<h2>Must know</h2>
<ul>
<li><strong>AI agent security is a mess</strong>: AI coding agents, like Stripe’s ‘Minions’ generating 1,000+ PRs weekly, are prioritizing speed over security, creating a ‘security debt crisis’ with issues like hardcoded API keys. Researchers demonstrated ChatGPT recommending malicious sites, highlighting ‘ChatGPT SEO’ vulnerabilities. Mitigations include spec-driven prompting and CI/CD scanners, but the problem is growing, as detailed by Columbia University research and various security disclosures. <a href="https://app.daily.dev/feed-by-ids?id=DD6LZnw0Q&amp;id=Et2d8uHu3&amp;id=LUO8dPKde&amp;id=RH1TYGIET&amp;id=UGo4Ea1t8&amp;id=bDJrJKMJX&amp;id=em7afQ5sm&amp;id=l0KLAjYGc&amp;id=ohhVPexbM&amp;id=vXBWwcwBk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google’s gemini 3.1 pro: high benchmarks, low usability</strong>: Despite Google touting Gemini 3.1 Pro with top benchmark scores (78% on ARC AGI 2), real-world coding and UI generation tests show it consistently underperforms Claude Opus 4.6. Issues include tool-calling failures, incomplete features, and poor environment detection. One author suggests Google is ‘benchmaxing’—optimizing for benchmarks at the expense of practical usability, making it a ‘pure disappointment’ for developers, according to articles from February 21-22, 2026. <a href="https://app.daily.dev/feed-by-ids?id=bIAFUawfa&amp;id=mGa3eARS3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic and figma team up for code-to-design workflow</strong>: Anthropic and Figma have collaborated on a new workflow that allows Claude Code to convert production code directly into editable Figma designs. Developers can send a live web page to a Figma file with a simple prompt, facilitating design collaboration. Updated designs can then be round-tripped back to code, with Claude Code updating the codebase to match the new design, as announced on February 21, 2026. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Elixir’s new language server, expert, reaches release candidate</strong>: Expert, the new official Elixir language server merging Lexical and Next LS, has released its first release candidate (v0.1.0-rc.1), featuring compiler diagnostics, completions, go-to-definition, and monorepo support. <a href="https://app.daily.dev/posts/ItEtgwo7d" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>WebMCP: browser-native execution for AI agents</strong>: Google’s WebMCP is a browser-native API allowing websites to expose schema-defined JavaScript tools for AI agents to invoke directly within the browser, improving determinism and reducing token overhead. <a href="https://app.daily.dev/posts/61GmU0Tcy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NTransformer enables llama 3.1 70B on a single rtx 3090</strong>: NTransformer, an open-source C++/CUDA LLM inference engine, claims to run Llama 3.1 70B on a single RTX 3090 (24GB VRAM) using adaptive caching and layer streaming, achieving 0.5 tok/s for 70B Q4_K_M. <a href="https://app.daily.dev/feed-by-ids?id=Ahx4zgQks&amp;id=WoJMuyqAQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent orchestration with mission control</strong>: Mission Control is an open-source Next.js 14 and TypeScript dashboard for orchestrating AI agents, managing tasks, and coordinating multi-agent collaboration via OpenClaw Gateway. <a href="https://app.daily.dev/posts/vyr9dJRQq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Go’s FFI now faster than GDScript and Rust for Godot</strong>: Go bindings for Godot via GDExtension now surpass GDScript and ‘safe’ Rust in FFI performance, thanks to call batching and Go 1.26’s cgo overhead reduction, as benchmarked with the spritebench tool. <a href="https://app.daily.dev/posts/j8AoMYJTv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>KittenTTS: ultra-compact text-to-speech for edge devices</strong>: KittenTTS by Kitten ML offers a family of ultra-compact text-to-speech models (down to 15M parameters, under 25MB quantized) optimized for CPU, ONNX format, and Apache 2.0 license, enabling client-side browser or mobile deployment. <a href="https://app.daily.dev/posts/aXA1rlMhE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent development tools and observability</strong>: The AI agent ecosystem is growing with tools like <code>cocoindex-code</code> for semantic code search and Hugging Face Skills for reusable AI/ML tasks, while agent observability now focuses on debugging reasoning rather than code. <a href="https://app.daily.dev/feed-by-ids?id=LY7p8IODB&amp;id=ZmcjIu8ai&amp;id=frjHTf2Ap&amp;id=xhDI1zj4i" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI agents will automate all your devops, probably</strong>: AI LABS is hyping OpenClaw as the ultimate autonomous AI agent, claiming it can handle everything from React dependency maintenance to cloud cost monitoring and even end-to-end app building from a PRD. Apparently, it just needs a Mac Mini and a ‘dangerously-skip-permissions’ flag. Because what could possibly go wrong with an AI agent running wild on your local machine, building apps and managing your infrastructure? <a href="https://app.daily.dev/posts/WCj2dFu7Q" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The great sqlite migration that wasn’t</strong>: <a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> attempted a live migration from MariaDB to SQLite, briefly went read-only, and then immediately rolled back due to ‘severe performance degradation.’ It’s almost like swapping out a robust database for a file-based one without extensive testing might not be the best idea for a high-traffic site. But hey, at least they tried, right? Plans to retry after ‘addressing the performance issues’ are, of course, underway. <a href="https://app.daily.dev/posts/RLQeRSeS0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The magic of free jackets: a business logic flaw</strong>: Someone managed to get a $1400 jacket for free by stacking two promotional coupons on an e-commerce platform, resulting in a $0 cart total. The article frames this as a ‘business logic flaw’ and a ‘valuable target for beginners.’ So, instead of fixing their coupon system, companies are now providing ‘bug bounty opportunities’ for ethical hackers to get free stuff. What a time to be alive! <a href="https://app.daily.dev/posts/inabaQdpD" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
