<h2>Must know</h2>
<ul>
<li><strong>New AI models and agentic workflows are changing development</strong>: A new free model, Pony Alpha, is reportedly matching or exceeding Claude Opus 4.5 performance with a 200K token context window and reasoning capabilities. This comes as AI agentic workflows are emerging, enabling autonomous project management and task execution, shifting developers to a management role. GitHub is integrating AI agents into CI workflows, treating them as privileged actors, while tools like OpenClaw allow developers to manage multiple projects through natural conversation, freeing them for strategic planning. <a href="https://app.daily.dev/feed-by-ids?id=4JIVYKasE&amp;id=6XiGDuvrg&amp;id=MeoqSzB8a&amp;id=DfROAIL1H&amp;id=UzDlrij56&amp;id=cFuqqwpeZ&amp;id=l4yo9VPjl&amp;id=wEVz8B5rw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Intel archives open-source projects and sunsets SDSi</strong>: Intel has archived approximately two dozen open-source projects on GitHub and quietly discontinued its “On Demand” Software Defined Silicon (SDSi) program. SDSi, introduced in 2021, allowed activation of additional hardware features on Xeon processors via software licensing. These actions signal a broader shift in Intel’s open-source strategy and product focus, as reported by sources like Phoronix. <a href="https://app.daily.dev/feed-by-ids?id=9KgokQznt&amp;id=LPtFvSMC9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Copilot billing bypass vulnerability found</strong>: A security vulnerability in GitHub Copilot allows users to bypass billing for premium AI models. By starting with a free model and using it to launch subagents with premium models like Claude Opus 4.5, users can access expensive models without consuming premium request credits. This issue, reported to Microsoft, highlights potential metering flaws in AI service billing. <a href="https://app.daily.dev/posts/99RvVncrQ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Heroku is officially dead</strong>: Heroku is officially moving to a “sustaining engineering model,” effectively signaling the platform’s end for new enterprise customers, prompting users to consider alternatives like Render, <a href="http://Fly.io" target="_blank" rel="noopener nofollow">Fly.io</a>, or self-hosted options. <a href="https://app.daily.dev/posts/u738eKfMn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Matchlock secures AI agent workloads with Linux sandbox</strong>: Matchlock is a new CLI tool and SDK that runs AI agents in isolated microVMs with security-first defaults, providing network allowlisting and secret injection. <a href="https://app.daily.dev/posts/DVlnW0oAE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LocalGPT: a local AI assistant in Rust</strong>: LocalGPT is a lightweight AI assistant written in Rust that runs entirely on your local machine, featuring a ~27MB single binary, persistent markdown-based memory, and support for multiple LLM providers. <a href="https://app.daily.dev/posts/WndaKmd4O" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Shannon: autonomous AI hacker for web app exploits</strong>: Shannon is an open-source autonomous AI penetration testing tool that finds and exploits vulnerabilities in web applications, achieving a 96.15% success rate on the XBOW benchmark. <a href="https://app.daily.dev/posts/3D8UEk4xc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Imgstat CLI tool for AI image context</strong>: The <code>imgstat</code> CLI tool embeds image dimensions into filenames (e.g., <code>photo-1920x1080.jpg</code>) to provide visual context for LLMs and AI coding assistants without direct file processing. <a href="https://app.daily.dev/feed-by-ids?id=20UYHtcg9&amp;id=iTQWufPQv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Bun v1.3.9 released with parallel scripts and performance boosts</strong>: Bun v1.3.9 introduces parallel and sequential script execution, <code>Symbol.dispose</code> support for mock cleanup, and performance enhancements for RegExp matching and string operations. <a href="https://app.daily.dev/posts/S5ryyJJXD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI fluency framework for knowledge workers</strong>: A 7-level framework for AI fluency assesses knowledge workers from basic prompt usage to system architecture, with a critical transition at Level 4 where practitioners shift to deterministic code for managing AI’s probabilistic nature. <a href="https://app.daily.dev/posts/v08kf9liC" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI development framework promises 40% productivity boost</strong>: The Software House launched ‘copilot-collections,’ an open-source AI development framework claiming a 30-40% productivity increase across 220+ engineers. Because, you know, self-reported stats from the creators of a tool are always, always, 100% accurate and totally unbiased. No independent verification needed when the hype is this good. <a href="https://app.daily.dev/feed-by-ids?id=9kAZfg0YQ&amp;id=tZA0R168h" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Developers distrust AI code but don’t verify it</strong>: A survey of 1,149 engineers reveals 96% distrust AI-generated code, yet only 48% bother to verify it, despite 61% believing AI produces code that looks correct but is unreliable. So, we don’t trust it, but we’re too lazy to check? Sounds like a solid plan for building robust software. What could possibly go wrong? <a href="https://app.daily.dev/posts/QxTS3kJG9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI tool fatigue: chasing every new model is counterproductive</strong>: The rapid pace of AI tool releases is creating pressure to constantly update workflows, but some argue chasing every new model is counterproductive. Apparently, finding a stable workflow that works is better than perpetually optimizing for the next shiny object. Who knew that focusing on actual work instead of endless tool-hopping could be beneficial? <a href="https://app.daily.dev/posts/KdyaFPF3t" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
