<h2>Must know</h2>
<ul>
<li><strong>OpenAI hits 730 billion valuation via circular compute deals</strong>: OpenAI secured $110 billion from Amazon, Nvidia, and SoftBank, but the deal structure is heavily tied to infrastructure commitments rather than pure cash. Amazon’s $50 billion contribution is largely a compute credit for its Trainium accelerators, while Nvidia’s $30 billion is linked to deploying five gigawatts of its upcoming Vera Rubin systems. This massive valuation leap comes despite OpenAI not expecting profitability until 2029, as reported by multiple financial outlets. <a href="https://app.daily.dev/feed-by-ids?id=8imM4YubL&amp;id=ShShzSJ3B" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google API keys silently granted Gemini access</strong>: Truffle Security researchers discovered that GCP API keys, originally intended for billing, were quietly upgraded by Google to act as authentication credentials for the Gemini AI API. This change occurred without developer notification, leading to nearly 3,000 exposed keys in the wild that could allow attackers to access private datasets or exhaust API quotas. Google initially dismissed the report as intended behavior before acknowledging it as a bug and restricting the leaked keys. <a href="https://app.daily.dev/feed-by-ids?id=InhdaYtFz&amp;id=0amBk9Q3T" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AirSnitch attack bypasses WiFi client isolation</strong>: Researchers at UC Riverside introduced AirSnitch, a set of vulnerabilities that bypasses client isolation on WPA2 and WPA3 networks to enable man-in-the-middle attacks. The flaws stem from a lack of standardization in IEEE 802.11 and allow for traffic interception and manipulation even on encrypted networks. All 11 tested devices from major vendors like Cisco, Ubiquiti, and Netgear were found vulnerable to at least one form of the attack. <a href="https://app.daily.dev/posts/tSnGJTDSb" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Vercel open sources React optimization rules for AI agents</strong>: A new repository of 40+ architectural rules designed to help LLMs like Claude and Cursor write more performant Next.js code. <a href="https://app.daily.dev/posts/J6VGCUvIf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Perplexity pivots to 200 dollar monthly enterprise agent</strong>: The company is abandoning ads to focus on ‘Perplexity Computer,’ an agentic tool that orchestrates 19 different models for complex workflows. <a href="https://app.daily.dev/posts/JYBUJ8DSb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Sandia Labs tests reconfigurable supercomputer to rival Nvidia</strong>: The Spectra system uses NextSilicon Maverick-2 chips that reconfigure hardware data flow dynamically, claiming half the power of Blackwell. <a href="https://app.daily.dev/posts/t9r2wgx3C" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>India blocks Supabase access under IT Act</strong>: The government issued a sudden blocking order affecting major ISPs, disrupting the fourth-largest market for the open-source Firebase alternative. <a href="https://app.daily.dev/posts/x9ggXCarx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubetorch brings remote execution to local PyTorch code</strong>: A new Kubernetes ecosystem project allows ML code to run remotely via a simple .to() API with sub-2-second hot reloading. <a href="https://app.daily.dev/posts/3II62MbL8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Netflix retreats from Warner Bros Discovery bidding war</strong>: Netflix withdrew its $82 billion bid after Paramount Skydance raised its offer to $31 per share, ending a nine-month acquisition battle. <a href="https://app.daily.dev/posts/JTzCMWvI7" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical RCE discovered in Juniper core routers</strong>: CVE-2026-21902 allows unauthenticated root access to PTX series routers via the On-Box Anomaly detection framework. <a href="https://app.daily.dev/posts/0X8b0bm1J" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The token economy is the new class warfare</strong>: A new narrative suggests that AI token costs are creating a ‘digital divide’ where only rich developers can afford the ‘scar tissue’ of failing with agents. While it’s a poetic metaphor, it conveniently ignores that local models are getting better and cheaper every week. It’s a classic attempt to gatekeep engineering experience behind a paywall that doesn’t actually exist for anyone with a decent GPU. <a href="https://app.daily.dev/posts/HFI3Yp8H0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Block cuts 4000 jobs and blames the AI boogeyman</strong>: Jack Dorsey’s Block is the latest to claim AI tools allow for smaller teams while laying off 10% of its workforce. Critics and analysts note this looks more like correcting for pandemic-era overhiring and boosting stock price—which jumped 24%—than a sudden leap in agentic productivity. Blaming the ‘efficiency’ of AI is the new corporate PR shield for standard bottom-line restructuring. <a href="https://app.daily.dev/posts/y7FHwZ7mA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The 6 hour AI built dashboard miracle</strong>: The Leet Container Dashboard is being marketed as ‘built in 6 hours’ thanks to Codex and Cline AI. While impressive for a weekend project, this ‘speed of AI’ hype ignores the years of TypeScript experience required to prompt it and the inevitable maintenance debt of unvetted, generated code. It’s a great demo, but building a dashboard isn’t the same as engineering a system. <a href="https://app.daily.dev/posts/GVRafeRV4" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
