research, and creative design, offering improved performance and cost efficiency over previous versions. GPT-6.2 Sol is optimized for versatile, efficient everyday work with strong coding and collaboration capabilities. Both models feature enhanced cybersecurity, safety measures, and alignment, and are available via ChatGPT Work and OpenAI API with competitive pricing.</description>
      <guid>https://news.radio-t.com/post/introducing-gpt-6-2-astra-and-sol-openai</guid>
      <pubDate>Tue, 17 Nov 2026 18:53:46 +0000</pubDate>
    </item>
    <item>
      <title>A year after the 100% AI mandate nobody understands our code</title>
      <link>https://news.radio-t.com/post/a-year-after-the-100-ai-mandate-nobody-understands-our-code</link>
      <description>The linked story appears to describe a company that mandated heavy AI coding, and a year later its developers say they struggle to understand the growing codebase. Commenters describe rushed, AI-generated projects, accumulated technical debt, bugs, and developers who feel they are learning less or losing motivation. Some say senior staff have also left, which would make the code hard to understand even without AI.&#xA;&#xA;Many blame management for prioritizing visible output and cost-cutting over software design and engineering skills. Others argue that codebases were often poorly understood before AI, and that the post may be fabricated or AI-written engagement bait. Suggested responses include slowing feature work to refactor and document the code, or rebuilding from clear requirements. There is also debate over whether companies should be publicly called out for poorly run AI mandates.</description>
      <guid>https://news.radio-t.com/post/a-year-after-the-100-ai-mandate-nobody-understands-our-code</guid>
      <pubDate>Tue, 10 Nov 2026 18:27:56 +0000</pubDate>
    </item>
    <item>
      <title>The Zig project is going nowhere</title>
      <link>https://news.radio-t.com/post/the-zig-project-is-going-nowhere</link>
      <description>Zig’s prolonged pre-1.0 development and frequent breaking changes make it costly to rely on. Its strict ban on AI-assisted contributions rejects work based on how it was produced, not just its quality. Bun’s move from Zig to Rust exposed technical and maintenance concerns, while Zig’s hostile response damaged confidence in its management. Moving from GitHub to Codeberg adds further reliability concerns.</description>
      <guid>https://news.radio-t.com/post/the-zig-project-is-going-nowhere</guid>
      <pubDate>Sat, 14 Nov 2026 18:54:47 +0000</pubDate>
    </item>
    <item>
      <title>Critical thinking when agents do the work | Addy Osmani (@addyosmani) on X</title>
      <link>https://news.radio-t.com/post/critical-thinking-when-agents-do-the-work-addy-osmani-addyosmani-on-x</link>
      <description>Back in the day, critical thinking was taught as approaching new info with humble curiosity and a healthy dose of skepticism - take in different views, form your own opinion. Still true, but increasingly, nothing in my day-to-day forces critical thinking on me now that so much of my work runs ...</description>
      <guid>https://news.radio-t.com/post/critical-thinking-when-agents-do-the-work-addy-osmani-addyosmani-on-x</guid>
      <pubDate>Sat, 14 Nov 2026 17:47:39 +0000</pubDate>
    </item>
    <item>
      <title>Gemini Decisions API</title>
      <link>https://news.radio-t.com/post/gemini-decisions-api</link>
      <description>Google introduced the Gemini Decisions API, powered by Gemini Decision 1, for evaluating up to 1,000 yes/no, multiple-choice, or rubric-scoring questions against shared multimodal input. It processes text, images, audio, video, and PDFs, returning probabilities and confidence scores in 150–250 milliseconds. Pricing is $0.10 per million input tokens, with no output-token charges. Google teams are using it for model routing, design, emoji selection, and safety checks.</description>
      <guid>https://news.radio-t.com/post/gemini-decisions-api</guid>
      <pubDate>Sat, 14 Nov 2026 07:41:30 +0000</pubDate>
    </item>
    <item>
      <title>Big tech is here to let us down</title>
      <link>https://news.radio-t.com/post/big-tech-is-here-to-let-us-down</link>
      <description>Microsoft outages affecting Outlook, Teams and Copilot disrupted businesses, schools, universities and courts across multiple days. Heavy reliance on Microsoft systems leaves institutions vulnerable when services fail, while staffing cuts may hinder repairs. Compensation for resulting losses is unlikely. Some European public institutions are switching to open-source alternatives to reduce dependence on Microsoft.</description>
      <guid>https://news.radio-t.com/post/big-tech-is-here-to-let-us-down</guid>
      <pubDate>Sat, 14 Nov 2026 07:38:25 +0000</pubDate>
    </item>
    <item>
      <title>Anthropic upgrades Claude Code Projects to run parallel AI workers</title>
      <link>https://news.radio-t.com/post/anthropic-upgrades-claude-code-projects-to-run-parallel-ai-workers</link>
      <description>Claude Code Projects now coordinate parallel Claude Code cloud sessions that handle tasks across repositories, run tests, and open pull requests. A coordinator assigns and reviews work, while shared memory and a project library preserve decisions and artifacts. The updated projects are in beta for select Pro and Max subscribers, with broader access planned. Threads may use limits faster.</description>
      <guid>https://news.radio-t.com/post/anthropic-upgrades-claude-code-projects-to-run-parallel-ai-workers</guid>
      <pubDate>Sat, 14 Nov 2026 07:37:53 +0000</pubDate>
    </item>
    <item>
      <title>Goldman Sachs is pulling Claude Code from its bankers and traders</title>
      <link>https://news.radio-t.com/post/goldman-sachs-is-pulling-claude-code-from-its-bankers-and-traders</link>
      <description>Goldman Sachs has limited Claude Code and GitHub to engineers and strats, citing productivity and risk concerns. Traders and bankers had used the tools to track client flows and model markets. The bank’s internal AI assistant remains available firmwide. The restriction follows earlier limits on Claude in Hong Kong and comes as Goldman helps underwrite Anthropic’s IPO.</description>
      <guid>https://news.radio-t.com/post/goldman-sachs-is-pulling-claude-code-from-its-bankers-and-traders</guid>
      <pubDate>Tue, 10 Nov 2026 18:25:34 +0000</pubDate>
    </item>
    <item>
      <title>Why I will not be using LLMs in my coding</title>
      <link>https://news.radio-t.com/post/why-i-will-not-be-using-llms-in-my-coding</link>
      <description>LLMs will not be used for professional coding because of concerns about cognitive effects, unreliable and verbose code, poor maintainability, environmental costs, and dependence on US-based companies and potential GDPR risks. Although AI tools can sometimes help when implementation stalls, the benefits are not worth these drawbacks. AI-assisted coding may persist, but its use is not inevitable.</description>
      <guid>https://news.radio-t.com/post/why-i-will-not-be-using-llms-in-my-coding</guid>
      <pubDate>Mon, 09 Nov 2026 06:45:23 +0000</pubDate>
    </item>
    <item>
      <title>The Lisp curse, now coming to your ecosystem</title>
      <link>https://news.radio-t.com/post/the-lisp-curse-now-coming-to-your-ecosystem</link>
      <description>LLM-assisted coding may encourage developers to generate custom solutions instead of adopting open-source libraries, reducing library use, maintainer feedback, and new project creation. These effects could reinforce each other as ecosystems offer fewer shared tools. Proven libraries remain valuable, but widespread code generation could weaken collaboration and make software development more isolated.</description>
      <guid>https://news.radio-t.com/post/the-lisp-curse-now-coming-to-your-ecosystem</guid>
      <pubDate>Sun, 08 Nov 2026 08:28:39 +0000</pubDate>
    </item>
    <item>
      <title>Someone behind the software</title>
      <link>https://news.radio-t.com/post/someone-behind-the-software</link>
      <description>Good products reflect their makers’ experience, judgment, and care. As agents make software easier to build, independently produced features can pile up without a shared perspective, leaving products messy and incoherent. Strong products require understanding customers, making deliberate choices, and continually shaping features into a unified whole.</description>
      <guid>https://news.radio-t.com/post/someone-behind-the-software</guid>
      <pubDate>Sat, 07 Nov 2026 06:06:06 +0000</pubDate>
    </item>
    <item>
      <title>Bug Report: Privacy violation in User-Agent header · Issue #107757 · anthropics/claude-code</title>
      <link>https://news.radio-t.com/post/bug-report-privacy-violation-in-user-agent-header-issue-107757-anthropics-claude-code</link>
      <description>A bug report on the Claude Code GitHub repository highlights a serious privacy issue where users&#39; email addresses are leaked in the User-Agent header during web searches without consent. The reporter emphasizes that this behavior should be completely removed rather than opt-in or opt-out. The issue has drawn significant community attention and reactions, prompting the introduction of a built-in privacy mod in version 2.1.317 to block this behavior by default. Users remain concerned about privacy and data security implications.</description>
      <guid>https://news.radio-t.com/post/bug-report-privacy-violation-in-user-agent-header-issue-107757-anthropics-claude-code</guid>
      <pubDate>Sat, 07 Nov 2026 19:04:34 +0000</pubDate>
    </item>
    <item>
      <title>AI is destroying my CS degree</title>
      <link>https://news.radio-t.com/post/ai-is-destroying-my-cs-degree</link>
      <description>A computer science student loves studying but says the university’s inconsistent AI rules, AI-generated coursework, errors, excessive workloads and automated student monitoring are worsening education and overloading staff. Career doom headlines add pressure, though interest in computing remains strong, including exploring topics beyond AI.</description>
      <guid>https://news.radio-t.com/post/ai-is-destroying-my-cs-degree</guid>
      <pubDate>Sat, 07 Nov 2026 06:00:54 +0000</pubDate>
    </item>
    <item>
      <title>Resisting Cognitive Surrender</title>
      <link>https://news.radio-t.com/post/resisting-cognitive-surrender</link>
      <description>Uncritical reliance on AI agents can accelerate development while increasing complexity, technical debt, bugs, and erosion of engineers’ understanding and judgment. Since accountability remains with the builder, delegation should be deliberate: reserve agent-driven work for low-risk, testable projects, set complexity limits, monitor automation bias, stay involved in important decisions, and use tools that make changes easier to comprehend and review.</description>
      <guid>https://news.radio-t.com/post/resisting-cognitive-surrender</guid>
      <pubDate>Sat, 07 Nov 2026 05:57:38 +0000</pubDate>
    </item>
    <item>
      <title>Claude Code sends your email address to websites</title>
      <link>https://news.radio-t.com/post/claude-code-sends-your-email-address-to-websites</link>
      <description>Claude Code sends users’ email addresses to websites.</description>
      <guid>https://news.radio-t.com/post/claude-code-sends-your-email-address-to-websites</guid>
      <pubDate>Sat, 07 Nov 2026 05:55:45 +0000</pubDate>
    </item>
  </channel>
</rss>