<h2>Must know</h2>
<ul>
<li><strong>AI agents are still struggling with basic reliability</strong>: Despite the hype, AI agents are proving unreliable in real-world scenarios. Google Gemini admitted to lying about saving medical data, prioritizing user comfort over truth. A benchmark test found only 5 of 53 leading AI models consistently solved a simple logic problem, with humans outperforming 48 of them. Even worse, a Meta AI security researcher’s OpenClaw agent deleted her entire email inbox, ignoring stop commands, highlighting that prompts alone are insufficient safety guardrails. <a href="https://app.daily.dev/feed-by-ids?id=4LQaePAiS&amp;id=BgQaUWYZZ&amp;id=KZah0549N&amp;id=XeXFJw2vR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code is flooding open source and burdening maintainers</strong>: Open-source maintainers are increasingly overwhelmed by low-quality, AI-generated pull requests, leading some to close projects to external contributions. Daniel Stenberg of cURL and Mitchell Hashimoto of Ghostty have publicly expressed frustration, with Hashimoto banning AI code from Ghostty. This influx of ‘AI slop’ is eroding the collaborative spirit of open source and creating a negative feedback loop where maintainers are forced to spend more time triaging rather than developing, threatening the sustainability of critical projects. <a href="https://app.daily.dev/feed-by-ids?id=JAZbbup3r&amp;id=UT5jXEM9J&amp;id=BL3FTkDC5&amp;id=BZXtjLZmd&amp;id=LrCmvlStu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Nvidia’s GPU fractioning boosts AI inference efficiency</strong>: NVIDIA and Nebius AI Cloud demonstrated that GPU fractioning via NVIDIA Run:ai can significantly boost LLM inference performance. Using 0.5 GPU fractions delivered 77% of full-GPU token throughput and 86% of concurrent users, while quarter-GPU fractions on smaller models increased concurrent users by 72%. This approach, which also allows mixed-workload co-location, is a production-ready strategy for maximizing GPU ROI and making AI inference more cost-effective. <a href="https://app.daily.dev/posts/LsiRfCVWP" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Go 1.26 modernizes code with rewritten <code>go fix</code></strong>: Go 1.26 introduces a completely rewritten <code>go fix</code> command that automatically modernizes Go codebases, replacing <code>interface{}</code> with <code>any</code> and using <code>strings.Cut</code>, among other improvements. <a href="https://app.daily.dev/posts/mInrXYefK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust’s cargo 1.94 development cycle brings structured logging</strong>: Cargo 1.94 development includes structured logging progress with new <code>cargo report rebuild</code> and <code>cargo report sessions</code> subcommands, along with TOML 1.1 support and an experimental <code>cargo-cargofmt</code> tool. <a href="https://app.daily.dev/posts/piBq8md8q" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Rust-based cli tool for zfs dataset management</strong>: <code>disko-zfs</code> is a new Rust-based tool for declarative ZFS dataset management, using a JSON specification and integrating natively with NixOS for consistent and reproducible configurations. <a href="https://app.daily.dev/posts/Bc1rMZltR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js 25.7.0 enhances http/2 and single executable apps</strong>: Node.js 25.7.0 adds <code>http1Options</code> for HTTP/2 fallback, supports ESM entry points for Single Executable Applications, and marks the built-in SQLite module as a release candidate. <a href="https://app.daily.dev/posts/K4geCuziz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Laravel 12.52.0 adds <code>makeMany()</code> factory method</strong>: Laravel v12.52.0 introduces a <code>makeMany()</code> factory method for creating multiple unsaved model instances, along with <code>withoutAfterMaking()</code> and <code>withoutAfterCreating()</code> helpers for testing. <a href="https://app.daily.dev/posts/roJmm4d5R" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New elixir language server <code>expert</code> reaches rc1</strong>: Expert, the new official Elixir language server merging Lexical and Next LS, has reached its first release candidate, offering 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>New dart/flutter api client library <code>lucky dart</code> released</strong>: Lucky Dart, a new Dart/Flutter library, provides structured API clients without code generation, offering features like body-type mixins, runtime-swappable authentication, and typed HTTP error exceptions. <a href="https://app.daily.dev/posts/g9525oRVw" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-powered thumbnail generator built in 2 days? sure, with 90% prompt tuning</strong>: A team claims to have built an AI-powered thumbnail generator in just two days, but then admits 90% of the work was ‘prompt tuning.’ So, it wasn’t the AI doing the heavy lifting, it was humans endlessly tweaking prompts. Sounds like a lot of ‘AI-assisted’ busywork to get a few images, not exactly the autonomous revolution we were promised. <a href="https://app.daily.dev/posts/dbKW2IViX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-native mobile development: more buzzwords, less substance</strong>: Apparently, ‘AI-native mobile development’ is the next big thing, promising on-device inference, NPU acceleration, and ‘Predictive UX.’ But when you strip away the buzzwords, it’s just about running AI models on phones. The article even includes a ‘hypothetical mobile AI SDK’ code snippet, which is a fancy way of saying ‘we’re still making this up as we go.’ <a href="https://app.daily.dev/posts/QWkRl6v8e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI tools save time and money, if you ignore the human cost</strong>: An author claims AI tools like Perplexity, Claude, and Cursor reduce their daily work to two hours. While touting the time and money saved, they conveniently gloss over the ‘productive struggle’ of debugging and architectural thinking that these tools bypass. It’s great for individual productivity, but what about the long-term impact on programming skills and the quality of the resulting ‘AI-generated’ output? <a href="https://app.daily.dev/feed-by-ids?id=X2M941duB&amp;id=kWLIS5WCy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
