<h2>Must know</h2>
<ul>
<li><strong>Cursor releases Composer 2 and challenges frontier model dominance</strong>: Cursor has launched Composer 2, a 1.04 trillion parameter Mixture-of-Experts model trained specifically on real-world engineering sessions rather than standard benchmarks. According to Cursor, the model outperforms Claude Opus 4.6 on their proprietary CursorBench by focusing on underspecified, high-churn tasks that mirror actual software development. This release signals a shift toward domain-specialized AI that can beat general-purpose frontier models while operating at a fraction of the inference cost. <a href="https://app.daily.dev/posts/osQspcxBm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic source code leak exposes internal agent architecture</strong>: An accidental inclusion of source maps in a Claude Code npm package has exposed over 500,000 lines of Anthropic’s proprietary TypeScript code. The leak reveals unreleased features like KAIROS, an autonomous background agent, and internal ‘anti-distillation’ mechanisms designed to poison competitor training data. While no customer data was compromised, security researchers are already using the blueprints to identify deterministic exploitation paths in AI context management pipelines. <a href="https://app.daily.dev/feed-by-ids?id=OULS6HmRL&amp;id=NF4kLGa2o&amp;id=OwgWFCuNb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Nvidia acquires SchedMD and tightens grip on supercomputing</strong>: Nvidia’s acquisition of SchedMD, the maintainers of the Slurm workload manager used by 60% of supercomputers, has triggered industry-wide concerns regarding hardware favoritism. While Slurm remains under a GPL v2.0 license, analysts at InfoWorld warn that Nvidia now employs the core developer pool, potentially prioritizing InfiniBand and GPU optimizations over competitors like AMD. Enterprises are being advised to seek vendor-neutral SLA guarantees or consider containerizing workloads to maintain future migration paths to Kubernetes or Flux. <a href="https://app.daily.dev/posts/RZr93bEOr" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Webflow open sources Codeflow for AI evaluative interviews</strong>: The environment allows interviewers to observe how candidates prompt and integrate AI output within their own local IDE setup. <a href="https://app.daily.dev/posts/zPppPEY2W" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Temporal Spring update adds Azure support and AI SDK integration</strong>: New features include worker versioning, a Terraform provider, and native integration with the OpenAI Agents SDK for durable agentic workflows. <a href="https://app.daily.dev/posts/RzyA9Q4cp" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft releases GitHub Copilot extension for .NET modernization</strong>: The tool generates migration plans and automated tasks to assist in upgrading legacy .NET applications to modern LTS versions. <a href="https://app.daily.dev/posts/cIRwdIriF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes community launches AI Conformance program</strong>: The program standardizes Dynamic Resource Allocation for GPUs and all-or-nothing scheduling to prevent deadlocks in distributed training. <a href="https://app.daily.dev/posts/kcqVDySrK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gemma Gem runs Google models entirely on device via WebGPU</strong>: This Chrome extension enables local LLM execution for webpage interaction without requiring API keys or sending data to the cloud. <a href="https://app.daily.dev/posts/bpv3Ekj3e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI retires older GPT 5.1 and 5.2 Codex models</strong>: Effective April 14, several legacy models will be removed for users signing in with ChatGPT accounts. <a href="https://app.daily.dev/posts/BRGw3uvEw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Netflix open sources VOID for physical interaction deletion</strong>: The model removes objects from video while intelligently inpainting the resulting physical changes, such as gravity-defying effects. <a href="https://app.daily.dev/posts/WK3IPorts" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The AI trust gap is widening despite the adoption hype</strong>: Stack Overflow’s latest data shows that while 84% of developers use AI tools, trust in their accuracy has plummeted to just 29%. We are currently in a ‘subprime technical debt’ phase where teams are shipping unreviewed ‘vibe code’ under the assumption that a future, smarter model will eventually fix the mess. <a href="https://app.daily.dev/feed-by-ids?id=pTGxr2aPB&amp;id=uLGIp8OgK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Copilot is legally just for entertainment</strong>: Despite being marketed as a ‘dependable workmate’ for the enterprise, Microsoft’s terms of service still explicitly state Copilot is ‘for entertainment purposes only.’ It is a classic case of corporate hedging: they want your subscription revenue for ‘productivity’ but accept zero liability when the probabilistic parrot hallucinates your quarterly projections. <a href="https://app.daily.dev/feed-by-ids?id=QPAxbn9Aj&amp;id=chPbgLDu0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The myth of the 10x AI productivity boost</strong>: A METR study found that experienced developers were actually 19% slower when using AI tools on mature codebases, despite believing they were 20% faster. The industry is mistaking ‘generation speed’ for ‘delivery speed,’ ignoring the massive comprehension debt and review tax that AI-generated spaghetti imposes on senior staff. <a href="https://app.daily.dev/feed-by-ids?id=0NLiScdCr&amp;id=hzhT2J9Is" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
