<h2>Must know</h2>
<ul>
<li><strong>AI code quality is a mess, but hybrid approaches offer hope</strong>: Startups are hiring developers to fix error-prone, insecure, and inefficient AI-generated code, reversing expected productivity gains. This ‘AI slop’ often comes from models trained on average public repositories, making it good for boilerplate but terrible for complex features, as reported by various industry sources. However, new tools like ProofOfThought are emerging, combining LLMs with theorem provers to bridge natural language intent with rigorous formal reasoning, making formal methods more accessible for robust AI applications.</li>
<li><strong>New AI programming protocols and agent frameworks emerge</strong>: The Model Context Protocol (MCP) is gaining traction as an AI-native protocol for LLMs, agents, and IDEs, offering standardized interfaces. Svelte now provides an MCP server for AI agents to analyze code and suggest fixes, while Phoenix creator Chris McCord advocates for Elixir’s superiority in agentic AI development due to its BEAM VM and ecosystem. These developments, highlighted by sources like TechCrunch, point to a growing focus on structured, robust frameworks for building production-ready AI systems.</li>
<li><strong>Netlify jumps on the AI agent bandwagon</strong>: Netlify has launched Agent Runners, AI agents designed to automate code fixes, updates, and deployments directly from its dashboard, supporting models like Claude, OpenAI, and Gemini. They also introduced an AI Gateway to consolidate inference management across multiple AI provider APIs. This move aims to streamline developer workflows and boost productivity, though details on cost and usage limitations remain vague.</li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Multiple security flaws found in AI systems and apps</strong>: Event planning app Partiful exposed user GPS data, Perplexity’s Comet AI browser was vulnerable to ‘CometJacking’ prompt injection, and OpenAI’s Sora had a jailbreak bug exposing sensitive audio recordings, all highlighting novel security challenges in AI-integrated systems.</li>
<li><strong>India’s tech talent pipeline is sputtering</strong>: Entry-level tech hiring in India has plummeted due to AI automation eliminating routine roles, forcing universities to overhaul curricula towards AI/ML skills and companies to prioritize experienced specialists.</li>
<li><strong>Linux 6.18 brings persistent caching and IOMMU updates</strong>: The new Linux kernel version includes DM-PCACHE for persistent block device caching, aiming to improve storage I/O performance, and IOMMU driver updates for Intel, AMD, Apple, and RISC-V, enhancing virtualization and memory management.</li>
<li><strong>Intel orphans numerous Linux software packages</strong>: Following internal restructuring, Intel has orphaned many critical Linux packages for Debian and Ubuntu, including tools for accelerators, performance monitoring, and power management, raising concerns about future support and compatibility.</li>
<li><strong>Functional programming’s impact on frontend development debated</strong>: An article argues that functional paradigms, particularly in frameworks like React, led to practices like CSS-in-JS and synthetic events that often fight native browser standards, though a shift towards leveraging web standards is now emerging.</li>
<li><strong>C# readonly vs const: a subtle performance trap</strong>: A production incident revealed that improper use of <code>readonly</code> in C# can cause unexpected memory allocations and performance degradation, highlighting a subtle but critical distinction from <code>const</code> often overlooked by developers.</li>
<li><strong>Piwik Pro eliminates free plan, impacting open-source tools</strong>: Piwik Pro has ended its free tier, introducing a minimum annual charge of €420, directly impacting individual developers and open-source projects that relied on free access for testing and support, with one R package maintainer re-evaluating support.</li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Flock’s gunshot detection now wants to listen for screams</strong>: Flock Safety, already under fire for its ineffective and privacy-invasive gunshot detection system, is now expanding its microphones to listen for human distress sounds like screaming. Because nothing says ‘public safety’ like a network of AI-powered microphones constantly eavesdropping on everyone, despite a track record of over 99% false alarms and police shooting at children after misidentifying fireworks. What could possibly go wrong with more audio surveillance and less actual problem-solving?</li>
<li><strong>Instacrops to demo ‘water-saving, crop-boosting AI’ at TechCrunch Disrupt</strong>: Instacrops, a startup that processes 15 million data points per hour to tell farmers how to water their plants, is heading to TechCrunch Disrupt to show off its ‘AI’ that supposedly reduces water use by 30% and boosts yields by 20%. They pivoted from hardware to software, which is usually a good sign they couldn’t make the hardware work. So, they’re basically selling fancy irrigation advice via WhatsApp, powered by ‘AI’ that analyzes soil moisture. Revolutionary.</li>
</ul>
