<h2>Must know</h2>
<ul>
<li><strong>OpenAI launches GPT-5.4 mini and nano models</strong>: OpenAI has released two smaller, faster variants of GPT-5.4 designed for high-volume subagent tasks. The nano model is now the cheapest entry point for classification and data extraction, undercutting Google’s Gemini 3.1 Flash-Lite on input token pricing. Both models feature a 400k context window and are already rolling out to GitHub Copilot users to improve codebase exploration speed. <a href="https://app.daily.dev/posts/6TGfnHW2k" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox 149 adds built-in VPN and security APIs</strong>: Launching March 24, Firefox 149 will include a free built-in VPN offering 50GB of monthly data for browser traffic in select regions. This version also marks the first browser implementation of the Sanitizer API, a new web standard specifically designed to neutralize XSS attacks. Additional UI updates include a new ‘Split View’ for side-by-side page browsing and a redesigned settings navigation. <a href="https://app.daily.dev/posts/nPBbJz5BW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mistral releases Leanstral for formal code verification</strong>: Mistral has introduced Leanstral, an open-weight coding agent that uses the Lean programming language to provide formal proofs for AI-generated code. By grounding output in mathematical specifications, Mistral aims to solve the reliability issues inherent in ‘vibe coding.’ The model is available under an Apache 2.0 license and claims significantly lower inference costs compared to Claude 3.5 Sonnet for verification tasks. <a href="https://app.daily.dev/posts/XdxpqgLtJ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Vite 8 switches to Rolldown for 30x faster builds</strong>: The major release integrates the Rust-based Rolldown bundler to drastically reduce build times for frontend projects. <a href="https://app.daily.dev/feed-by-ids?id=CZHeTFzvD&amp;id=db4oA9qt6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Bun 1.3.11 introduces native OS cron job scheduling</strong>: The update adds a Bun.cron API for cross-platform task scheduling and fixes over 100 bugs in the Windows and macOS runtimes. <a href="https://app.daily.dev/posts/DuG54Q8tB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pyodide brings full Python data science stacks to the browser</strong>: The WebAssembly-based distribution now supports NumPy, pandas, and Scikit-learn running directly in client-side JavaScript. <a href="https://app.daily.dev/posts/GRvcdA8Dr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LeapKit offers a batteries-included Go web framework</strong>: This new framework bundles HTMX, Tailwind, and PostgreSQL support into a single CLI-driven workflow for rapid SaaS development. <a href="https://app.daily.dev/posts/02k5z7UCY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA CloudXR 6.0 brings RTX streaming to Vision Pro</strong>: Developers can now stream high-fidelity 3D applications from workstations to visionOS using a native Swift framework. <a href="https://app.daily.dev/posts/Lajg4CAxX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Project Detroit revives Java interoperability with Python and JS</strong>: Oracle is embedding V8 and CPython runtimes directly into the JVM to allow Java apps to call AI libraries natively. <a href="https://app.daily.dev/posts/aQEMjwL5W" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PostgreSQL 18 begins deprecating MD5 password encryption</strong>: The database will now issue warnings for MD5-encrypted users, pushing the ecosystem toward the more secure SCRAM-SHA-256 standard. <a href="https://app.daily.dev/posts/nekgWU5A0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The $12 million band-aid for AI slop</strong>: Big Tech is patting itself on the back for a $12.5 million fund to help open-source maintainers triage the mountain of garbage bug reports their own AI tools are generating. Instead of fixing the models that hallucinate security vulnerabilities, they’re funding ‘tools and strategies’ so humans can spend more time cleaning up after the machines. It’s a classic case of selling the poison and then charging for the antidote. <a href="https://app.daily.dev/posts/LS9KJQCC0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>World ID’s desperate bid for agentic relevance</strong>: The company formerly known as Worldcoin is launching ‘AgentKit’ to verify that your AI shopping agent is actually yours via iris scans. After their token tanked 76%, they’ve pivoted to ‘agentic commerce,’ hoping you’ll give an AI ‘power of attorney’ to buy things using a blockchain protocol. It’s a solution in search of a problem, wrapped in enough buzzwords to make a VC blush. <a href="https://app.daily.dev/posts/63OP7gZFo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The myth of the 5x productivity gain</strong>: Meta claims their ‘Ranking Engineer Agent’ doubled model accuracy and gave engineers a 5x productivity boost, while other studies show AI code actually increases the review burden and bug counts. These ‘productivity’ metrics conveniently ignore the long-term cost of maintaining unvetted, AI-generated SQL and the ‘AI blind spot debt’ accumulating in enterprise databases. Shipping code faster isn’t a win if you’re just accelerating the rate at which you break things. <a href="https://app.daily.dev/feed-by-ids?id=085ZOrcxF&amp;id=bE9k3kakw" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
