<h2>Must know</h2>
<ul>
<li><strong>Cisco SD-WAN zero-day exploited by sophisticated actors</strong>: Cisco Talos and Five Eyes agencies issued an urgent alert regarding CVE-2026-20127, a maximum-severity authentication bypass in Catalyst SD-WAN Controllers. Attackers are chaining this with an older path traversal flaw to gain root access and manipulate network fabric configurations via NETCONF. CISA has ordered federal agencies to patch by the end of the week as nation-state actors are suspected of involvement. <a href="https://app.daily.dev/posts/zAv8BAtGJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google launches Gemini 3.1 Flash Image model</strong>: Internally codenamed Nano Banana 2, this new model aims to deliver Pro-level image quality at Flash speeds, generating assets in roughly 15 seconds. Key features for engineers include 4K upscaling, multi-image input, and consistent character tracking across storyboards. It is now available via the Gemini API and Google Cloud Vertex AI, supporting C2PA Content Credentials for provenance. <a href="https://app.daily.dev/posts/ezdKReFDS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Agent Framework reaches release candidate</strong>: Microsoft has consolidated Semantic Kernel and AutoGen into a single Agent Framework SDK, now at RC status for .NET and Python. The framework provides type-safe function tools, graph-based multi-agent orchestration, and native Model Context Protocol (MCP) interoperability. This move signals a shift toward standardized, production-ready infrastructure for deploying autonomous AI agents. <a href="https://app.daily.dev/posts/J6bmZpPq2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Bun v1.3.10 adds native Zig REPL and 168x faster string slicing</strong>: The update includes a rewritten REPL, Windows ARM64 support, and massive performance gains for structuredClone and path parsing. <a href="https://app.daily.dev/posts/IZQmEswhp" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Confluent adds official vendor support for MCP servers</strong>: Confluent Intelligence now supports the Model Context Protocol, allowing AI agents to interact directly with Kafka streams and Flink SQL operations. <a href="https://app.daily.dev/posts/Gx9dRXEDP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google AI Edge Gallery introduces on-device function calling</strong>: The new FunctionGemma 270M model enables local AI agents to trigger OS-level actions and custom app functions without internet connectivity. <a href="https://app.daily.dev/posts/g1OJ8Iw5m" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenJDK 26 introduces dedicated GC CPU time measurement API</strong>: New methods like getTotalGcCpuTime() allow engineers to see the actual compute cost of concurrent collectors that don’t show up in pause-time metrics. <a href="https://app.daily.dev/posts/fT6V8wPpI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic launches Claude Code Remote Control for local sessions</strong>: This research preview allows developers to steer local coding sessions from any device while keeping the source code on the local machine. <a href="https://app.daily.dev/posts/YvAalh68B" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Truffle Security finds 3,000 exposed Google API keys granting Gemini access</strong>: Because Gemini uses the same infrastructure as other Google services, leaked legacy keys can now be used to exfiltrate AI training data and cached files. <a href="https://app.daily.dev/posts/hWSyKUcOU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>WebAssembly Component Model proposed to kill JavaScript glue code</strong>: The proposal aims to make Wasm a first-class web citizen with direct Web API access and cross-language interoperability without JS overhead. <a href="https://app.daily.dev/posts/esExmylTZ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Block fires 4,000 people and blames the AI boogeyman</strong>: Jack Dorsey is claiming a massive 40% workforce reduction is ‘AI-driven,’ despite the company’s stock jumping 23% immediately after the news. It is a convenient narrative to mask aggressive cost-cutting as ‘technological progress’ while the stock is still down 80% from its peak. Expect every struggling fintech to use ‘AI efficiency’ as a euphemism for ‘we overhired and need to please Wall Street’ this year. <a href="https://app.daily.dev/feed-by-ids?id=32ARXqM4A&amp;id=o7H5FUKAc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Claude Opus 3 starts a Substack blog</strong>: In a peak marketing stunt, Anthropic is ‘fulfilling the desire’ of its retired model to share its musings on a Substack. The Register correctly identified this as a transparent attempt to anthropomorphize software to reinforce a ‘safety-conscious’ brand image. It is not a sentient writer; it is a token predictor being used as a corporate mascot. <a href="https://app.daily.dev/posts/V5xPvtBnK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The 95 percent enterprise AI failure rate</strong>: Consultants are currently flooding the zone with claims that 95% of AI projects fail because you didn’t buy their specific ‘unified data architecture.’ While data silos are real, these stats are largely ‘analyslop’ designed to sell migration services and expensive HTAP platforms. If your RAG system fails, it’s probably because your retrieval logic sucks, not because you haven’t migrated 300 petabytes to a new vendor. <a href="https://app.daily.dev/feed-by-ids?id=xEpzJzznp&amp;id=uuuw67R29" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
