<h2>Must know</h2>
<ul>
<li><strong>Iran jams Starlink internet with military-grade GPS</strong>: Iran successfully used military-grade GPS jammers to disrupt Starlink satellite internet, cutting performance by up to 80% during a nationwide internet shutdown. This technique targeted GPS signals essential for Starlink terminals, proving effective where similar Russian attempts failed. The incident raises serious concerns about Starlink’s reliability in conflict zones and its touted censorship-resistant capabilities, as reported by various sources. <a href="https://app.daily.dev/posts/VVK9EFJ8w" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Snap Store publisher domains hijacked for malware distribution</strong>: Scammers are exploiting the Canonical Snap Store by registering expired domains previously used by legitimate publishers, then using these hijacked accounts to distribute malicious updates to trusted snap applications. This new tactic, targeting cryptocurrency wallet users, undermines trust in the Snap Store’s publisher reputation signals. Publishers are advised to maintain domain control and enable 2FA, while users should be cautious with crypto apps from Snap, according to cybersecurity reports. <a href="https://app.daily.dev/posts/8RJLmL5a0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>China plans massive 193,428-satellite constellation</strong>: China has filed ITU requests for two satellite constellations, CTC-1 and CTC-2, totaling an astounding 193,428 satellites—four times larger than SpaceX’s Starlink plans. The filings, made by a newly established entity, have raised national security concerns for the US FCC, which has accelerated approvals for domestic satellite networks in response. While China’s current deployment is limited, the move signals a significant escalation in the space race, as reported by various news outlets. <a href="https://app.daily.dev/posts/9Dqo8zul1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Erlang socket throughput gets a boost</strong>: An Erlang developer implemented sendmmsg and recvmmsg support for OTP’s socket module, improving throughput by 50% over gen_udp for batch processing of network packets by reducing context switches. <a href="https://app.daily.dev/posts/k54Xnpxbv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Figma gets a command-line tool for AI and developers</strong>: dannote/figma-use is a new CLI tool offering full read/write access to Figma, supporting over 100 commands and JSX syntax for programmatic design creation, optimized for token efficiency with AI agents. <a href="https://app.daily.dev/posts/DfAoRLrdU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Curl drops OpenSSL-QUIC backend</strong>: The curl project is removing support for the OpenSSL-QUIC backend in version 8.19.0 due to its inadequate API, extremely poor performance (3x slower), and excessive memory usage (20x more) compared to alternatives. <a href="https://app.daily.dev/posts/As2f3e0AW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open source models: cathedral, megachurch, and bazaar</strong>: A new analysis categorizes open source projects into three models: exclusive ‘cathedrals,’ large ‘megachurches’ like the Linux Foundation, and the vast majority of ‘bazaar’ projects, which are often single-maintainer and critically underfunded. <a href="https://app.daily.dev/posts/rr8Vo8FaG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Compiling Scheme to WebAssembly with GC</strong>: A detailed account explores adding a WebAssembly compiler to Bob, a 15-year-old Scheme implementation, tackling challenges like representing Scheme objects as WASM GC structs and implementing built-ins directly in WASM text. <a href="https://app.daily.dev/posts/oy0rppWeR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Physical AI and AV industry shifts at CES 2026</strong>: CES 2026 highlighted Physical AI, with Chinese automotive and AV tech companies dominating, Mobileye acquiring Mentee Robotics for $900 million, and Motional shifting to an ‘AI-first’ AV development process. <a href="https://app.daily.dev/posts/MtQzzuuIN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vertex AI pricing for 2026</strong>: Vertex AI pricing for its 16 core services in 2026 shows Gemini models ranging from $0.10 to $120 per million tokens, with custom training and prediction costs varying significantly by machine type and usage. <a href="https://app.daily.dev/posts/AslvqVnql" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>GPT-5.2 Pro solves Erdos problem #281, or so they say</strong>: Apparently, GPT-5.2 Pro has ‘solved’ Erdos problem #281, a feat previously beyond human mathematicians, with Terence Tao himself calling it ‘perhaps the most unambiguous instance’ of AI solving an open problem. While impressive, the complete lack of technical details on the solution or any independent verification makes one wonder if this is a genuine breakthrough or just another AI hype cycle fueled by a single social media post. Let’s see the math, not just the marketing. <a href="https://app.daily.dev/posts/pRkd3CPEF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-enforced RFCs: because undisciplined AI code is a problem now?</strong>: govctl is a new CLI tool that ‘enforces RFC-driven development workflows for AI-assisted coding,’ because apparently, AI-generated code is so undisciplined it needs a Rust-based babysitter to ensure it follows specs. It’s designed for teams who want ‘structured, spec-first development rather than rapid prototyping,’ which sounds suspiciously like adding more bureaucracy to the already over-engineered process of managing AI’s ‘creativity.’ <a href="https://app.daily.dev/posts/oBJvqLGNT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Multi-agent AI for SRE: because humans are just too slow for incidents</strong>: Multi-agent AI systems are now emerging to ‘assist’ SRE teams during incident response, automating ‘tedious investigation tasks’ while supposedly keeping humans ‘in control.’ Research suggests centralized architectures work best, but production readiness is still a ‘challenge.’ So, we’re building complex AI systems to debug other complex systems, with the caveat that they still ‘fall short of human-guided analysis’ in complex scenarios. What could possibly go wrong? <a href="https://app.daily.dev/posts/T6lGgvwh" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
