<h2>Must know</h2>
<ul>
<li>AI coding agents are still a mess, but now they’re making your code worse: Remember when AI was supposed to make us all 10x developers? Turns out, AI coding assistants like GitHub Copilot and Cursor are indeed generating code faster, but they’re also introducing more bugs, security vulnerabilities, and architectural debt, according to Cursor and The New Stack. Dropbox, for instance, saw a million lines of AI-suggested code monthly, but the overall impact on code quality and architectural integrity remains questionable. It seems we’re trading typing speed for long-term headaches, with some studies even suggesting developers using AI are 19% slower overall due to the need for extensive review and rework. <a href="https://app.daily.dev/feed-by-ids?id=J24eplQdr&amp;id=iT5DCqanN&amp;id=EHuoSn58k&amp;id=pc6GKGFyk&amp;id=qu9MmthdT&amp;id=yJ8WtVNig&amp;id=5joRPhN74&amp;id=1w09x1vo7&amp;id=6aVFrCy8v" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>OpenAI’s market share is tanking, but they’re still trying to make money: OpenAI’s enterprise LLM market share has plummeted from 50% in 2023 to a mere 27% by late 2025, with Anthropic and Google eating their lunch. In response, OpenAI is launching ChatGPT Go at $8/month with ads, and plans to introduce ads for free users, as reported by AI News. Meanwhile, they’re trying to shore up their enterprise game by hiring new sales leadership and expanding partnerships with companies like ServiceNow. It seems the ‘move fast and break things’ ethos is now ‘move fast and try to find a business model that sticks.’ <a href="https://app.daily.dev/feed-by-ids?id=KwZZ88381&amp;id=9fjLFvZxt&amp;id=JTChiywvF&amp;id=MkQL2zAyT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI is making public knowledge platforms obsolete, threatening future models: The rise of AI tools like ChatGPT and Claude is causing a significant drop in user engagement on public knowledge platforms like Stack Overflow (down 78%) and Wikipedia. Developers are increasingly using AI privately to solve problems, reducing collaborative knowledge sharing. This creates a ‘knowledge collapse’ feedback loop: AI models trained on public data are now driving users away from those sources, potentially degrading the quality of future training data and consolidating power with corporate AI providers, according to the article ‘AI Tools Threaten Sustainability of Public Knowledge Platforms’. <a href="https://app.daily.dev/feed-by-ids?id=3Qlqmk7FI&amp;id=p61M36Xco" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li>Microsoft’s new AI chip aims to dethrone Nvidia: Microsoft has unveiled the Maia 200, a custom AI inference chip boasting over 100 billion transistors and claiming 3x the performance of Amazon’s Trainium3 and superior FP8 performance to Google’s TPUs, signaling a major push to reduce reliance on Nvidia. <a href="https://app.daily.dev/posts/fDSp3OUaH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Vercel’s new json-render tool lets AI generate UIs: Vercel has released json-render, an open-source tool that allows AI to dynamically generate user interfaces based on natural language prompts and predefined components, shifting frontend development towards agentic behavior and context engineering. <a href="https://app.daily.dev/posts/5RKyfi3u8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI agents are exposing cloud infrastructure weaknesses: Autonomous AI agents are revealing fundamental flaws in existing cloud infrastructure, including inadequate network segmentation, insufficient identity-based security, and a lack of real-time cost monitoring, demanding a new discipline in cloud operations. <a href="https://app.daily.dev/feed-by-ids?id=1w09x1vo7&amp;id=5joRPhN74&amp;id=6aVFrCy8v" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>DuckDB adds support for Vortex columnar file format: DuckDB now supports Vortex, a new open-source columnar data file format that claims to be 18% faster than Parquet v2 and 35% faster than Parquet v1 for analytical queries, offering lightweight compression and compute expressions on compressed data. <a href="https://app.daily.dev/posts/zpcWwZt9t" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>GitHub CLI integrates Copilot CLI directly: GitHub CLI now supports direct integration with the Copilot CLI via <code>gh copilot</code>, replacing the deprecated extension and streamlining AI-assisted command-line workflows for developers. <a href="https://app.daily.dev/posts/381VmPpMD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Python gets a new JIT compiler and Pandas 3 is out: Python is introducing a new JIT compiler aimed at significantly boosting code execution speed, with a packaging library reportedly running three times faster, alongside the release of Pandas version 3 for data analysis. <a href="https://app.daily.dev/posts/0TjKgLuas" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Vercel Sandboxes now support filesystem snapshots: Vercel Sandboxes have introduced filesystem snapshots, allowing developers to capture and reuse the complete state of a sandbox, eliminating repeated setup steps and enhancing parallel testing and rapid experimentation workflows. <a href="https://app.daily.dev/posts/Uv3Zx9otG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li>AI-powered microsegmentation: the new magic bullet for cyberattacks: Apparently, shutting down operations during a cyberattack is ‘outdated.’ Now, with AI-driven microsegmentation, you can just isolate the compromised bits and keep 80% of your business running. Because, you know, AI just knows how to perfectly segment your network in hours, and attackers will totally respect those boundaries. It’s almost like they’re saying we can have our cake and eat it too, without actually fixing the underlying security issues. <a href="https://app.daily.dev/posts/ijIOxfZ4T" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI-generated code: the gift that keeps on giving… tech debt: A developer built an app in a month using AI, and guess what? Massive tech debt. Turns out, AI is great at generating code, but not so great at generating good code without constant human babysitting and strict guidance. So, while you might get a prototype faster, you’ll be paying for it in maintenance and refactoring down the line. It’s like hiring a junior developer who never sleeps, but also never learns from their mistakes unless you explicitly tell them every single time. <a href="https://app.daily.dev/posts/2tsKdG55u" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI is the new ‘agile’ for executives who want to code: Forget those pesky developers, now executives can ‘vibecode’ their own apps with AI tools, building hackathon platforms in 5 hours for $300. Because nothing says ‘strategic leadership’ like an executive spending their time building a metric alignment app instead of, you know, leading. It’s a hybrid executive role: part-time individual contributor, part-time manager, full-time delusion that AI makes everyone a coder. <a href="https://app.daily.dev/posts/McILqauKu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>