<h2>Must know</h2>
<ul>
<li><strong>Anthropic and OpenAI release new models, AWS invests big in AI infrastructure</strong>: Anthropic launched Claude Opus 4.6 with ‘agent teams’ for parallel task execution and an expanded 1 million token context window, aiming to appeal to a broader range of knowledge workers. Not to be outdone, OpenAI released GPT 5.3 Codex, which reportedly outperformed Claude Opus 4.6 on benchmarks and offers real-time steering. Meanwhile, AWS reported strong Q4 2025 growth and plans to invest $200 billion in 2026 to double its datacenter capacity by 2027, primarily for AI infrastructure, indicating a massive bet on the future of AI compute. <a href="https://app.daily.dev/feed-by-ids?id=G7Sre82A1&amp;id=Ijw4mCG7B&amp;id=1BAlXIloL&amp;id=CZUzCGBQG&amp;id=I2PdjppNx&amp;id=IDVrBk7Fu&amp;id=J2clcSBHP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI bot traffic escalates, sparking an arms race with publishers</strong>: AI bots are now a significant portion of web traffic, with two main types: those scraping for model training and those retrieving real-time info for AI agents. Data from Akamai and TollBit shows a rise in both since mid-2025. Publishers are deploying aggressive defenses, leading to an ‘arms race’ as bots develop sophisticated tactics to bypass protections. This conflict is predicted to fundamentally reshape the internet, potentially changing who the primary users are. <a href="https://app.daily.dev/posts/5O2tzW5oh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft develops LLM backdoor scanner, Google launches developer knowledge API</strong>: Microsoft has developed a scanner to detect backdoors in open-weight AI models by identifying malicious triggers embedded during training, addressing a critical supply chain risk for enterprises. Concurrently, Google launched its Developer Knowledge API and Model Context Protocol (MCP) server in public preview. This API provides programmatic access to official Google developer documentation, aiming to keep AI-powered developer tools updated with accurate, current information and prevent issues with outdated training data. <a href="https://app.daily.dev/feed-by-ids?id=O9jvh85A8&amp;id=X3xuKKIe1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Databricks introduces MemAlign for LLM evaluation</strong>: Databricks launched MemAlign, a new MLflow framework that uses a dual-memory system (semantic and episodic) to reduce the cost and latency of training LLM-based judges, enabling rapid adaptation to new domains. <a href="https://app.daily.dev/posts/eCdeDEih6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Enterprise AI agent security gets a dedicated solution</strong>: Operant AI launched Agent Protector, a real-time security solution for autonomous AI agents, addressing challenges like shadow agent discovery, privilege escalation, and data exfiltration within enterprise environments. <a href="https://app.daily.dev/feed-by-ids?id=0e28QTlYb&amp;id=2vNCw5dB4&amp;id=the2EenMO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Postgres and ScyllaDB enhance vector search capabilities</strong>: Postgres is being positioned as a consolidated solution for various database needs through extensions, while ScyllaDB integrated vector search using the USearch library, claiming 10x faster performance than Faiss. <a href="https://app.daily.dev/feed-by-ids?id=QtM7zNGDf&amp;id=VBHSJlMha&amp;id=pKNE7iWlw&amp;id=xEaqTmGWC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Valve delays hardware due to AI-driven component shortages</strong>: Valve has delayed pricing and shipping for its Steam Machine, Steam Frame VR headset, and new Steam Controller, attributing the delays to industry-wide memory and storage component shortages driven by AI data center demand. <a href="https://app.daily.dev/posts/iXYPYa4qJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Old driver flaw disables modern EDR</strong>: Threat actors exploited a legitimate, decade-old EnCase forensic tool’s Windows kernel driver to disable security products, bypassing modern verification due to Windows’ backward compatibility policies. <a href="https://app.daily.dev/posts/9GskJTmwz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Actions usability issues highlighted</strong>: A critical assessment argues that GitHub Actions, despite its market share, suffers from poor log viewing, slow feedback loops, complex YAML, and security risks, suggesting alternatives like Buildkite offer a superior developer experience for production teams. <a href="https://app.daily.dev/posts/L0ni7jj2k" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Java and C# design philosophies compared</strong>: A talk by Adele Carpenter at NDC London 2026 provocatively titled ‘Java Sucks (So C# Didn’t Have To)’ provides a balanced technical comparison, attributing Java’s design choices to backward compatibility and community evolution, while C# benefited from a second-mover advantage. <a href="https://app.daily.dev/feed-by-ids?id=MWj0SYCOr&amp;id=PNWmQO4PV&amp;id=SvHH72Vwr&amp;id=Vsxz1h8Es&amp;id=qw41IUDAz" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Vibe coding: the next big thing or just prompt engineering with extra steps?</strong>: Apparently, ‘Vibe coding’ is the ‘evolution’ of prompt engineering, where AI agents autonomously operate within predefined frameworks. So, instead of writing code, we’re now writing specifications for AI to implement. It’s almost like we’re just calling it something new to make it sound more profound than ‘telling the AI what to do, but with more structure.’ <a href="https://app.daily.dev/posts/wypgtlifg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-optimized phone plans: because your phone needs more AI, not better battery life</strong>: SpaceX is reportedly exploring a Starlink-branded mobile phone ‘optimized for AI neural networks.’ Because what everyone really needs is a phone that can run more AI locally, rather than, you know, just having a reliable signal or a battery that lasts more than a day. Clearly, the future of mobile is less about communication and more about on-device inference. <a href="https://app.daily.dev/feed-by-ids?id=MXFPbStCq&amp;id=afU8DAFQ0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI research newsletter from OpenJDK: because Java developers are now AI experts?</strong>: A ‘curated newsletter’ from ‘foojay – a place for friends of OpenJDK’ is covering deep AI research, including agent-based prediction market manipulation and LLM hallucination control. It’s great to see OpenJDK branching out, but one has to wonder if the ‘friends of OpenJDK’ are truly the target audience for cutting-edge AI agent research. Or maybe they just ran out of Java news. <a href="https://app.daily.dev/posts/nRGZFexXd" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
