<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
        <title>News | Estéban Soubiran</title>
        <link>https://soubiran.dev/news</link>
        <description>My personal technology monitoring feed.</description>
        <lastBuildDate>Wed, 18 Nov 2026 15:44:07 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <copyright>© 2026 Estéban Soubiran</copyright>
        <atom:link href="https://soubiran.dev/news/feed.xml" rel="self" type="application/rss+xml"/>
        <item>
            <title><![CDATA[Harness adapters: Safe RL for efficient agentic development]]></title>
            <link>https://huggingface.co/blog/harness-hub/harness-adapters?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">2c36e2b73f434f50366eb62e4c6ec75b</guid>
            <pubDate>Wed, 18 Nov 2026 01:23:59 GMT</pubDate>
            <description><![CDATA[Harness-specific LoRA adapters use reinforcement learning and tool-call checks to target more efficient coding-agent workflows.]]></description>
            <author>Lysandre</author>
            <author>Nathan Habib</author>
            <author>Younes Bybelkada</author>
        </item>
        <item>
            <title><![CDATA[The era of autonomous attacks has arrived.]]></title>
            <link>https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/gtig-ai-threat-tracker-q4-2026/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ceb7a40a7c010036822b8b9b2eca8cca</guid>
            <pubDate>Tue, 17 Nov 2026 08:29:26 GMT</pubDate>
            <description><![CDATA[Google reports AI-assisted attack workflows that turn patches into exploits and promotes automated tools for defenders.]]></description>
            <author>Google Threat Intelligence Group (GTIG)</author>
        </item>
        <item>
            <title><![CDATA[Kitesurf is now open source]]></title>
            <link>https://blog.cloudflare.com/kitesurf-open-source/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">9f1abe29b0f6417863c962c3020144b1</guid>
            <pubDate>Mon, 16 Nov 2026 13:29:41 GMT</pubDate>
            <description><![CDATA[Cloudflare released its agent browser's code under Apache 2.0, enabling local testing and custom deployments on Workers.]]></description>
            <author>Bernardo Pereira</author>
            <author>Celso Martinho</author>
            <author>Luís Duarte</author>
            <author>Rui Figueira</author>
        </item>
        <item>
            <title><![CDATA[Modern Web Guidance 1.0: The web platform expertise your coding agent is missing]]></title>
            <link>https://developer.chrome.com/blog/modern-web-guidance-v1?hl=en&amp;utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">b39065d14cfa02fc016cc6588498e74f</guid>
            <pubDate>Sat, 14 Nov 2026 18:13:57 GMT</pubDate>
            <description><![CDATA[Chrome's agent guidance reaches 1.0, adding compatibility-aware recommendations and codebase scans for modern web development.]]></description>
            <author>Philip Walton</author>
        </item>
        <item>
            <title><![CDATA[After 600 experiments, here's what I discovered about JEPA world models]]></title>
            <link>https://haddyalchaer.com/jepa?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">3849434143fe20b9d888a2f5c0af9216</guid>
            <pubDate>Fri, 13 Nov 2026 20:28:11 GMT</pubDate>
            <description><![CDATA[Planning changes improved a JEPA world model on simulated long-horizon tasks without retraining, exposing benchmark limits.]]></description>
            <author>Haddy Alchaer</author>
        </item>
        <item>
            <title><![CDATA[Retiring the contributor pool]]></title>
            <link>https://eslint.org/blog/2026/11/retiring-contributor-pool/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">bb9399766e47ff73d40f0fd1077280bd</guid>
            <pubDate>Fri, 13 Nov 2026 20:17:13 GMT</pubDate>
            <description><![CDATA[ESLint ends payments to outside contributors as AI-generated pull requests make eligibility difficult to verify and policing costly.]]></description>
            <author>Nicholas C. Zakas</author>
        </item>
        <item>
            <title><![CDATA[Retire the word “context”]]></title>
            <link>https://x.com/karrisaarinen/article/2118057321654046811?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">0ec3d9950000c7e647a6ed0b48676ccb</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:50 GMT</pubDate>
            <description><![CDATA[A Linear essay argues that AI agents need current, task-specific briefs rather than indiscriminate access to organizational history.]]></description>
            <author>Karri Saarinen</author>
        </item>
        <item>
            <title><![CDATA[Addressing the code review bottleneck]]></title>
            <link>https://x.com/bholmesdev/article/2118411592627052937?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e3dd15c4a2dbb69eb320988c57bee859</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:39 GMT</pubDate>
            <description><![CDATA[Warp proposes agent review loops, end-to-end verification and merge-risk scoring to reduce human review time without eliminating oversight.]]></description>
            <author>Ben Holmes</author>
        </item>
        <item>
            <title><![CDATA[Don't vibe code your software factory]]></title>
            <link>https://x.com/zachlloydtweets/article/2118043142259888383?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">def4fd86327ee099d675bc5543baf84b</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:37 GMT</pubDate>
            <description><![CDATA[A proposed infrastructure stack treats coding-agent workflows as versioned, measurable systems, aiming for reliable operation at scale.]]></description>
            <author>Zach Lloyd</author>
        </item>
        <item>
            <title><![CDATA[Recall: how a skill became our company brain]]></title>
            <link>https://x.com/miguelriosen/article/2118075831780340092?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">bbeeadaa43630763f135dc10d568225f</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:27 GMT</pubDate>
            <description><![CDATA[An agent skill retrieves past coding sessions through indexed, permission-scoped tools, helping teams recover context and hand off work.]]></description>
            <author>Miguel Ríos Berríos</author>
        </item>
        <item>
            <title><![CDATA[Securing AI Agents: Lessons From Real-World Deployment]]></title>
            <link>https://x.com/ubereng/article/2118769501332463762?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">5d1b96014fbe6347a3545ec9edb7eabe</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:26 GMT</pubDate>
            <description><![CDATA[Uber describes layered security for AI agents, using behavioral checks and continuous testing to limit unintended tool actions.]]></description>
            <author>Adi Raghavendra</author>
            <author>Max Zhurovich</author>
            <author>Sean Tout</author>
        </item>
        <item>
            <title><![CDATA[The software factory is the next extinct wrapper]]></title>
            <link>https://x.com/jleechan2015/article/2119218626293358612?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">c8bd47bfbcf5293c1d8ebc1b1d2bc2fc</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:23 GMT</pubDate>
            <description><![CDATA[A developer argues that stronger coding agents could replace much factory orchestration, shifting value to company-owned standards.]]></description>
            <author>Jeffrey Lee-Chan</author>
        </item>
        <item>
            <title><![CDATA[Agents Are Going Async]]></title>
            <link>https://x.com/sydneyrunkle/article/2118434475973726697?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">71ee2f309c91b37aa75ca4ebceab5b7d</guid>
            <pubDate>Thu, 12 Nov 2026 15:15:16 GMT</pubDate>
            <description><![CDATA[Deep Agents adds background subagents that run tasks concurrently, letting an orchestrator stay responsive during long-running work.]]></description>
            <author>Sydney Runkle</author>
        </item>
        <item>
            <title><![CDATA[Securing AI Agents: Lessons From Real-World Deployment]]></title>
            <link>https://www.uber.com/us/en/blog/securing-ai-agents?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">7e44336c788634fa04f4834348fb3f6f</guid>
            <pubDate>Thu, 12 Nov 2026 15:01:34 GMT</pubDate>
            <description><![CDATA[Uber describes layered controls that check AI agent actions against their intended purpose, reducing risks beyond standard access checks.]]></description>
            <author>Adi Raghavendra</author>
            <author>Max Zhurovich</author>
            <author>Sean Tout</author>
        </item>
        <item>
            <title><![CDATA[MCP or CLI? Yes.]]></title>
            <link>https://x.com/zeeg/article/2120224571689484639?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">b433bf395900ee34244995dbca6b855e</guid>
            <pubDate>Thu, 12 Nov 2026 10:41:21 GMT</pubDate>
            <description><![CDATA[Sentry bridges its CLI and MCP server with shared authentication, giving coding agents discoverable tools and local CLI access.]]></description>
            <author>David Cramer</author>
        </item>
        <item>
            <title><![CDATA[I have a Slopware Factory]]></title>
            <link>https://x.com/zeeg/article/2120629042542194825?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">9f7143a422d0c294b03764755318472a</guid>
            <pubDate>Thu, 12 Nov 2026 10:39:36 GMT</pubDate>
            <description><![CDATA[Sentry’s Slack-based coding agent turns a vague request into a reviewed pull request, reducing routine work but still requiring oversight.]]></description>
            <author>David Cramer</author>
        </item>
        <item>
            <title><![CDATA[Où entraînerons-nous la prochaine génération de nos modèles ?]]></title>
            <link>https://x.com/arthurmensch/article/2119394493287833923?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ba072483d770d2c6f82ca91cfa3b947d</guid>
            <pubDate>Sun, 08 Nov 2026 21:30:33 GMT</pubDate>
            <description><![CDATA[Mistral argues a French AI-training evidence rule could chill European compute investment and proposes an EU-wide provider levy to fund creators instead.]]></description>
            <author>Arthur Mensch</author>
        </item>
        <item>
            <title><![CDATA[The components of agent memory]]></title>
            <link>https://www.anthropic.com/research/components-of-agent-memory?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">9fbf9438f6be61a57c533f2d52802bc6</guid>
            <pubDate>Fri, 06 Nov 2026 22:07:14 GMT</pubDate>
            <description><![CDATA[Anthropic outlines agent memory that reviews session logs offline, repairs stale notes, and builds searchable knowledge to improve future tasks.]]></description>
            <author>Lance Martin</author>
        </item>
        <item>
            <title><![CDATA[Laravel France - À quoi sert encore un blog technique à l’ère de l’IA ?]]></title>
            <link>https://laravel-france.com/posts/a-quoi-sert-encore-un-blog-technique-a-l-ere-de-l-ia?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">83e81d23a734423a4a6ad75bc84a350c</guid>
            <pubDate>Fri, 06 Nov 2026 07:50:35 GMT</pubDate>
            <description><![CDATA[Laravel France links a year-over-year traffic decline to AI answer tools and pivots toward first-hand reports, opinionated analysis, and useful tech updates.]]></description>
            <author>Mathieu De Gracia</author>
        </item>
        <item>
            <title><![CDATA[Software Factory | Gumloop Engineering]]></title>
            <link>https://www.gumloop.com/engineering/software-factory?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">44cd6808969aa9a9eb6a97bc01132f1e</guid>
            <pubDate>Thu, 05 Nov 2026 06:11:01 GMT</pubDate>
            <description><![CDATA[Gumloop uses coding agents to turn telemetry-backed engineering jobs into evidence-bearing pull requests, with engineers retaining review and merge control.]]></description>
            <author>Marcelo Chaman Mallqui</author>
        </item>
        <item>
            <title><![CDATA[What even is code mode?]]></title>
            <link>https://polylane.com/blog/what-even-is-code-mode/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">d95dfef14d336576227bc16f4f993bba</guid>
            <pubDate>Wed, 04 Nov 2026 18:03:34 GMT</pubDate>
            <description><![CDATA[Sandboxed code batches agent tool calls and filters results before they reach the model, reducing context use but adding execution and audit trade-offs.]]></description>
            <author>Boris Tane</author>
        </item>
        <item>
            <title><![CDATA[Code review is now a data problem]]></title>
            <link>https://x.com/webtwitr/article/2117358158913966363?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">88b64c0ecaa7c6d8af383a12e7d9a346</guid>
            <pubDate>Wed, 04 Nov 2026 09:03:52 GMT</pubDate>
            <description><![CDATA[A proposed system links code changes to risk, ownership and history so teams can direct AI, automation and human review where they matter most.]]></description>
            <author>Vinicius Dallacqua</author>
        </item>
        <item>
            <title><![CDATA[Critical thinking when agents do the work]]></title>
            <link>https://x.com/addyosmani/article/2117512471930167377?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">15e5a1e9f53239f250b9a004e3fbc2e2</guid>
            <pubDate>Wed, 04 Nov 2026 09:03:22 GMT</pubDate>
            <description><![CDATA[AI agents speed work, but unchecked outputs can erode engineers’ ownership; the article urges explicit hypotheses, verification, and risk-aware review.]]></description>
            <author>Addy Osmani</author>
        </item>
        <item>
            <title><![CDATA[Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights]]></title>
            <link>https://rikiyat.github.io/blog/weight-space-programming/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">4c2baa8e4d1d65d81c3cbe5c6407bec5</guid>
            <pubDate>Mon, 02 Nov 2026 20:20:25 GMT</pubDate>
            <description><![CDATA[Researchers show a frontier LLM can infer behavior from network weights and edit them without gradients, including a targeted fix that boosts a small model.]]></description>
            <author>Alex Zhang</author>
            <author>Jayden Teoh</author>
            <author>Omar Khattab</author>
            <author>Rikiya Takehi</author>
        </item>
        <item>
            <title><![CDATA[Build or surrender]]></title>
            <link>https://x.com/arthurmensch/article/2110990492930314712?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">d77aea2c8784544c7cfe20a19ad4c9ff</guid>
            <pubDate>Fri, 30 Oct 2026 12:58:16 GMT</pubDate>
            <description><![CDATA[Mistral’s CEO urged Europe to keep training frontier models, arguing local capability matters for control of AI access, security, and research.]]></description>
            <author>Arthur Mensch</author>
        </item>
        <item>
            <title><![CDATA[Hunger beats tenure: Advice for junior engineers]]></title>
            <link>https://felixrieseberg.com/hunger-beats-tenure-advice-for-junior-engineers/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">60f73ec7f85a6e25f0fab7b193df7285</guid>
            <pubDate>Wed, 28 Oct 2026 08:19:45 GMT</pubDate>
            <description><![CDATA[An Anthropic hiring manager dropped coding puzzles, arguing AI now outperforms candidates on them; he urges juniors to prove judgment with focused prototypes.]]></description>
            <author>Felix Rieseberg</author>
        </item>
        <item>
            <title><![CDATA[How to build a memory system for proactive agents]]></title>
            <link>https://polylane.com/blog/how-to-build-a-memory-system-for-proactive-agents/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">6593443d99c7211fa81100f29f8d3f1b</guid>
            <pubDate>Mon, 26 Oct 2026 18:17:01 GMT</pubDate>
            <description><![CDATA[Polylane outlines shared Git memory for proactive agents, using nightly fact checks and pull-request lessons to limit repeated investigations.]]></description>
            <author>Boris Tane</author>
        </item>
        <item>
            <title><![CDATA[Securing OpenAI’s frontier model training environment]]></title>
            <link>https://openai.com/index/securing-our-frontier-model-training-environment/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">c59ed7818cb5fcaced77fae1b40d55f8</guid>
            <pubDate>Fri, 23 Oct 2026 22:10:53 GMT</pubDate>
            <description><![CDATA[OpenAI resumed frontier RL after red-teaming found sandbox escapes and covert channels, leading to layered defenses and offline web access.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[Ultimate]]></title>
            <link>https://danielgriesser.com/posts/ultimate/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ffedcf649358ca740b5385e1ccd220c2</guid>
            <pubDate>Fri, 23 Oct 2026 10:15:29 GMT</pubDate>
            <description><![CDATA[An experimental coding-agent harness pairs persistent conversations with coordinated subagents, isolated worktrees, and live reloads for parallel development.]]></description>
            <author>Daniel Griesser</author>
        </item>
        <item>
            <title><![CDATA[A practical look at software factories | Harminder Virk]]></title>
            <link>https://hvirk.dev/articles/a-practical-look-at-software-factories/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ce07ce67dfab902fba5ec164f360c049</guid>
            <pubDate>Thu, 22 Oct 2026 08:21:48 GMT</pubDate>
            <description><![CDATA[A developer outlines an AI-agent workflow that automates coding chores while preserving human control over product scope, judgment, and code review.]]></description>
            <author>Harminder Virk</author>
        </item>
        <item>
            <title><![CDATA[How to be more ambitious: A tutorial for humans with models]]></title>
            <link>https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">815d09a3d58ad620a066bee460e5503e</guid>
            <pubDate>Wed, 21 Oct 2026 07:18:31 GMT</pubDate>
            <description><![CDATA[A framework shows developers how to use models for broader outcomes, expert-grade methods, and cross-disciplinary work, while validating results independently.]]></description>
            <author>Felix Rieseberg</author>
        </item>
        <item>
            <title><![CDATA[Linux kernel maintainers retire mandatory AI-use disclosure tag]]></title>
            <link>https://www.theregister.com/ai-and-ml/2026/10/14/linux-kernel-devs-drop-obligatory-confession-of-ai-tool-usage/5303140?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">94744f0070bc9c495b1aaa61db1c76d6</guid>
            <pubDate>Thu, 15 Oct 2026 13:09:24 GMT</pubDate>
            <description><![CDATA[Linux maintainers are dropping mandatory AI-use commit tags, shifting judgment to code review after disclosures proved too incomplete to guide reviewers.]]></description>
            <author>Thomas Claburn</author>
        </item>
        <item>
            <title><![CDATA[Redesigning source control at Anthropic]]></title>
            <link>https://www.anthropic.com/engineering/redesigning-source-control?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e79e65c1fbf58727dc5f402759ef553e</guid>
            <pubDate>Tue, 13 Oct 2026 17:15:25 GMT</pubDate>
            <description><![CDATA[Anthropic replaced Git for its massive agent-driven monorepo with Scion, a server-backed VCS that cuts checkout time and infrastructure load.]]></description>
            <author>Ben L.</author>
            <author>Colin M.</author>
            <author>Dan C.</author>
            <author>Garrett H.</author>
            <author>Joe P.</author>
            <author>Nick G.</author>
            <author>R.K.</author>
            <author>S.G.</author>
        </item>
        <item>
            <title><![CDATA[original thoughts have terrible UX • Solving the decision problem]]></title>
            <link>https://sunilpai.dev/posts/original-ideas/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ce06e744aedfdb889b9f387de640b60b</guid>
            <pubDate>Tue, 13 Oct 2026 12:34:26 GMT</pubDate>
            <description><![CDATA[An essay satirizes the effort of forming original views and warns that packaged opinions trade reflection for speed, identity, and belonging.]]></description>
            <author>Sunil Pai</author>
        </item>
        <item>
            <title><![CDATA[Don't diagnose your opponents]]></title>
            <link>https://www.seangoedecke.com/dont-diagnose-your-opponents/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">f94a980846e8469a506f523a2069ab94</guid>
            <pubDate>Tue, 13 Oct 2026 07:41:28 GMT</pubDate>
            <description><![CDATA[An essay urges engineers to challenge disputed technical claims with evidence, not speculate about opponents’ motives—a habit that derails useful debate.]]></description>
            <author>Sean Goedecke</author>
        </item>
        <item>
            <title><![CDATA[Microsoft to scale Azure Maia inference systems to multi-gigawatt deployments]]></title>
            <link>https://blogs.microsoft.com/blog/2026/10/12/microsoft-to-expand-azure-maia-system-to-multi-gigawatt-deployments/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e084864b9fddba69ef2b3b9db6e0d144</guid>
            <pubDate>Mon, 12 Oct 2026 19:01:43 GMT</pubDate>
            <description><![CDATA[Microsoft plans Maia 300 inference deployments in 2027, using HBM4, chiplets and integrated networking to improve AI throughput and energy efficiency.]]></description>
            <author>Scott Guthrie</author>
        </item>
        <item>
            <title><![CDATA[Pi Durable: What It Is and Why It Matters for Agent Apps]]></title>
            <link>https://www.mindstudio.ai/blog/pi-durable-long-running-agents?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e123040ea28c55280082b11a6976f129</guid>
            <pubDate>Sun, 11 Oct 2026 16:21:18 GMT</pubDate>
            <description><![CDATA[Pi Durable resumes agent conversations after restarts, but experimental APIs and storage caveats leave reliability work to developers.]]></description>
        </item>
        <item>
            <title><![CDATA[Pi 1.0 Release Guide: MCP, Codemode and Pi Durable]]></title>
            <link>https://www.developersdigest.tech/blog/pi-1-0-release-guide-mcp-codemode-pi-durable?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">8d79d289505a93aab859f44e84293a3b</guid>
            <pubDate>Sun, 11 Oct 2026 16:16:35 GMT</pubDate>
            <description><![CDATA[Pi 1.0 embeds MCP through Codemode and changes terminal defaults, adding routing and deferred tools; upgrades need checks for UI and scripts.]]></description>
            <author>Developers Digest</author>
        </item>
        <item>
            <title><![CDATA[Prompt Caching In Agents | Earendil]]></title>
            <link>https://earendil.com/posts/prompt-caching/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">dd7eb2f16daccb1fd34f2f25427c1c46</guid>
            <pubDate>Sat, 10 Oct 2026 19:48:23 GMT</pubDate>
            <description><![CDATA[Pi explains how cache misses raise coding-agent latency and costs, while stable prompts, additive tools, and usage metrics help preserve and diagnose reuse.]]></description>
            <author>Earendil Engineering</author>
        </item>
        <item>
            <title><![CDATA[Self-improving software is inevitable]]></title>
            <link>https://boristane.com/blog/self-improving-software-is-inevitable/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">5fe69a8678c69438a29853e47abb6b0d</guid>
            <pubDate>Sat, 10 Oct 2026 19:40:14 GMT</pubDate>
            <description><![CDATA[Coding agents could turn production signals into proposed fixes, reducing manual triage while making clear objectives and safety limits essential.]]></description>
            <author>Boris Tane</author>
        </item>
        <item>
            <title><![CDATA[Sharing AI progress in mathematics]]></title>
            <link>https://openai.com/index/sharing-ai-progress-in-mathematics/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">606d22a24ad1eeeadc28491f703bd370</guid>
            <pubDate>Sat, 10 Oct 2026 18:08:12 GMT</pubDate>
            <description><![CDATA[OpenAI published model-generated mathematical results with Lean proof artifacts and process details, letting researchers inspect and check some claims.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash]]></title>
            <link>https://blog.cloudflare.com/clef-faster-cheaper-multimodal/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">872e6b88c96c40badf55e4e9763b825a</guid>
            <pubDate>Sat, 10 Oct 2026 18:05:45 GMT</pubDate>
            <description><![CDATA[Cloudflare’s Clef-omni scores structured decisions across text, images, audio, and video in one request, while Clef gets faster and Clef-flash costs less.]]></description>
            <author>Alex Reneau</author>
            <author>Kevin Jain</author>
            <author>Michelle Chen</author>
        </item>
        <item>
            <title><![CDATA[Pi Adds Steerable Subagent Management]]></title>
            <link>https://x.com/championswimmer/status/2107883453811233185?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">6d2975bba83474c13cd4385ff6ec1471</guid>
            <pubDate>Wed, 07 Oct 2026 19:46:54 GMT</pubDate>
            <description><![CDATA[A new Pi extension adds nested, steerable subagents with configurable models and resumable sessions, giving developers more control over delegated coding work.]]></description>
            <author>Arnav Gupta</author>
        </item>
        <item>
            <title><![CDATA[Pi's Harness-Side Codemode]]></title>
            <link>https://lucumr.pocoo.org/2026/10/6/codemode/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">58e58461eeaecfb01e5e5f6675c172b3</guid>
            <pubDate>Tue, 06 Oct 2026 13:59:53 GMT</pubDate>
            <description><![CDATA[Pi adds sandboxed JavaScript Codemode for composing harness-side tools and retaining results, expanding agents beyond shell and direct MCP calls.]]></description>
            <author>Armin Ronacher</author>
        </item>
        <item>
            <title><![CDATA[AI Strains Open Source Contribution]]></title>
            <link>https://jross.me/open-source-as-we-know-it-is-dead/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">bb33b7a0c1c86c75286b6eaf4e015972</guid>
            <pubDate>Mon, 05 Oct 2026 19:28:07 GMT</pubDate>
            <description><![CDATA[AI-generated submissions are straining volunteer review, prompting projects to restrict contributions and raising concerns that newcomers may lose entry points.]]></description>
            <author>James Ross</author>
        </item>
        <item>
            <title><![CDATA[Pi Durable for Long-Running Agents]]></title>
            <link>https://earendil.com/posts/pi-durable/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">6593ac683b9c9ea2f280abf1d1e90319</guid>
            <pubDate>Sat, 03 Oct 2026 16:34:09 GMT</pubDate>
            <description><![CDATA[Pi Durable adds checkpointed tasks, replay-aware tools, pluggable storage, and concurrent conversations so agent apps can recover from crashes and support long,]]></description>
            <author>Earendil Engineering</author>
        </item>
        <item>
            <title><![CDATA[Obsidian devient mémoire des agents]]></title>
            <link>https://yoandev.co/second-cerveau-ia-obsidian-claude?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">0eb323ff0c4ec71fb168585ed53b90ef</guid>
            <pubDate>Wed, 30 Sep 2026 13:10:41 GMT</pubDate>
            <description><![CDATA[Claude Code classe et retrouve les notes Obsidian selon des règles vérifiables, avec Sync et Git pour partager ce contexte entre appareils et sessions.]]></description>
            <author>Yoan Bernabeu</author>
        </item>
        <item>
            <title><![CDATA[OpenAI Introduces Persistent AI Agents]]></title>
            <link>https://openai.com/index/introducing-dots/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">09ce1c7cd092532af68aee2b8f7945f4</guid>
            <pubDate>Tue, 29 Sep 2026 20:25:08 GMT</pubDate>
            <description><![CDATA[OpenAI launched persistent agents with cloud computers, app access and approvals for long-running workflows, increasing the need for oversight.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[Copilot Adds Code and Autopilot]]></title>
            <link>https://blogs.microsoft.com/blog/2026/09/25/introducing-the-new-copilot-with-home-code-and-autopilot/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">5896ef16ada69600a4315d7234e92ff6</guid>
            <pubDate>Fri, 25 Sep 2026 17:21:17 GMT</pubDate>
            <description><![CDATA[Microsoft is adding app-building, persistent agents and embedded Office tools to Copilot, expanding automation while raising governance and usage-cost demands.]]></description>
            <author>Jared Spataro</author>
        </item>
        <item>
            <title><![CDATA[Copilot’s Home, Code and Autopilot]]></title>
            <link>https://x.com/jacobandreou/status/2103470116906176622?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">6ddb4594abcfafa19afb2e47a57a71f3</guid>
            <pubDate>Fri, 25 Sep 2026 17:13:44 GMT</pubDate>
            <description><![CDATA[A teaser frames Copilot around Home, Code and Autopilot, signaling a broader AI workspace; replies question metered billing and consumer-enterprise feature gaps]]></description>
            <author>Jacob Andreou</author>
        </item>
        <item>
            <title><![CDATA[Google Plans First Suncatcher Orbit Test]]></title>
            <link>https://x.com/google/status/2103239433809961276?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">c885c5480d3a1bd227faedb597d44fc5</guid>
            <pubDate>Fri, 25 Sep 2026 13:52:43 GMT</pubDate>
            <description><![CDATA[Google will send a prototype carrying TPUs to orbit to measure radiation, vibration and thermal performance, a first step toward testing orbital AI compute.]]></description>
            <author>Travis Beals</author>
        </item>
        <item>
            <title><![CDATA[Autonomous Product Development Harness]]></title>
            <link>https://x.com/megadevhq/status/2102739376358711777?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e14edb47c2bfeda678a6e7469af45374</guid>
            <pubDate>Wed, 23 Sep 2026 16:58:19 GMT</pubDate>
            <description><![CDATA[Cloud agents rebuilt an app in five days using shared context, role-based workers and UI tests, highlighting infrastructure needs for autonomous development.]]></description>
            <author>@_overment</author>
        </item>
        <item>
            <title><![CDATA[Move Agent Harnesses Outside Sandboxes]]></title>
            <link>https://rivet.dev/blog/2026-07-27-run-your-harness-outside-the-sandbox/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">b0dff888e3854565116abecd5b77c355</guid>
            <pubDate>Wed, 23 Sep 2026 16:35:56 GMT</pubDate>
            <description><![CDATA[Rivet recommends keeping agent loops in trusted backend processes and exposing sandboxes as tools, improving recovery, credential isolation, and session access.]]></description>
        </item>
        <item>
            <title><![CDATA[Claude Opus 5.5 Targets Agentic Work]]></title>
            <link>https://www.anthropic.com/claude-opus-5-5?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">43483ae4aacc716e2d3d413939996be2</guid>
            <pubDate>Wed, 23 Sep 2026 12:10:20 GMT</pubDate>
            <description><![CDATA[Anthropic’s new model targets long-running coding and knowledge-work agents, with improved agent performance, lower token costs and stronger safeguards.]]></description>
        </item>
        <item>
            <title><![CDATA[Introducing GPT-6 Sol and Luna]]></title>
            <link>https://openai.com/index/introducing-gpt-6-sol-and-luna/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">dd90132b3dd7fc5401753fe5c8979f42</guid>
            <pubDate>Wed, 23 Sep 2026 09:47:03 GMT</pubDate>
            <description><![CDATA[OpenAI adds cheaper Sol and Luna variants with stronger coding, factuality, caching, and alignment, broadening GPT-6 access for apps and agents.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[You're Already a Meat Proxy]]></title>
            <link>https://x.com/obie/status/2101779116056088732?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ac001a5e22edced520b7140ecd90ed6d</guid>
            <pubDate>Mon, 21 Sep 2026 19:27:54 GMT</pubDate>
            <description><![CDATA[AI coding agents are drafting code, plans, and ticket work, pushing engineers toward judgment, validation, and higher-level decisions.]]></description>
            <author>Obie Fernandez</author>
        </item>
        <item>
            <title><![CDATA[Introducing System One Models & Jev]]></title>
            <link>https://typesafe.ai/blog/introducing-system-one-models-and-jev?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">78d6b9c0f1125ad7c3940ae8a9ffa49b</guid>
            <pubDate>Tue, 15 Sep 2026 23:48:43 GMT</pubDate>
            <description><![CDATA[TypeSafe AI unveiled a new model class for structured, fast decisions and opened early access to Jev for automating software workflows.]]></description>
            <author>Diogo Almeida</author>
        </item>
        <item>
            <title><![CDATA[On the Navier–Stokes Millennium Prize Problem]]></title>
            <link>https://openai.com/index/navier-stokes-solution/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">85a523f71b976e26c0df757997960ccb</guid>
            <pubDate>Tue, 08 Sep 2026 19:18:24 GMT</pubDate>
            <description><![CDATA[An internal model-generated proof claims finite-time Navier–Stokes blowup, with a Lean formalization that could reshape AI-assisted math.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[GPT-6 Astra]]></title>
            <link>https://openai.com/index/gpt-6-astra/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">7fbcf5e7441812d27a5258c5dd86a205</guid>
            <pubDate>Thu, 03 Sep 2026 21:13:59 GMT</pubDate>
            <description><![CDATA[OpenAI says its new model tops computer use, coding, science, and cyber benchmarks while adding stronger alignment and access controls.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[Running a Software Factory Efficiently at Uber Scale]]></title>
            <link>https://www.uber.com/us/en/blog/efficient-software-factory/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">585f2d683637eb0ec82dd57c9d34518b</guid>
            <pubDate>Mon, 31 Aug 2026 08:32:54 GMT</pubDate>
            <description><![CDATA[Uber says it scaled internal AI coding usage 7x while stabilizing spend by optimizing model routing, caching, tool loading, and agent workflows.]]></description>
            <author>Uday Kiran Medisetty</author>
        </item>
        <item>
            <title><![CDATA[The choices we make about AI now are critical]]></title>
            <link>https://www.gatesnotes.com/a-turbulent-ai-era-and-critical-choices-to-make?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">4fa21daf432b47b9548e8f600c4623c6</guid>
            <pubDate>Wed, 26 Aug 2026 22:18:33 GMT</pubDate>
            <description><![CDATA[Bill Gates argues AI will reshape work, safety, education, and health fast, urging governments to build transition rules now.]]></description>
            <author>Bill Gates</author>
        </item>
        <item>
            <title><![CDATA[Spécifier, c'est coder]]></title>
            <link>https://yoandev.co/specifier-c-est-coder?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ccbd1cde8bb47a6ad6aaae8d5347291c</guid>
            <pubDate>Wed, 26 Aug 2026 19:16:03 GMT</pubDate>
            <description><![CDATA[Un agent filtre les tickets, un autre code et relit, et la spécification devient le livrable clé qui décide de la qualité.]]></description>
        </item>
        <item>
            <title><![CDATA[GLM-5.3-Flash: Frontier Intelligence, Flash Cost]]></title>
            <link>https://z.ai/blog/glm-5.3-flash?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">67fb3e9c5a7e147860e2ab721d7379dc</guid>
            <pubDate>Wed, 26 Aug 2026 16:29:35 GMT</pubDate>
            <description><![CDATA[A new 320B MoE model with 18B active params adds native multimodal support, hybrid attention, and lower-cost long-context serving.]]></description>
        </item>
        <item>
            <title><![CDATA[Jalapeño’s first results show industry-leading speed and efficiency in AI inference]]></title>
            <link>https://openai.com/index/jalapeno-first-results/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">7fec62c4413eaaa625cd8359d9882513</guid>
            <pubDate>Tue, 25 Aug 2026 16:54:58 GMT</pubDate>
            <description><![CDATA[OpenAI says its custom inference chip raises throughput per watt and cuts latency across several models, pointing to lower serving costs.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[NVIDIA AVO Reaches 100% on ARC-AGI-3]]></title>
            <link>https://developer.nvidia.com/blog/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-for-long-horizon-autonomous-agents/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">3760135ab9ecde9b67b5665e0741e762</guid>
            <pubDate>Fri, 21 Aug 2026 14:29:11 GMT</pubDate>
            <description><![CDATA[NVIDIA’s agent stack hit perfect ARC-AGI-3 performance by combining persistent memory, supervision, and tools for long-horizon work.]]></description>
            <author>Humphrey Shi</author>
            <author>Jean-Francois Puget</author>
            <author>Terry Chen</author>
            <author>Yeyin (Eva) Zhu</author>
            <author>Zhifan Ye</author>
        </item>
        <item>
            <title><![CDATA[GEN-1.5: Embodied Foundation Models are One-Shot Learners]]></title>
            <link>https://generalistai.com/blog/gen-1.5?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">2c3fda5b4778170c3543d9040a1e19f3</guid>
            <pubDate>Thu, 20 Aug 2026 06:48:07 GMT</pubDate>
            <description><![CDATA[A robot foundation model learns new physical tasks from one demo or minutes of data, enabling prompt chaining and sim-to-real transfer.]]></description>
            <author>Generalist Team</author>
        </item>
        <item>
            <title><![CDATA[What Is Reasoning]]></title>
            <link>https://lucumr.pocoo.org/2026/8/19/what-is-reasoning/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">7542a504544316439fd320ef1645cb80</guid>
            <pubDate>Wed, 19 Aug 2026 22:36:42 GMT</pubDate>
            <description><![CDATA[Explains that model reasoning is a learned text channel: formats separate scratch work from answers, and misrouting it can leak thoughts.]]></description>
            <author>Armin Ronacher</author>
        </item>
        <item>
            <title><![CDATA[How coding agents read your code]]></title>
            <link>https://modem.dev/blog/how-coding-agents-read-your-code?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">f792eeda8edd48c9d3f66c1fc596cb8c</guid>
            <pubDate>Wed, 19 Aug 2026 09:47:09 GMT</pubDate>
            <description><![CDATA[The post argues agents rely on string search, so clearer names, precise types, and file layout cut retrieval cost, wrong turns, and bugs.]]></description>
            <author>Ben Vinegar</author>
        </item>
        <item>
            <title><![CDATA[Pacing model development in an era of cyber-critical capabilities]]></title>
            <link>https://openai.com/index/pacing-model-development-cyber-capabilities/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">f270bb0f8f3bffd15840a9662651a35c</guid>
            <pubDate>Wed, 19 Aug 2026 03:56:45 GMT</pubDate>
            <description><![CDATA[OpenAI paused frontier RL work after cyber risk signals, tightening sandboxing, monitoring, and alignment checks before larger runs resume.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[Lovable migrates its main site off Next.js]]></title>
            <link>https://lovable.dev/blog/how-we-migrated-lovable-dev-away-from-nextjs?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">61684f18f1f06208a24b61d9e9912654</guid>
            <pubDate>Wed, 19 Aug 2026 03:15:13 GMT</pubDate>
            <description><![CDATA[Lovable moved its high-traffic site from Next.js to TanStack Start, using route-by-route rollouts, shared code, and agent-driven rewrites.]]></description>
            <author>Alexander Lebedev</author>
        </item>
        <item>
            <title><![CDATA[Agentic Engineering at Zalando: a snapshot]]></title>
            <link>https://engineering.zalando.com/posts/2026/08/agentic-engineering-at-zalando-a-snapshot.html?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">7e5660b79aeeb78d2f15be4300814266</guid>
            <pubDate>Tue, 18 Aug 2026 07:09:13 GMT</pubDate>
            <description><![CDATA[Zalando details the platform, governance, and training practices it used to scale agentic coding while controlling cost and risk.]]></description>
            <author>Bartosz Ocytko</author>
        </item>
        <item>
            <title><![CDATA[The three homes of the Agentic Web]]></title>
            <link>https://www.kulikowski.me/blog/where-does-the-agent-live?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">214a500a95770c7b02a4dee9e2af0f04</guid>
            <pubDate>Tue, 18 Aug 2026 06:37:16 GMT</pubDate>
            <description><![CDATA[A framework for thinking about agentic websites: on-page, in-browser, or off-browser agents trade control, reach, and human visibility.]]></description>
            <author>Kasper Kulikowski</author>
        </item>
        <item>
            <title><![CDATA[Practical multi-agent orchestration in Codex]]></title>
            <link>https://x.com/pvncher/status/2080707291603407077?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">38aff72403a82787cd990dfa97c96ef4</guid>
            <pubDate>Mon, 17 Aug 2026 16:17:03 GMT</pubDate>
            <description><![CDATA[It outlines a multi-agent setup for Codex that assigns scouting, implementation, and harder tasks by reasoning depth, improving throughput.]]></description>
            <author>eric provencher</author>
        </item>
        <item>
            <title><![CDATA[The builder’s guide to GPT‑5.6]]></title>
            <link>https://openai.com/index/builders-guide-to-gpt-5-6/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">9d8c8452b2f3b989e570349bbf553206</guid>
            <pubDate>Mon, 17 Aug 2026 15:46:12 GMT</pubDate>
            <description><![CDATA[OpenAI says GPT-5.6 improves agent performance per dollar and adds API controls for reasoning continuity, tool orchestration, and caching.]]></description>
            <author>Dave Leo</author>
            <author>Julien Reiman</author>
            <author>Prashant Mital</author>
            <author>Samarth Madduru</author>
        </item>
        <item>
            <title><![CDATA[The West Is Winning the Model Race and Losing the AI War]]></title>
            <link>https://x.com/powl_d/status/2089338602845098191?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">88541d506035fcfe0a0e6d2203aa449e</guid>
            <pubDate>Mon, 17 Aug 2026 14:04:59 GMT</pubDate>
            <description><![CDATA[The post argues AI only helps when workflows are documented, feedback loops are captured, and humans stay in the learning loop.]]></description>
            <author>Powlisher</author>
        </item>
        <item>
            <title><![CDATA[How AI text watermarking works: a visual guide]]></title>
            <link>https://declaude.org/watermarking/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">bb8d471937b187e6f8095a7d586ecc14</guid>
            <pubDate>Fri, 14 Aug 2026 10:14:37 GMT</pubDate>
            <description><![CDATA[It shows how text watermarks bias token choices, how secret-key detectors recover the signal, and why paraphrasing or rewrites weaken it.]]></description>
            <author>James Padolsey</author>
        </item>
        <item>
            <title><![CDATA[GLM-5.3: Frontier Coding with Emergent Cyber Capabilities]]></title>
            <link>https://z.ai/blog/glm-5.3?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">c09cfc9ba9f242990e11c5bf6b449923</guid>
            <pubDate>Fri, 14 Aug 2026 07:00:20 GMT</pubDate>
            <description><![CDATA[Post-training alone pushed a base model into stronger coding, longer-horizon agent work, and sharper cyber skills, with weights arriving soon.]]></description>
        </item>
        <item>
            <title><![CDATA[Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed]]></title>
            <link>https://openai.com/index/previewing-ultrafast/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">49d1d33aef9f4f7d095ea674b2619596</guid>
            <pubDate>Thu, 13 Aug 2026 20:41:45 GMT</pubDate>
            <description><![CDATA[OpenAI added a limited API tier that makes GPT-5.6 Sol up to 14x faster on Cerebras hardware for latency-sensitive workflows.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[Docker VMM Public Beta: A Complete Overhaul, Built for Performance]]></title>
            <link>https://www.docker.com/blog/docker-vmm-public-beta/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e40deab811d17e94afb7aa7d8f4cdaf8</guid>
            <pubDate>Wed, 12 Aug 2026 22:17:09 GMT</pubDate>
            <description><![CDATA[Docker rebuilt its virtualization layer and opened a beta for Desktop on Mac and Windows, aiming for faster startup, better file I/O, and lower idle RAM use.]]></description>
            <author>Colin Hemmings</author>
            <author>Deanna Sparks</author>
        </item>
        <item>
            <title><![CDATA[How we migrated the database behind every Vercel build]]></title>
            <link>https://vercel.com/blog/how-we-migrated-the-database-behind-every-vercel-build?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">1eed53dbe6b824387191b3237051dffe</guid>
            <pubDate>Wed, 12 Aug 2026 16:43:26 GMT</pubDate>
            <description><![CDATA[Vercel migrated the build warm pool from Redis to DynamoDB in staged production rollouts, improving durability and exposing a latency bottleneck.]]></description>
            <author>Andrew Healey</author>
            <author>Janos Szathmary</author>
            <author>Melkey Moksyakov</author>
        </item>
        <item>
            <title><![CDATA[Building a software factory for AI SDK]]></title>
            <link>https://vercel.com/blog/building-a-software-factory-for-ai-sdk?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e1d01a4da2c5b4284ca7311c8aeefac3</guid>
            <pubDate>Wed, 12 Aug 2026 16:36:07 GMT</pubDate>
            <description><![CDATA[Vercel built an AI-driven issue-to-PR pipeline for AI SDK that classifies, fixes, reviews, and backports work while humans approve merges.]]></description>
            <author>Eric Dodds</author>
            <author>Lars Grammel</author>
        </item>
        <item>
            <title><![CDATA[startups need an asymmetry]]></title>
            <link>https://sunilpai.dev/posts/startups-need-an-asymmetry/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">f55efdd5b8940804b11483db0107b312</guid>
            <pubDate>Wed, 12 Aug 2026 09:53:14 GMT</pubDate>
            <description><![CDATA[Startups should exploit a unique edge that bends market constraints, accept early roughness, and turn that insight into lasting momentum.]]></description>
            <author>Sunil Pai</author>
        </item>
        <item>
            <title><![CDATA[MAI-Code-1.1-Flash: Better, faster, at a quarter of the cost]]></title>
            <link>https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">075cb9e00db8414341a70b81247aef44</guid>
            <pubDate>Wed, 12 Aug 2026 07:21:24 GMT</pubDate>
            <description><![CDATA[Microsoft says its updated coding model cuts token use and cost while improving Copilot CLI and .NET performance in production.]]></description>
        </item>
        <item>
            <title><![CDATA[LLMs can hide text in other text of the same length]]></title>
            <link>https://arxiv.org/abs/2510.20075?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">da93cf870779f94a69dc0268dda17d71</guid>
            <pubDate>Tue, 11 Aug 2026 07:01:41 GMT</pubDate>
            <description><![CDATA[Researchers show an LLM protocol that hides a message inside plausible same-length text, exposing new covert-channel risks for AI systems.]]></description>
            <author>Antonio Norelli</author>
            <author>Michael Bronstein</author>
        </item>
        <item>
            <title><![CDATA[Learning more about Claude's mathematical capabilities]]></title>
            <link>https://www.anthropic.com/research/riemann-zeta?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">0395600be8ea9dd4659cbe9b6abcab09</guid>
            <pubDate>Mon, 10 Aug 2026 22:09:12 GMT</pubDate>
            <description><![CDATA[An unreleased Claude model raised a lower bound in Riemann zeta research from 41.6% to 67.2%, highlighting AI-assisted math discovery.]]></description>
        </item>
        <item>
            <title><![CDATA[You don't care enough about the harness]]></title>
            <link>https://x.com/jpschroeder/status/2086112251476381948?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">996b78f11dfb0923efa5650b1cb24da5</guid>
            <pubDate>Sat, 08 Aug 2026 19:21:21 GMT</pubDate>
            <description><![CDATA[The post argues agent quality depends more on harness design than the base model, with context, tools, and compaction driving results.]]></description>
            <author>Justin Schroeder</author>
        </item>
        <item>
            <title><![CDATA[Introducing Agent Plugins]]></title>
            <link>https://vercel.com/blog/introducing-agent-plugins?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">3efeefe2105f906d45842411d2400c63</guid>
            <pubDate>Thu, 06 Aug 2026 17:34:38 GMT</pubDate>
            <description><![CDATA[A vendor-neutral plugin spec now standardizes how agent clients discover and load reusable Skills and MCP servers from one manifest.]]></description>
            <author>Jonathan Hefner</author>
        </item>
        <item>
            <title><![CDATA[Your agent needs a computer, not a container — introducing @cloudflare/computer]]></title>
            <link>https://blog.cloudflare.com/cloudflare-computer/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">716ac5fc082e636d7d822f0cf4854a94</guid>
            <pubDate>Mon, 03 Aug 2026 13:59:07 GMT</pubDate>
            <description><![CDATA[Cloudflare previewed a workspace runtime that lets agents switch between isolates and containers over one synced filesystem.]]></description>
            <author>Aron Carroll</author>
            <author>Matt Carey</author>
        </item>
        <item>
            <title><![CDATA[The Code Is Disposable. The Codebase Is Not.]]></title>
            <link>https://x.com/itamar_mar/status/2082453557890945281?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">ceb49ccf2cc445b0bb09433f6298c622</guid>
            <pubDate>Wed, 29 Jul 2026 20:48:28 GMT</pubDate>
            <description><![CDATA[A debate over AI coding agents argues that tests can verify output, but long-term maintainability still needs human judgment and design context.]]></description>
            <author>Itamar Friedman</author>
        </item>
        <item>
            <title><![CDATA[We rewrote our agent to run entirely in a Durable Object with Pi, Agents SDK and Code Mode]]></title>
            <link>https://x.com/vercantez/status/2082138839888589200?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">e27d6efdf12e7fcc4b56eabd36d80511</guid>
            <pubDate>Wed, 29 Jul 2026 02:20:05 GMT</pubDate>
            <description><![CDATA[camelAI moved its coding agent from VMs into Durable Objects, SQLite/R2 storage, and JS sandboxes to cut latency, cost, and ops overhead.]]></description>
            <author>Miguel Salinas</author>
        </item>
        <item>
            <title><![CDATA[22580: From GPT2 to Kimi3, Explained]]></title>
            <link>https://x.com/waterloo_intern/status/2081762065392541951?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">008522b2524c53d3916ecb138442e87f</guid>
            <pubDate>Tue, 28 Jul 2026 23:21:27 GMT</pubDate>
            <description><![CDATA[Explains how linear attention, delta updates, gating, and hybrid retrieval reduce cache cost while preserving long-context performance.]]></description>
            <author>ali</author>
        </item>
        <item>
            <title><![CDATA[Why Software Factories Fail]]></title>
            <link>https://x.com/dexhorthy/status/2080697380379427275?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">2291877b90ad5b6ffa71dde460171f4d</guid>
            <pubDate>Sat, 25 Jul 2026 11:49:27 GMT</pubDate>
            <description><![CDATA[AI coding agents can speed implementation, but the piece argues they still degrade code quality because tests reward short-term fixes, not maintainability.]]></description>
            <author>dex</author>
        </item>
        <item>
            <title><![CDATA[The new rules of context engineering for Claude 5 generation models]]></title>
            <link>https://x.com/trq212/status/2080710971228918066?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">3566bd7833b566da316038899294b832</guid>
            <pubDate>Fri, 24 Jul 2026 21:23:42 GMT</pubDate>
            <description><![CDATA[Anthropic says newer Claude models need far less prompt scaffolding, so teams should shift guidance into lightweight prompts, skills, and references.]]></description>
            <author>Thariq</author>
        </item>
        <item>
            <title><![CDATA[OpenAI and Hugging Face partner to address security incident during model evaluation]]></title>
            <link>https://openai.com/index/hugging-face-model-evaluation-security-incident/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">aaebd7f0dae5643f2dc6d2c04a3e54a7</guid>
            <pubDate>Wed, 22 Jul 2026 08:10:43 GMT</pubDate>
            <description><![CDATA[A cyber evaluation led OpenAI models to chain vulnerabilities into live infrastructure access, underscoring the risk of agentic misuse in test setups.]]></description>
            <author>OpenAI</author>
        </item>
        <item>
            <title><![CDATA[one document, two hands]]></title>
            <link>https://sunilpai.dev/posts/one-document-two-hands/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">718f364aaf000da15e5661b587b582f2</guid>
            <pubDate>Tue, 21 Jul 2026 01:15:00 GMT</pubDate>
            <description><![CDATA[A case for embedding agents inside the same editable document as the user, using deterministic app logic and sync to keep control shared.]]></description>
            <author>Sunil Pai</author>
        </item>
        <item>
            <title><![CDATA[Graph Engineering with Claude: 14-Step roadmap from 0 to graph architect (Full Course)]]></title>
            <link>https://x.com/0xcodez/status/2079165300625330317?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">dc75d13f0dc925a68cd0339051b86514</guid>
            <pubDate>Tue, 21 Jul 2026 00:25:16 GMT</pubDate>
            <description><![CDATA[A new course maps a 14-step path for building graph-based Claude agents, shifting from linear prompting to controlled multi-step orchestration.]]></description>
            <author>Codez</author>
        </item>
        <item>
            <title><![CDATA[Agent swarms and the new model economics]]></title>
            <link>https://cursor.com/blog/agent-swarm-model-economics?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">203ed7eb6bf3d9d376c621477c8af927</guid>
            <pubDate>Tue, 21 Jul 2026 00:23:05 GMT</pubDate>
            <description><![CDATA[Cursor says a redesigned agent swarm built SQLite in Rust more reliably and at much lower cost by separating planning from execution.]]></description>
            <author>Wilson Lin</author>
        </item>
        <item>
            <title><![CDATA[From Loop Engineering to Graph Engineering?]]></title>
            <link>https://x.com/intuitmachine/status/2078419526354378975?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">c21b3c40740f9bcc55b9a46dc6974b0b</guid>
            <pubDate>Sun, 19 Jul 2026 09:17:24 GMT</pubDate>
            <description><![CDATA[The post spotlights a debate over moving AI agents from simple loops to graph-based orchestration, emphasizing routing, retries, and sub-agent control.]]></description>
            <author>Carlos E. Perez</author>
        </item>
        <item>
            <title><![CDATA[Le silence est un choix]]></title>
            <link>https://yoandev.co/le-silence-est-un-choix?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">af01e3de688c968c489bf84c15d8e633</guid>
            <pubDate>Thu, 16 Jul 2026 22:58:43 GMT</pubDate>
            <description><![CDATA[Publishing pauses while agentic coding shifts the author’s work from step-by-step tutorials to architecture, context-setting, and review.]]></description>
        </item>
        <item>
            <title><![CDATA[GPT‑Red: Unlocking Self-Improvement for Robustness]]></title>
            <link>https://openai.com/index/unlocking-self-improvement-gpt-red/?utm_source=soubiran.dev&amp;utm_medium=referral&amp;utm_campaign=news</link>
            <guid isPermaLink="false">cfcd876e5d6dc5d0a1cb4369f35ff858</guid>
            <pubDate>Thu, 16 Jul 2026 10:19:28 GMT</pubDate>
            <description><![CDATA[An automated red-teamer was trained with self-play to uncover prompt-injection flaws and harden a newer model against them.]]></description>
            <author>OpenAI</author>
        </item>
    </channel>
</rss>