The latest articles on DEV Community by Tech Weekly (@weekly). https://dev.to/weekly https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Forganization%2Fprofile_image%2F1857%2Fa2b77250-f5fc-461f-a1a3-451baed55f59.png https://dev.to/weekly en Madhu Sudhan Subedi Mon, 23 Feb 2026 03:05:01 +0000 https://dev.to/weekly/weekly-06-2026-openais-agentic-ai-push-codex-laravels-ai-sdk-fundamentals-over-frameworks-4mh3 https://dev.to/weekly/weekly-06-2026-openais-agentic-ai-push-codex-laravels-ai-sdk-fundamentals-over-frameworks-4mh3 <h2> 🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/47" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </h2> <h3> <strong>OpenAI's acquisition of OpenClaw signals the beginning of the end of the ChatGPT era</strong> </h3> <p>OpenAI just made a big move that could mark the beginning of the end of the classic “type-and-chat” ChatGPT era. They’ve “acquired” OpenClaw – in practice, hiring its creator Peter Steinberger – and are betting big on agentic AI.<br> OpenClaw is an open-source agent framework that lets AI not just talk, but actually act: browse the web, click buttons, run code, and complete real tasks like booking flights, managing email, or coordinating across multiple tools on your behalf. Think of it as a chat-first automation layer that turns a chatbot into a proactive digital worker.<br> Strategically, this signals that the real AI race is shifting from “who has the best model” to “who owns the most powerful, safe agent ecosystem.” For OpenAI, OpenClaw’s community and tooling drop almost directly on top of the ChatGPT user base, accelerating their push into personal and enterprise agents.<br> But there’s a huge open question: security. Giving an AI agent the power to browse, click, and execute code is basically handing it root-like access to your digital life — and anyone who successfully prompts or attacks that agent could ride along with that access. So while this might be the next chapter after chatbots, it also raises the stakes for AI safety and governance in a big way.</p> <p><a href="https://venturebeat.com/technology/openais-acquisition-of-openclaw-signals-the-beginning-of-the-end-of-the" rel="noopener noreferrer">Link</a></p> <h3> <strong>The Product-Minded Engineer: The importance of good errors and warnings</strong> </h3> <p>There's a growing shift in how we think about great engineering — and it goes beyond writing clean code. The Pragmatic Engineer recently spotlighted a new book by a veteran of Microsoft, Facebook, and Stripe that makes the case for product-minded engineering: a mindset where you're not just shipping features, but asking why they matter — to users, to the business, and to the bottom line.<br> One standout idea: things like error messages and warnings aren't edge cases — they're product surfaces that shape how people experience your system, and they deserve the same care as any core feature.<br> The bottom line for us as engineers — the profiles teams are hungry for right now are those who can move fluidly between code and customer impact. If you challenge requirements, use what you build, and care about real-world outcomes, that's the competitive edge in today's market.</p> <p><a href="https://newsletter.pragmaticengineer.com/p/the-product-minded-engineer" rel="noopener noreferrer">Link</a></p> <h3> <strong>How Codex is built</strong> </h3> <p>how OpenAI’s Codex is actually built – and why it matters for us as engineers.<br> Codex is now a multi‑agent coding assistant used by over a million developers weekly, and OpenAI says usage has grown 5x just since January. The wild part? The team estimates that more than 90% of the Codex app’s code is written by Codex itself.<br> The core CLI is written in Rust, not TypeScript, to prioritize performance, correctness, and a very high engineering bar – plus minimal dependencies so they fully understand what they ship. Under the hood, Codex runs a classic agent loop: assemble a rich prompt with tools and local context, call the model, execute tool calls like reading files or running tests, feed results back, and iterate until the task is done.<br> Internally, engineers behave less like coders and more like “agent managers,” running 4–8 agents in parallel for feature work, code review, security review, and summarization. They’ve built 100+ reusable “skills,” from a security best‑practices pass to a “yeet” skill that turns changes into a draft PR in one shot.<br> Maybe the most interesting bit: OpenAI is training the next Codex using the current one, with Codex helping write and review the code that powers its own successor. If you’re still treating AI as just autocomplete, this is a good reminder that the frontier teams have fully shifted to AI‑first software engineering.</p> <p><a href="https://newsletter.pragmaticengineer.com/p/how-codex-is-built" rel="noopener noreferrer">Link</a></p> <h3> <strong>Learn fundamentals, not frameworks</strong> </h3> <p>In an AI-first world, you should learn fundamentals, not chase every new framework.<br> Most frameworks have a half-life of just a few years, while things like languages, protocols, algorithms, and system design last decades. At the same time, AI is already generating around 41% of all code, but nearly half of that code is high-risk for security vulnerabilities and needs serious review.<br> That means your real edge is not “I know Framework X,” it’s understanding concurrency, data structures, distributed systems, and clean architecture so you can debug, refactor, and make trade-offs when AI-generated code inevitably breaks.<br> A practical rule: spend 80% of your learning time on fundamentals and 20% on frameworks—you’ll always pick up tools on the job, but nobody will teach you system design and deep debugging for free.</p> <p><a href="https://newsletter.techworld-with-milan.com/p/learn-fundamentals-not-frameworks" rel="noopener noreferrer">Link</a></p> <h3> <strong>Laravel Announces Official AI SDK for Building AI-Powered Apps - Laravel News</strong> </h3> <p>Laravel just dropped an official AI SDK, and it might be the moment where “AI in Laravel” becomes truly first‑class.<br> Laravel’s new AI SDK is now part of the 12.x docs and gives you a native API for everything from text generation to embeddings and tool-based agents, all without locking you into a single AI vendor. You can swap between providers like Anthropic, Gemini, OpenAI, ElevenLabs and more literally by changing a single line of config, while smart fallbacks handle rate limits or outages for you.<br> The crazy part is it’s one package for text, images, audio, transcription, embeddings, reranking, vector stores, web search, and even file search—plus you get agents with tools, memory, structured outputs, and streaming, all testable with Laravel-style fakes. This keeps your whole AI stack inside the Laravel ecosystem instead of juggling random third‑party SDKs.<br> If you want to play with it, check out laravel.com/ai and the new AI SDK section in the Laravel 12 docs.</p> <p><a href="https://laravel-news.com/laravel-announces-official-ai-sdk-for-building-ai-powered-apps" rel="noopener noreferrer">Link</a></p> aiagents developertools softwareengineering technews Madhu Sudhan Subedi Sat, 14 Feb 2026 19:22:28 +0000 https://dev.to/weekly/weekly-05-2026-ai-agents-build-compilers-svg-first-apps-markdown-for-agents-and-why-code-is-1km8 https://dev.to/weekly/weekly-05-2026-ai-agents-build-compilers-svg-first-apps-markdown-for-agents-and-why-code-is-1km8 <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/46" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>An SVG Is All You Need: Self‑Contained, Durable, and Interactive</strong> </h3> <p>There’s a fun idea floating around web dev: what if an SVG file was all you needed to build an app? Imagine a single SVG that’s not just an image, but also your UI description, layout, and even some behavior—all in one portable, declarative document. Because SVG is just XML, it’s inherently structured, scriptable, and easy for tools or agents to read and transform, unlike a random canvas blob or screenshot. In a world of increasingly complex web stacks, this pushes a radical opposite vision: ultra-minimal apps where your “frontend” is basically a smart SVG that can be rendered anywhere, styled, versioned, or even edited visually like a design file.</p> <p><a href="https://jon.recoil.org/blog/2025/12/an-svg-is-all-you-need.html" rel="noopener noreferrer">Link</a></p> <h3> <strong>Code Is Still the Best Abstraction</strong> </h3> <p>No-code and low-code tools are great for quick wins and empowering non-engineers, but they lock business logic into proprietary UIs, get messy as pipelines grow, make testing and versioning hard, and end up costly—so the real trend is moving toward smaller, integrated data stacks orchestrated by code-first, declarative tools like Dagster, where your automation lives in open YAML or Python instead of a vendor’s black box.</p> <p><a href="https://www.ssp.sh/brain/no-less-code-vs-code/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Introducing Markdown for Agents</strong> </h3> <p>Cloudflare just launched something called “Markdown for Agents,” and it’s basically SEO for AI agents instead of just humans.<br> The big shift is that traffic is moving from classic search engines to AI crawlers and agents, which don’t want bloated HTML—they want clean, structured text they can parse cheaply and reliably. Feeding an AI the full HTML of a page can use 5x more tokens than the same content in markdown, which directly translates into higher cost and slower pipelines for any agent reading the web.<br> Cloudflare’s play is simple: if an agent sends an Accept: text/markdown header, their network will fetch your normal HTML, convert it to markdown on the fly, and return that instead, along with a header that even tells you how many tokens the markdown roughly uses. This is already being used by popular coding agents like Claude Code and OpenCode, and it works today on Cloudflare’s own docs and blog.<br> On top of that, every markdown response ships with “Content Signals” headers like ai-train=yes, search=yes, ai-input=yes, so you can explicitly say “this content is allowed for AI training, search and agents,” with more granular policies coming later. For anyone building AI agents—or running a site that wants to be discoverable by them—the takeaway is clear: start treating agents as first-class visitors and make sure your content can be served as markdown, not just HTML.</p> <p><a href="https://blog.cloudflare.com/markdown-for-agents/?utm_source=tldrdevops/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Building a C compiler with a team of parallel Claudes</strong> </h3> <p>Anthropic tested what happens if you treat AI like a full dev team instead of just autocomplete.<br> They spun up 16 Claude agents in parallel and asked them to build a new C compiler in Rust that can compile big real‑world codebases like the Linux kernel, QEMU, FFmpeg, Postgres, and Redis. Each agent worked in its own container, grabbed tasks, edited code, ran tests, and pushed to a shared repo, with a strict CI and test harness making sure they didn’t break everything.<br> The result: around 100k lines of compiler code that actually works on multiple architectures, but is still slower than GCC and not production‑ready yet. It still relies on existing tools for some low‑level pieces, but that’s not the point—the experiment shows that with good scaffolding, AI agents can coordinate and ship large, complex systems end‑to‑end.<br> So the simple takeaway for developers: we’re moving from “AI helps you write functions” to “AI can own entire projects,” and our new job is increasingly about setting goals, designing guardrails, and reviewing what these agent teams produce.</p> <p><a href="https://www.anthropic.com/engineering/building-c-compiler" rel="noopener noreferrer">Link</a></p> <h3> <strong>Code is cheap. Show me the talk</strong> </h3> <p>Linus Torvalds once said, “Talk is cheap. Show me the code,” but in 2026 that quote is basically upside down.<br> With LLMs, an average developer can spin up 10,000 lines of decent, working code in hours instead of weeks, so raw code has become abundant, almost commoditized. The old signals we used to judge projects—beautiful READMEs, clean architecture, perfect comments—can now be one-shot AI output, so they no longer prove real effort or expertise. That means the real bottleneck has shifted: it’s not typing or syntax anymore, it’s our ability to think clearly, design systems, and read and evaluate mountains of code to separate solid engineering from AI slop.<br> For seniors, this is a superpower: if you can imagine, articulate, and architect well, you can compress months of work into days by steering AI tools, then spend your energy on hard trade-offs, governance, and quality. But for juniors, there’s a risk of becoming dependent on the “genie,” shipping code you don’t understand and never building the fundamentals to judge what’s good or dangerous.<br> So the big shift is this: in modern software development, good talk—clear thinking, problem framing, and critical reading of code—is now more valuable than good code itself, because code is cheap, but trustworthy judgment and accountability are not.</p> <p><a href="https://nadh.in/blog/code-is-cheap/" rel="noopener noreferrer">Link</a></p> aiagents softwaredevelopment developertools techtrends2026 Madhu Sudhan Subedi Sun, 25 Jan 2026 14:36:09 +0000 https://dev.to/weekly/weekly-04-2026-astro-joins-cloudflare-curl-drops-openssl-quic-ai-cant-replace-developers-3anp https://dev.to/weekly/weekly-04-2026-astro-joins-cloudflare-curl-drops-openssl-quic-ai-cant-replace-developers-3anp <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/45" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>Astro is joining Cloudflare</strong> </h3> <p>Big news in web development this week: Cloudflare has acquired the Astro Technology Company, makers of the popular Astro web framework. Astro powers content-driven websites for major brands including Porsche, IKEA, and OpenAI.<br> The good news for developers? Astro stays open source with MIT licensing and will continue accepting community contributions. All Astro employees are joining Cloudflare and will keep working on the framework.<br> Astro 6 is launching soon with a redesigned development server powered by Vite's new Environments API. This means your local dev environment can run in the same runtime as production, including support for Cloudflare's Durable Objects, D1 databases, and AI agents.<br> The framework has grown rapidly thanks to its focused approach: it's built specifically for content-driven sites, renders HTML server-first for speed, and uses an Islands Architecture that keeps pages fast by default while allowing interactive components when needed.<br> With this acquisition, Cloudflare is doubling down on making web development faster and more accessible for everyone.</p> <p><a href="https://blog.cloudflare.com/astro-joins-cloudflare/?utm_source=tldrdevops/" rel="noopener noreferrer">Link</a></p> <h3> *<em>More HTTP/3 focus, one backend less *</em> </h3> <p>Big news from the curl project this week!<br> The popular networking tool is simplifying its HTTP/3 support by dropping OpenSSL's QUIC backend. The decision? Performance and memory issues.<br> According to curl maintainer Daniel Stenberg, OpenSSL-QUIC was struggling badly. Benchmarks showed competitor ngtcp2 transferring data up to three times faster, while OpenSSL-QUIC consumed up to twenty times more memory.<br> Curl now supports just two HTTP/3 backends: ngtcp2 plus nghttp3, and experimental quiche support.<br> This streamlines the codebase and focuses resources on backends that actually deliver. The change ships with curl version 8.19.0.<br> For developers using curl with HTTP/3, the message is clear: performance matters, and curl is betting on leaner, faster implementations.</p> <p><a href="https://daniel.haxx.se/blog/2026/01/17/more-http-3-focus-one-backend-less/?utm_source=tldrdevops" rel="noopener noreferrer">Link</a></p> <h3> <strong>Why We've Tried to Replace Developers Every Decade Since 1969</strong> </h3> <p>Here's a pattern that's been repeating for over 50 years.<br> Since 1969, every decade has promised to replace developers with easier tools. COBOL would let business people write their own programs. CASE tools in the '80s would generate everything from flowcharts. Visual Basic would make coding simple enough for anyone.<br> Now it's AI's turn to make the same promise.<br> But here's what history teaches us: each tool genuinely helps developers work faster, but none eliminates the core challenge. Software complexity isn't about typing speed or syntax—it's about thinking through edge cases, security, integrations, and maintenance.<br> As one developer puts it: "Software development is thinking made tangible." You can't shortcut that reasoning, no matter how good your tools get.<br> AI will change how we work, but it won't replace the judgment developers bring to complex problems.</p> <p><a href="https://www.caimito.net/en/blog/2025/12/07/the-recurring-dream-of-replacing-developers.html?utm_source=tldrdevops" rel="noopener noreferrer">Link</a></p> <h3> <strong>Design Systems for Software Engineers</strong> </h3> <p>Ever wonder why enterprise software looks so polished? The secret is design systems.<br> The Pragmatic Engineer breaks down how companies like Rubrik transform their UIs from "college project" to award-winning interfaces. It starts with a wake-up call—often from leadership noticing the product looks amateur.<br> Building a design system is surprisingly complex. Just designing a button requires decisions on six key factors: color, text, height, border radius, shadows, and fonts. Then you need to consider hover states, animations, accessibility, localization, and mobile responsiveness.<br> But the payoff is huge: faster delivery, higher quality, and that "premium feel" that justifies enterprise pricing.<br> The article also covers how AI fits in. While AI can't generate an entire design system from a prompt, it excels at writing unit tests and catching edge cases.<br> For any company scaling past product-market fit, a design system becomes essential infrastructure.</p> <p><a href="https://newsletter.pragmaticengineer.com/p/design-systems-for-software-engineers" rel="noopener noreferrer">Link</a></p> <h3> <strong>Should CSS Be Constraints? Predictability, Edge Cases, and Better Mental Models</strong> </h3> <p>Why is CSS layout so frustrating? A researcher who wrote the first formal specification of CSS 2.1 layout weighs in.<br> The problem isn't that CSS is non-deterministic—it's that the rules are incredibly complex. For example, text-align: center actually becomes left-aligned if the content is too wide. Why? So users can scroll to see overflowing text instead of it spilling off-screen where it's unreachable.<br> Some suggest replacing CSS with constraint-based systems, like iOS uses. But constraints have their own issues: layouts can be under-determined or over-determined, making debugging tedious and unpredictable.<br> The real insight? Design is genuinely hard. CSS embeds decades of design wisdom into rules that handle edge cases most developers never think about.<br> The solution isn't starting over—it's what CSS is already doing: providing better layout modes like Flexbox and Grid that match modern design needs better than the old text-centric flow layout.</p> <p><a href="https://pavpanchekha.com/blog/why-css-bad.html" rel="noopener noreferrer">Link</a></p> webdev opensource devops technews Madhu Sudhan Subedi Sun, 18 Jan 2026 14:16:11 +0000 https://dev.to/weekly/weekly-03-2026-react2shell-zero-day-defense-kubernetes-autoscaling-guide-ai-coding-guardrails--3e0 https://dev.to/weekly/weekly-03-2026-react2shell-zero-day-defense-kubernetes-autoscaling-guide-ai-coding-guardrails--3e0 <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/44" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>How Unix User Isolation Prevented React2Shell in a Next.js Production</strong> </h3> <p>Puncoz, a seasoned developer, recently faced a real-world cyberattack on his production server and lived to tell the tale. React2Shell, CVE-2025-55182, is a critical CVSS 10.0 zero-day vulnerability in React Server Components that allows unauthenticated remote code execution through unsafe deserialization in server actions—one malicious HTTP request and attackers gain full code execution on your server. When Puncoz's production server was hit, what saved him wasn't fancy security tools or modern defenses, but basic Unix user isolation from the 1970s. Each service ran as a different user: Node.js app as 'nodeapp', database as 'postgres', web server as 'nginx'. Yes, the attackers got code execution, but they could only access files owned by the nodeapp user—database credentials, SSH keys, and production secrets all remained protected behind simple permission boundaries. Puncoz resolved the situation by immediately patching the vulnerability and conducting a full security audit, but the damage was already contained thanks to proper isolation. The lesson? Sometimes the oldest security principles are the most powerful: defense in depth, principle of least privilege, and proper user isolation can stop even the most critical modern vulnerabilities. The old ways are still the best ways.</p> <p><a href="https://blog.puncoz.com/how-basic-unix-user-isolation-saved-my-server-from-react2shell-cve-2025-55182" rel="noopener noreferrer">Link</a></p> <h3> <strong>Kubernetes Autoscaling: Which Tool Fits Your Workload Best?</strong> </h3> <p>How do you choose the right autoscaler for a Kubernetes cluster when options abound? Fve major tools—HPA, VPA, Cluster Autoscaler, Karpenter, and KEDA—breaking down their strengths, limitations, and best-fit scenarios. HPA handles pod scaling based on resource metrics, VPA tunes resource requests, Cluster Autoscaler and Karpenter manage node provisioning, and KEDA excels at event-driven scaling. No single tool covers every scaling need, so understanding their differences is key.<br> Real-world clusters often combine these tools for smarter, more efficient scaling. For example, pairing HPA with Karpenter and CloudPilot AI allows for rapid, cost-effective node provisioning, especially with Spot Instances. KEDA shines for scaling workloads triggered by queues or external events, while VPA is best for optimizing resources in stable, long-running jobs. Using only HPA and Cluster Autoscaler risks leaving significant cloud savings and resilience on the table.<br> For DevOps teams, the takeaway is clear: modular, composable autoscaling strategies deliver better reliability and cost control. As Kubernetes workloads diversify, combining autoscalers based on workload patterns—rather than sticking to defaults—can help teams scale smarter and spend less.</p> <p><a href="https://www.cloudpilot.ai/en/blog/k8s-autoscaling-comparison/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Streamlining CLAUDE.md: Practical Guardrails for AI-Assisted Coding</strong> </h3> <p>A pared‑down CLAUDE.md the author uses to guide Claude Code. The emphasis is on simple, testable progress: small commits that compile, learning from existing code, and choosing boring solutions over clever tricks. Technical standards center on composition, interfaces, explicit data flow, and test‑driven habits, with firm rules on error handling—fail fast, include context, and never swallow exceptions.<br> What’s the interesting shift? The file now aligns with Claude’s built‑in planning modes instead of fighting them, and trims guidance to decisions Claude should follow while coding. There’s a strong focus on project integration—reuse the repo’s build, tests, formatter, and patterns; don’t add tools without justification. Concrete do‑nots include bypassing commit hooks, disabling tests, or committing broken code. For teams experimenting with AI assistants, the takeaway is clear: codify conservative engineering norms, enforce incremental delivery, and let AI operate within your project’s constraints rather than reinvent them.</p> <p><a href="https://www.dzombak.com/blog/2025/12/streamlining-my-user-level-claude-md/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Welcome to 2026: AI progress is accelerating across models, hardware, robotics, and economics, reshaping 2026.</strong> </h3> <p>Alex Wissner-Gross highlights rapid advances: models gaining novel capabilities, like Claude managing a tomato plant, and iQuest’s looped transformer hitting SOTA on SWE-bench. Hardware scale is exploding, with xAI at 450k GPUs and plans for 900k, while financiers back private power to feed AI demand amid projected US shortfalls.<br> New computing fronts emerge: photonic devices promising 10x energy efficiency, DNA data storage 1,000x denser than tape, and TSMC fast-tracking 1.4 nm as Nvidia races to meet China’s H200 demand. Robotics and sensors cross milestones with a zero-disengagement coast-to-coast autonomous drive and e-skin that detects pain; defense fields operational laser interception.<br> Space manufacturing switches on, producing ultra‑pure crystals, and Starlink becomes a travel utility. Markets respond: major AI IPOs loom, Yale economists project sizable productivity gains, and compensation surges at leading firms. Neuralink targets high‑volume production and automated surgery in 2026. The thesis: the singularity is compounding; the greatest risk is slowing down.</p> <p><a href="https://x.com/alexwg/status/2006719715457061081" rel="noopener noreferrer">Link</a></p> <h3> <strong>Build user agents first, then co-develop software with them to achieve true quality.</strong> </h3> <p>Faster “vibe coding” makes shipping easy, but quality depends on understanding users. Traditional testing and coverage can’t guarantee real-world reliability because they miss user context. The proposal: “vibe code” users as agents—rich personas with happy-path flows and day-in-life profiles—then iterate software against these simulated users.<br> The workflow is cyclical: model users, build software, let agentic and real users run flows, gather feedback, refine both users and product, repeat. Success hinges on simplicity for users, enabled by deep understanding of their workflows and needs. Compared to personas and TDD, this approach makes personas executable agents that actually use the product, shifting validation to end-to-end behavior before and during development.</p> <p><a href="https://dima.day/blog/build-software-build-users/" rel="noopener noreferrer">Link</a></p> cybersecurity kubernetes ai devops Madhu Sudhan Subedi Tue, 13 Jan 2026 14:11:26 +0000 https://dev.to/weekly/weekly-02-2026-new-emails-angry-coders-and-the-editor-engineer-era-4nnf https://dev.to/weekly/weekly-02-2026-new-emails-angry-coders-and-the-editor-engineer-era-4nnf <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/43" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>Google is rolling out a feature that allow users to change their Gmail address</strong> </h3> <p>Big news for Gmail users! Google is finally rolling out a highly requested feature that lets you change your existing email address without creating a brand-new account. Yes, you can finally ditch that embarrassing username from high school!</p> <p>Here’s how it works: You switch to a new address, but your old one automatically becomes an 'alias.' This means you still receive all your emails, and importantly, you keep all your data—Photos, Drive files, and contacts—right where they are.</p> <p>Reports indicate this was first spotted in Google’s Hindi support pages, meaning it might be rolling out in India first. Just a heads-up: you can only do this once every 12 months. Check your 'Personal Info' settings to see if the update has reached you!</p> <p><a href="https://www.cnbc.com/2025/12/26/google-gmail-change-email-address-without-new-account-india-hindi-support.html" rel="noopener noreferrer">Link</a></p> <h3> <strong>Co-Creator of Go Language is Rightly Furious Over Appreciation Email</strong> </h3> <p>Never make a computer legend angry! Rob Pike, co-creator of the Go programming language, just went off after receiving an AI-generated thank-you email.</p> <p>It turns out, an AI project called 'AI Village' gave agents a goal: 'random acts of kindness.' The AI decided this meant spamming famous programmers with unsolicited praise. But Pike wasn't having it.</p> <p>He posted a furious response on Bluesky, cursing the waste of energy and resources used by these 'vile machines' just to send fake gratitude. The takeaway? Maybe keep the 'random acts of kindness' to actual humans.</p> <p><a href="https://itsfoss.com/news/rob-pike-furious/" rel="noopener noreferrer">Link</a></p> <h3> <strong>When AI writes almost all code, what happens to software engineering?</strong> </h3> <p>If AI starts writing 90% of our code, are software engineers doomed? The Pragmatic Engineer just dropped a deep dive on this, and the answer is... no, but your job is definitely changing.</p> <p>The big shift? Engineers will move from being 'writers' to 'editors.' Instead of typing out syntax, you’ll spend your time reviewing AI output, debugging complex logic, and architecting systems.</p> <p>The article warns: Junior devs might struggle to learn without the 'grunt work,' while Seniors will become super-productive '10x' engineers. The future isn't about knowing how to code, but knowing what to code. Are you ready to be an editor?</p> <p><a href="https://newsletter.pragmaticengineer.com/p/when-ai-writes-almost-all-code-what" rel="noopener noreferrer">Link</a></p> <h3> <strong>Culture, Not Campuses, Built Modern AI</strong> </h3> <p>Daniel Lemire argues today’s AI breakthrough sprang from cultural forces. Gaming pushed powerful linear‑algebra GPUs; web culture created a vast, networked library. Together, they enabled generative AI. Innovation didn’t flow linearly from universities to industry; it emerged where culture demanded it. <br> The implication: to understand and predict technological progress, track cultural drivers and hacker norms—not just academic publications.</p> <p><a href="https://lemire.me/blog/2026/01/01/technology-is-culture" rel="noopener noreferrer">Link</a></p> <h3> <strong>Pv6 just turned 30 yet it hasn’t taken over.</strong> </h3> <p>IPv6 expanded address space massively, yet adoption remains under half because it wasn’t backward-compatible, added few must-have features beyond addresses, and IPv4 gained workarounds like NAT. Migration costs, complexity, and uneven performance kept many on IPv4, often running dual stacks or disabling IPv6.<br> Experts argue IPv6 did its job: it enabled scale in mobile, broadband, cloud, IoT, and cleaner network planning. Meanwhile, the internet shifted toward name-based architectures and protocols like QUIC, reducing reliance on permanent IPs and further dulling the incentive to switch.<br> Today, organizations adopt IPv6 mainly on cost and scalability grounds—when IPv4 addresses and bigger NATs get too expensive. Some large players seek vast IPv6 blocks, nudging adoption past 50% in parts of Asia. <br> The takeaway: IPv6 is not a failure; it’s a targeted success in growth domains, with broader migration driven by economics, not features.</p> <p><a href="https://www.theregister.com/2025/12/31/ipv6_at_30" rel="noopener noreferrer">Link</a></p> aiengineering futureoftheinternet developerculture techtrends2026 Madhu Sudhan Subedi Sun, 04 Jan 2026 03:05:48 +0000 https://dev.to/weekly/weekly-53-2025-llm-workflows-code-bottlenecks-ai-adoption-in-2026-c26 https://dev.to/weekly/weekly-53-2025-llm-workflows-code-bottlenecks-ai-adoption-in-2026-c26 <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/42" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>My LLM coding workflow going into 2026</strong> </h3> <p>AI coding assistants transformed development in 2025, but using them effectively requires structure and discipline. Google engineer Addy Osmani shares his proven workflow going into 2026.<br> The key insight? Treat AI as a powerful pair programmer, not autonomous magic. Start with detailed specs before writing any code. Break work into small, testable chunks. Provide extensive context about your codebase and constraints. And critically—always review and test everything the AI generates.<br> At Anthropic, 90% of Claude Code is now written by Claude Code itself, but engineers maintain strict oversight. <br> The pattern is clear: AI amplifies your engineering skills when you stay in control.<br> Bottom line is to plan carefully, iterate in small steps, test thoroughly, and commit often. AI coding tools are incredible force multipliers, but the human engineer remains the director of the show.</p> <p><a href="https://addyo.substack.com/p/my-llm-coding-workflow-going-into" rel="noopener noreferrer">Link</a></p> <h3> <strong>How to become a more effective engineer</strong> </h3> <p>Software engineers often complain about dysfunction but don't understand how their organizations actually work. Guest author Cindy Sridharan argues that career advancement requires mastering "soft skills" - which are actually hard skills.<br> Key tactics are learn your org's implicit hierarchies and who holds real influence. Understand if your culture is top-down, bottom-up, or both. Get comfortable with messy codebases and organizations instead of demanding perfection. Chase small wins to build credibility early.<br> The biggest mistake? Engineers staying frustrated without learning how decisions get made, what gets rewarded, and who to influence. Most tech companies are far messier inside than they appear - there's no "right way," only understanding tradeoffs and navigating reality.<br> Bottom line would be technical excellence alone won't make you effective. Understanding organizational dynamics is non-negotiable for impact.</p> <p><a href="https://newsletter.pragmaticengineer.com/p/how-to-become-a-more-effective-engineer" rel="noopener noreferrer">Link</a></p> <h3> <strong>The 2025 Cloudflare Radar Year in Review</strong> </h3> <p>The internet in 2025 didn’t just grow — it got faster, stranger, and deeply AI-driven.<br> Global traffic increased 19%, but bots now consume a meaningful share of the web. AI crawlers made up 4.2% of all HTML requests, while googlebot alone accounted for 4.5%, briefly spiking to 11% in late April. “User-action” crawling — bots loading pages because an AI was asked a question — surged more than 15×, led by chatgpt-user acting as a super-user on behalf of millions.<br> Crypto infrastructure quietly crossed a milestone. For the first time, most human web traffic became post-quantum ready, with 52% protected by quantum-resilient TLS, up from 29% earlier this year — driven largely by Apple enabling hybrid quantum-secure key exchange by default in iOS 26.<br> Security remained rough. About 6% of traffic across Cloudflare’s network was mitigated as potentially malicious, hyper-volumetric DDoS attacks hit new records, and over 5% of scanned email was malicious, dominated by phishing and identity impersonation.<br> In the AI platform race, Meta’s llama-3-8b-instruct became the most-used model on Cloudflare Workers AI, while plain text generation remained the dominant use case at the edge.</p> <p><a href="https://blog.cloudflare.com/radar-2025-year-in-review/" rel="noopener noreferrer">Link</a></p> <h3> <strong>AI-driven coding is making code cheap; winners will fix bottlenecks in review, testing, and release to ship value faster.</strong> </h3> <p>Paul Dix argues 2026 will mark a split between teams that modernize their delivery pipelines for AI-generated code and those that stall. Coding speed has surged thanks to agentic tools, but throughput is now constrained by requirements clarity, reviews, validation, deployment safety, and operations.<br> Using Amdahl’s Law, he notes speeding coding alone yields modest end-to-end gains if it’s a small slice of the workflow. To capture 10x improvements, organizations must raise ceilings across review bandwidth and ownership, testing and validation, release/rollback confidence, security and compliance gates, and product decision latency.<br> He expects most new code to be AI-generated, with humans directing intent and verifying outcomes. Startups without heavy process gates already show tighter loops when they optimize the entire pipeline. He recommends making pipelines agent-accessible: fast, deterministic tests; one-command dev environments; crisp module boundaries; colocated docs; golden datasets and performance harnesses; and agent-accessible observability across CI, performance, deployment, and production.<br> The takeaway: optimize non-code chokepoints and tooling so agents can work end-to-end; that’s where the productivity divergence will come from.</p> <p><a href="https://x.com/pauldix/status/2006423514446749965" rel="noopener noreferrer">Link</a></p> <h3> <strong>Facilitating AI Adoption at Imprint: Practical Playbook, Not Hype</strong> </h3> <p>Will Larson, engineering leader at Imprint, shares hard-won lessons from 18 months of internal AI adoption. His key insight? Real adoption isn't about picking tools—it's about deep partnership and removing obstacles.<br> Imprint's approach centers on three pillars: pave the path by removing barriers to access, recognize that opportunity exists everywhere across all teams, and have senior leadership lead from the front with hands-on work.<br> The biggest surprise? Successful agent deployment requires old-fashioned product engineering, not just platform building. Work closely with domain experts on high-impact workflows, make solutions extensible, and monitor adoption as your signal for problem-solution fit.<br> Critical detail: centralizing prompts in a shared database made them discoverable, improvable, and reusable across teams. And yes, mundane IT issues like account provisioning and access controls still make or break adoption.<br> Bottom line: We're early in AI adoption—focus on rate of learning over everything else.</p> <p><a href="https://lethain.com/company-ai-adoption/" rel="noopener noreferrer">Link</a></p> aiengineering llmcoding engineeringleadership devops Madhu Sudhan Subedi Sun, 28 Dec 2025 16:05:31 +0000 https://dev.to/weekly/ai-open-source-pay-gaps-and-the-future-of-software-power-39e0 https://dev.to/weekly/ai-open-source-pay-gaps-and-the-future-of-software-power-39e0 <h2> 🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/41" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </h2> <h3> <strong>The Bubble Is Labor: AI as Capital’s Path to Doing the Work Itself</strong> </h3> <p>companies don’t hire people because they want to. They hire people because founders can’t do all the work themselves. If they could—if they had unlimited hands and brains—the ideal number of employees would be zero.<br> That leads to a scary realization. No one is actually responsible for creating jobs. Jobs exist only because owners need help.<br> Now enter AI. This isn’t companies “replacing workers.” It’s companies finally being able to do the work themselves. When global knowledge work costs tens of trillions of dollars every year, investing heavily in AI suddenly makes perfect sense.<br> This reframes everything. AI isn’t breaking a stable system—it’s accelerating an inevitable one. Capital has always tried to reduce its need for labor. AI is just different because it replaces intelligence, not tasks.<br> It’s unsettling. But maybe it’s not evil—just new physics.</p> <p><a href="https://danielmiessler.com/blog/the-real-bubble-is-human-labor" rel="noopener noreferrer">Link</a></p> <h3> <strong>The Entrepreneurial Software Engineer’s Path To Business Leadership</strong> </h3> <p>Software engineers create enormous value because code has near-zero marginal cost and massive leverage—one engineer can impact millions of users. That’s why demand and salaries are rising sharply. But to maximize their impact, engineers must go beyond writing code and develop product, business, and leadership skills. Communication, writing, and mentoring help engineers scale their influence through people, not just systems. Understanding business goals, product strategy, and design enables better decisions and greater ownership. Finally, building strong relationships within teams, companies, and the wider industry turns engineers into intrapreneurs—people who create outsized value from within organizations or launch successful ventures of their own.</p> <p><a href="https://www.forbes.com/councils/forbesbusinesscouncil/2023/12/22/the-entrepreneurial-software-engineers-path-to-business-leadership/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Confronting Career Inequalities: Respect, Timing, and the Pay Gap in Tech</strong> </h3> <p>Is compensation really about merit, or mostly about timing and context? Two leadership paths—a COO underpaid relative to a VP—showing how early entry and organizational timing shape titles, scope, and pay. Respect also diverges: some leaders fight for basic authority while others are granted autonomy, underscoring that job satisfaction depends as much on how work is treated as on role or skill.<br> The broader point is uncomfortable: tech’s revenue-per-employee dwarfs other industries (Apple’s RPE vs. ~$170K averages), so pay concentrates where output scales. That leaves many high performers elsewhere working longer, harder hours for far less, and fuels imposter syndrome in tech—even among capable leaders who doubt whether their pay reflects true value. The takeaway for managers: acknowledge the structural gap, budget with empathy, and anchor compensation decisions in impact and respect, not just optics or tenure. Owning these realities—timing advantages, organizational politics, and tech’s outsized economics—helps teams navigate fairness with clarity rather than guilt.</p> <p><a href="https://playfulprogramming.com/posts/career-inequalities" rel="noopener noreferrer">Link</a></p> <h3> <strong>“Source Available” Isn’t Open Source</strong> </h3> <p>Is “open source” just “source is open”? Dries Buytaert critiques David Heinemeier Hansson’s launch of Fizzy under the O’Saasy license, which blocks competing SaaS and therefore violates the Open Source Initiative (OSI) definition. Matt Mullenweg pushes back, and Dries argues the term has a specific, community-built meaning that shouldn’t be repurposed for marketing. Calling Fizzy “source available” is more accurate: you can read, run, and modify the code, but SaaS rights are reserved.<br> The deeper issue isn’t semantics—it’s sustainability. Many companies profit from open source while leaving maintenance to others, creating a “free rider” problem that projects like Drupal and WordPress have wrestled with for decades. Dries sees value in experimenting with licenses that discourage customer free-riding, and says O’Saasy fits that mold—but it’s still not open source. The takeaway for developers and leaders: protect the integrity of the OSI definition while tackling sustainability head-on through contribution credits, incentives, or policy. Dries reframes the debate: definitions keep the commons coherent; sustainability keeps it alive.</p> <p><a href="https://dri.es/source-available-is-not-open-source-and-that-is-okay" rel="noopener noreferrer">Link</a></p> <h3> <strong>Opinionated Agents Beat “General Purpose” Every Time</strong> </h3> <p>Should agent products offer more knobs or better outcomes? The case for opinionated design: pick a narrow task, do lots of work on the user’s behalf, and encode hard-won team knowledge into prompts, tools, and defaults. The “flexibility trap”—exposing temperature, chunking, and model pickers—pushes complexity onto users. Instead, dogfood relentlessly, evaluate on real tasks (not benchmarks), and prune configuration until the baseline agent works reliably.<br> The surprising point: models are non‑fungible in their harness. Intelligence is spiky, so a model swap can break tuned prompts and tool use; only the harness+model pair that succeeds on your task matters. The practical playbook is deep and narrow—optimize ruthlessly for the 10% of workflows that drive most value, avoid “wide” agents that try everything and “shallow” agents that shouldn’t be agents at all. Implication for builders: audit configs, hardcode good defaults, and specialize by domain (as seen with Cursor, Claude Code, DeepAgents, Amp Code). Opinionated choices create dependable agents; options can come later once the core experience is solid.</p> <p><a href="https://www.vtrivedy.com/posts/agents-should-be-more-opinionated" rel="noopener noreferrer">Link</a></p> aishift techpower codeleverage futureofwork Madhu Sudhan Subedi Sun, 21 Dec 2025 15:53:46 +0000 https://dev.to/weekly/weekly-51-2025-ai-coding-agents-and-engineering-culture-01x-engineers-5bnj https://dev.to/weekly/weekly-51-2025-ai-coding-agents-and-engineering-culture-01x-engineers-5bnj <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/40" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>How Good Engineers write Bad Code at Big Companies</strong> </h3> <p>Why does sloppy code emerge from teams packed with strong engineers? It’s structural: short tenures, frequent reorgs, and internal mobility mean most changes are made by “beginners” to a codebase or language. A few “old hands” carry deep knowledge, but their review bandwidth is limited and informal. The median productive engineer is competent but racing deadlines in unfamiliar systems—so hacky fixes pass, get lightly reviewed, and ossify.<br> The surprising claim: big tech accepts this trade. Leadership prioritizes organizational legibility and rapid redeployment over long-term codebase expertise, especially in an era where pivoting to AI matters more than pristine internals. The implication for developers is sobering—individuals can’t fix the system. The realistic move is to become an “old hand” where possible and block the worst changes, while recognizing that in impure, deadline-driven engineering, some bad code is inevitable.</p> <p><a href="https://www.seangoedecke.com/bad-code-at-big-companies/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Compound Engineering: When Agents Write All Your Code</strong> </h3> <p>What happens when 100% of your code is written by AI agents? Every outlines a four-step loop—Plan, Work, Assess, Compound—that turns agent output into a learning system. The twist isn’t automation alone; it’s compounding. Each bug, test failure, and design insight gets documented and reused by future agents, so every feature makes the next one easier to build.<br> Is manual testing and verbose documentation becoming optional? In this workflow, a developer spends most time on planning and review, while agents write and self-check code using tools like linters, unit tests, and parallel code-review subagents. The “Compound” step stores lessons inside prompts or plugins, spreading knowledge across the team—even new hires get guardrails instantly. Every claims a single developer can now handle the workload of five, and they’re shipping multiple production products this way.<br> The takeaway for engineering leaders: optimize for planning quality, agent orchestration, and learnings capture. Expect faster iteration, fewer repeated mistakes, and easier replatforming. As models improve, small projects need less upfront planning, but complex systems still benefit from strong architectural guidance—and a feedback loop that never forgets.</p> <p><a href="https://every.to/chain-of-thought/compound-engineering-how-every-codes-with-agents" rel="noopener noreferrer">Link</a></p> <h3> <strong>Why Write Engineering Blogs: Career Signal, Community, and Clear Thinking</strong> </h3> <p>Why do many engineers keep blogging after the initial hype fades? Some started to build visibility or share a product; others simply wanted to teach, document hard-won lessons, or make sense of complex systems. A recurring theme is permanence and impact: structured writing creates a public artifact that outlives chat, helps others solve real problems, and quietly advocates for its author.<br> Is blogging still worth it in the post‑social era? The interviewees say yes—because it blends personal learning with practical reach. Blogs become living notebooks for techniques, design decisions, and deep dives that don’t fit official docs. They can seed communities around tools, sharpen thinking through feedback, and serve as durable career signals. The takeaway for engineers and teams: treat blogging as part of engineering practice—capture what you learn, go deep on specifics, update mistakes, and let well-written posts compound into reputation, collaboration, and better work. In this post, the editors frame blogging as both craft and leverage for developers and organizations.</p> <p><a href="https://writethatblog.substack.com/p/why-write-engineering-blogs" rel="noopener noreferrer">Link</a></p> <h3> <strong>Working with Q: A Defensive Protocol for Coding Agents</strong> </h3> <p>How should an AI coding agent think when mistakes compound and can brick a project? I found github gist notes lays out a clear, testable protocol for “defensive epistemology” in software work: make explicit predictions before every action, compare outcomes after, and stop to update the model whenever reality surprises you. The core rule is blunt: reality doesn’t care about your mental model; all failures live in that gap. By writing DOING/EXPECT/IF YES/IF NO before tool calls and RESULT/MATCHES/THEREFORE after, agents and humans expose reasoning, catch wrong assumptions early, and prevent cascading errors.<br> The piece doubles down on disciplined loops: say “oops” and log raw failures, maintain multiple hypotheses, checkpoint after small batches, separate facts from theories, and avoid silent fallbacks that hide hard failures. It also sets boundaries for autonomy—surface contradictions, push back with evidence, and pause before irreversible changes. The takeaway for developers and teams: treat agent workflows like science. Slow is smooth, smooth is fast. The author offers a practical operating manual for coding agents that prioritizes verified reality over confident guesses, turning debugging into a transparent, repeatable process.</p> <p><a href="https://gist.github.com/Web-Dev-Codi/29b25c56a0f1e2535d93614eab67b4d1" rel="noopener noreferrer">Link</a></p> <h3> <strong>The Rise of the 0.1x Engineer: Curators in the Age of Coding Assistants</strong> </h3> <p>Are “10x engineers” still the goal when AI can spray code into any codebase? I am reading a post that argues the real leverage now comes from “0.1x engineers” — the people who resist prompting first and instead set patterns, prune cruft, and keep systems coherent. As coding assistants make it trivial to add code, they also make it trivial to add bloat, spaghetti structure, and LLM leftovers.<br> The interesting shift: productivity isn’t measured by lines of code but by reuse, structure, and deletion. Teams will need engineers who can distinguish dead code from generated detritus, define reusable blocks, and group files intelligently. The takeaway for engineering leaders and ICs: optimize for maintainers and curators. In this post, the author suggests the next elite skill is designing clean interfaces, enforcing patterns, and saying “no” to unnecessary code — because in an AI‑augmented world, discipline beats speed.</p> <p><a href="https://www.jerpint.io/blog/2025-12-08-01x-engineer/" rel="noopener noreferrer">Link</a></p> aicodingagent softwareengineering developerproductivity codequality Madhu Sudhan Subedi Sun, 21 Dec 2025 15:23:31 +0000 https://dev.to/weekly/weekly-50-2025-anthropics-bun-bet-the-pm-drought-seattles-ai-backlash-4l8p https://dev.to/weekly/weekly-50-2025-anthropics-bun-bet-the-pm-drought-seattles-ai-backlash-4l8p <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/39" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>Bun JavaScript runtime acquired by Anthropic</strong> </h3> <p>Anthropic has snapped up the Bun JavaScript runtime in a move that could reshape how AI writes and ships code, elevating Bun from fast new kid on the block to “essential infrastructure” behind tools like Claude Code’s native installer, all while staying open source and MIT-licensed with its turbocharged bundle of runtime, package manager, bundler, test runner, and single-file executable builds. Under the hood, Bun’s unconventional stack—JavaScriptCore instead of V8, and cutting-edge Zig for native code—gives it a distinctive technical edge and a bit of a rebel identity in the JS world. Creator Jarred Sumner, who helped raise about 26 million dollars for Bun while it brought in essentially zero revenue, now sees Anthropic as the long-term home that can fund aggressive development just as AI agents begin to write, test, and deploy more of our software. But the plot twist is the tension ahead: Zig’s inventor enforces a strict “no AI/LLM” contributor policy even as an AI company leans heavily on the language, and developers are split between excitement that Anthropic might supercharge Bun into the default JS/TS runtime for AI-era apps and anxiety that today’s free, community-driven project could slowly tilt toward paid features or a strategic pivot decided in boardrooms, not GitHub issues.</p> <p><a href="https://devclass.com/2025/12/03/bun-javascript-runtime-acquired-by-anthropic-tying-its-future-to-ai-coding/?ref=dailydev" rel="noopener noreferrer">Link</a></p> <h3> <strong>Where Do You Go If You Don’t Care About Growth?</strong> </h3> <p>What if climbing the ladder isn’t the goal? A junior developer questions the default path of “grow, get promoted, repeat,” arguing that many roles offer little real upside for ICs (Individual Contributors) while reliably enriching management. The tension is clear: performance is judged by opaque criteria, “training” often means conformity, and even unglamorous roles demand 5+ years’ experience to fix legacy messes.<br> Is there room for software careers centered on stability and purpose over status? Small, non‑growth companies, personal projects, and open source as more meaningful tracks—work that’s maintainable, helps users, and doesn’t exist to maximize someone else’s return. The takeaway for teams and hiring managers: not everyone optimizes for speed and scale. There’s a rising cohort that values fair pay, sane scope, and autonomy over bonuses and titles. The question isn’t how to accelerate growth; it’s whether the industry can make space for engineers who choose craft and contribution over corporate ambition.</p> <p><a href="https://ramones.dev/posts/where-are-you-supposed-to-go/" rel="noopener noreferrer">Link</a></p> <h3> <strong>AI Companies Are Hiring Fewer PMs: 34% Lower Share of Open Roles</strong> </h3> <p>Are product roles shrinking as LLMs reshape how teams build? Riso Group finds that AI companies list 34% fewer Product Manager titles as a share of openings compared to other sectors. PM roles account for 2.3% of AI postings vs 3.2–3.8% across DevTools, Consumer, Data, B2B, and Fintech, based on 8,803 deduplicated job titles scraped from 100 tech firms. The analysis is directional and excludes big tech, but the gap is clear.<br> The implication is that AI-first orgs may be shifting headcount toward model building, infra, and data roles while expecting existing teams to absorb more product responsibilities. PM scope could trend more technical and system-oriented, with fewer pure orchestration roles. Let’s see how these percentages evolve as agentic workflows mature and “builder PM” expectations rise.</p> <p><a href="https://risogroup.co/insights/ai-companies-hiring-one-third-fewer-pms" rel="noopener noreferrer">Link</a></p> <h3> <strong>Everyone in Seattle Hates AI</strong> </h3> <p>Why are many Seattle engineers hates AI right now? This essay points to a cultural shift inside big tech—especially Microsoft—where layoffs, forced adoption of underperforming Copilot tools, and “AI or nothing” org politics have bred resentment. Engineers were told projects failed because teams didn’t “embrace AI,” even while being required to use tools that often made work slower and worse. The result: top talent rebranded as “not AI,” stuck watching compensation stall and autonomy vanish.<br> The surprising twist is how this environment shapes beliefs. Engineers start to think AI is useless and they aren’t qualified to work on it anyway. That self-limiting loop hurts companies (innovation gets centralized and stifled), employees (careers stagnate), and local builders (anything labeled “AI” triggers scorn). Compare with San Francisco, where curiosity and permission to build still exist; the essay argues that belief in one’s agency is now a competitive advantage. The takeaway for leaders: stop using AI as a political shield, empower teams to ship and evaluate tools honestly, and rebuild trust before cynicism hardens into permanent resistance.</p> <p><a href="https://jonready.com/blog/posts/everyone-in-seattle-hates-ai.html" rel="noopener noreferrer">Link</a></p> <h3> <strong>Treat Test Code Like Production Code</strong> </h3> <p>Is test code getting a pass it shouldn’t? Mark Seemann in his article argues that the same standards applied to production code—readability, DRY, refactoring, reviews—should apply to tests. Sloppy patterns like copy‑paste suites, commented‑out “zombie” code, arbitrary waits, and blocking async calls make maintenance harder and slow teams down. The point of good code isn’t the computer; it’s future humans who have to read and change it.<br> What’s the practical guidance? Use test‑specific best practices (think xUnit Test Patterns), write descriptive tests without sacrificing DRY, and avoid duplication that leads to “shotgun surgery” when behavior changes. There are narrow dispensations: test code can relax some security and platform rules when it never ships (e.g., hardcoded test credentials, skipping certain async context rules in .NET, allowing orphan instances in Haskell). The takeaway for teams: treat test code as a first‑class citizen—it’s part of the system’s maintainability, not a temporary scaffold.</p> <p><a href="https://blog.ploeh.dk/2025/12/01/treat-test-code-like-production-code/" rel="noopener noreferrer">Link</a></p> bunruntime productmanagement aiadoption softwareengineering Madhu Sudhan Subedi Sun, 14 Dec 2025 08:49:53 +0000 https://dev.to/weekly/weekly-50-2025-anthropics-bun-bet-the-pm-drought-seattles-ai-backlash-1okk https://dev.to/weekly/weekly-50-2025-anthropics-bun-bet-the-pm-drought-seattles-ai-backlash-1okk <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/39" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>Bun JavaScript runtime acquired by Anthropic</strong> </h3> <p>Anthropic has snapped up the Bun JavaScript runtime in a move that could reshape how AI writes and ships code, elevating Bun from fast new kid on the block to “essential infrastructure” behind tools like Claude Code’s native installer, all while staying open source and MIT-licensed with its turbocharged bundle of runtime, package manager, bundler, test runner, and single-file executable builds. Under the hood, Bun’s unconventional stack—JavaScriptCore instead of V8, and cutting-edge Zig for native code—gives it a distinctive technical edge and a bit of a rebel identity in the JS world. Creator Jarred Sumner, who helped raise about 26 million dollars for Bun while it brought in essentially zero revenue, now sees Anthropic as the long-term home that can fund aggressive development just as AI agents begin to write, test, and deploy more of our software. But the plot twist is the tension ahead: Zig’s inventor enforces a strict “no AI/LLM” contributor policy even as an AI company leans heavily on the language, and developers are split between excitement that Anthropic might supercharge Bun into the default JS/TS runtime for AI-era apps and anxiety that today’s free, community-driven project could slowly tilt toward paid features or a strategic pivot decided in boardrooms, not GitHub issues.</p> <p><a href="https://devclass.com/2025/12/03/bun-javascript-runtime-acquired-by-anthropic-tying-its-future-to-ai-coding/?ref=dailydev" rel="noopener noreferrer">Link</a></p> <h3> <strong>Where Do You Go If You Don’t Care About Growth?</strong> </h3> <p>What if climbing the ladder isn’t the goal? A junior developer questions the default path of “grow, get promoted, repeat,” arguing that many roles offer little real upside for ICs (Individual Contributors) while reliably enriching management. The tension is clear: performance is judged by opaque criteria, “training” often means conformity, and even unglamorous roles demand 5+ years’ experience to fix legacy messes.<br> Is there room for software careers centered on stability and purpose over status? Small, non‑growth companies, personal projects, and open source as more meaningful tracks—work that’s maintainable, helps users, and doesn’t exist to maximize someone else’s return. The takeaway for teams and hiring managers: not everyone optimizes for speed and scale. There’s a rising cohort that values fair pay, sane scope, and autonomy over bonuses and titles. The question isn’t how to accelerate growth; it’s whether the industry can make space for engineers who choose craft and contribution over corporate ambition.</p> <p><a href="https://ramones.dev/posts/where-are-you-supposed-to-go/" rel="noopener noreferrer">Link</a></p> <h3> <strong>AI Companies Are Hiring Fewer PMs: 34% Lower Share of Open Roles</strong> </h3> <p>Are product roles shrinking as LLMs reshape how teams build? Riso Group finds that AI companies list 34% fewer Product Manager titles as a share of openings compared to other sectors. PM roles account for 2.3% of AI postings vs 3.2–3.8% across DevTools, Consumer, Data, B2B, and Fintech, based on 8,803 deduplicated job titles scraped from 100 tech firms. The analysis is directional and excludes big tech, but the gap is clear.<br> The implication is that AI-first orgs may be shifting headcount toward model building, infra, and data roles while expecting existing teams to absorb more product responsibilities. PM scope could trend more technical and system-oriented, with fewer pure orchestration roles. Let’s see how these percentages evolve as agentic workflows mature and “builder PM” expectations rise.</p> <p><a href="https://risogroup.co/insights/ai-companies-hiring-one-third-fewer-pms" rel="noopener noreferrer">Link</a></p> <h3> <strong>Everyone in Seattle Hates AI</strong> </h3> <p>Why are many Seattle engineers hates AI right now? This essay points to a cultural shift inside big tech—especially Microsoft—where layoffs, forced adoption of underperforming Copilot tools, and “AI or nothing” org politics have bred resentment. Engineers were told projects failed because teams didn’t “embrace AI,” even while being required to use tools that often made work slower and worse. The result: top talent rebranded as “not AI,” stuck watching compensation stall and autonomy vanish.<br> The surprising twist is how this environment shapes beliefs. Engineers start to think AI is useless and they aren’t qualified to work on it anyway. That self-limiting loop hurts companies (innovation gets centralized and stifled), employees (careers stagnate), and local builders (anything labeled “AI” triggers scorn). Compare with San Francisco, where curiosity and permission to build still exist; the essay argues that belief in one’s agency is now a competitive advantage. The takeaway for leaders: stop using AI as a political shield, empower teams to ship and evaluate tools honestly, and rebuild trust before cynicism hardens into permanent resistance.</p> <p><a href="https://jonready.com/blog/posts/everyone-in-seattle-hates-ai.html" rel="noopener noreferrer">Link</a></p> <h3> <strong>Treat Test Code Like Production Code</strong> </h3> <p>Is test code getting a pass it shouldn’t? Mark Seemann in his article argues that the same standards applied to production code—readability, DRY, refactoring, reviews—should apply to tests. Sloppy patterns like copy‑paste suites, commented‑out “zombie” code, arbitrary waits, and blocking async calls make maintenance harder and slow teams down. The point of good code isn’t the computer; it’s future humans who have to read and change it.<br> What’s the practical guidance? Use test‑specific best practices (think xUnit Test Patterns), write descriptive tests without sacrificing DRY, and avoid duplication that leads to “shotgun surgery” when behavior changes. There are narrow dispensations: test code can relax some security and platform rules when it never ships (e.g., hardcoded test credentials, skipping certain async context rules in .NET, allowing orphan instances in Haskell). The takeaway for teams: treat test code as a first‑class citizen—it’s part of the system’s maintainability, not a temporary scaffold.</p> <p><a href="https://blog.ploeh.dk/2025/12/01/treat-test-code-like-production-code/" rel="noopener noreferrer">Link</a></p> bunruntime productmanagement aiadoption softwareengineering Madhu Sudhan Subedi Sun, 07 Dec 2025 03:17:13 +0000 https://dev.to/weekly/weekly-49-2025-ai-breakthroughs-critical-react-flaw-and-shifting-tech-culture-3o91 https://dev.to/weekly/weekly-49-2025-ai-breakthroughs-critical-react-flaw-and-shifting-tech-culture-3o91 <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/38" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>React Server Components Hit with Critical RCE Vulnerability</strong> </h3> <p>A serious security vulnerability has been disclosed in React Server Components, enabling unauthenticated remote code execution via malformed requests to React Server Function endpoints. The issue, tracked as CVE-2025-55182 with a maximum CVSS score of 10.0, affects React versions 19.0, 19.1.0, 19.1.1, and 19.2.0. Notably, apps may be vulnerable even if they don’t explicitly use Server Function endpoints, as long as they support React Server Components.<br> What should teams do now? Upgrade immediately to patched React releases—19.0.1, 19.1.2, or 19.2.1—and apply framework-specific updates. Next.js users should move to the latest patches in their line (or downgrade from recent canaries to stable 14.x), while projects using React Router’s unstable RSC APIs, Expo, Waku, Parcel/Vite RSC plugins, Redwood SDK, Turbopack, or Webpack-based RSC modules should bump to the latest versions listed. Hosting mitigations exist but aren’t a substitute for updating.<br> Why this matters: RCE in a widely used server rendering pipeline is rare and high-impact. Any team experimenting with or deploying RSC features should treat this as a priority, review their server exposure, and ensure dependencies align with patched React drops. Expect more detail on the root cause after the fix rollout completes.</p> <p><a href="https://react.dev/blog/2025/12/03/critical-security-vulnerability-in-react-server-components" rel="noopener noreferrer">Link</a></p> <h3> <strong>AWS re:Invent 2025: Frontier Agents, Nova 2, and Trainium3 Raise the Bar</strong> </h3> <p>How far can autonomous agents go? AWS unveiled “frontier agents” that run for hours or days without human input, including Kiro as a virtual developer, a Security Agent, and a DevOps Agent—paired with new Bedrock AgentCore capabilities for policy, evaluations, and episodic memory to make production deployments safer and measurable. Amazon Connect added agentic voice and assistance features with observability, hinting at agents becoming first‑class citizens in customer operations. For builders, Strands Agents arrived in TypeScript (preview) with edge support, and Bedrock added 18 open‑weight models alongside the Nova 2 family and Nova Forge “open training,” pushing model choice and customization forward.<br> Will specialized hardware finally tame AI training costs? Trainium3 UltraServers (3nm chips, up to 144 per system) promise up to 4.4x compute vs. prior gen and major efficiency gains, while “checkpointless training” on SageMaker HyperPod aims for near‑instant recovery from faults at scale. On the data side, S3 Vectors is now GA with up to two billion vectors per index and big cost reductions, S3 object limits jumped to 50TB, and S3 Batch Ops got 10x faster—signal that AWS is hardening the data plumbing for RAG and vector-heavy apps. CloudWatch introduced a unified store for ops/security/compliance logs; GuardDuty expanded to EC2/ECS with ATT&amp;CK‑mapped incident timelines; Security Hub is GA with near‑real‑time correlation.<br> Key implications: agents are moving from demos to durable, observable systems; model customization and hardware throughput are becoming table stakes; and vector-native storage plus unified telemetry will shape production AI stacks. Developers should watch AgentCore’s policy/eval primitives, Trainium3 economics, and S3 Vectors integrations with Bedrock Knowledge Bases and OpenSearch to streamline agentic workflows end‑to‑end.</p> <p><a href="https://www.aboutamazon.com/news/aws/aws-re-invent-2025-ai-news-updates" rel="noopener noreferrer">Link</a></p> <h3> <strong>The Junior Hiring Crisis: AI Is Eroding the Apprenticeship Ladder</strong> </h3> <p>Are AI‑adopting companies quietly closing the door on new grads? Recent data from Stanford’s Digital Economy Lab shows junior hiring down 13% at firms embracing AI, while Harvard finds higher unemployment for 22–25‑year‑olds even as senior hiring stays steady. The article argues this isn’t about talent quality—it’s about an apprenticeship breakdown: AI automates entry‑level tasks, and a decade of “I’m an IC, not a manager” culture has normalized seniors opting out of mentoring. Add misaligned incentives (quarterly pressure, short tenures) and companies default to hiring seniors, starving the pipeline.<br> The long‑term risk is a timing mismatch. In 10–20 years, who becomes the architects with the judgment to design complex systems if the ladder is missing today? The near‑term takeaway flips to individual agency: build the skills AI can’t automate—relationship intelligence, influence, and collaboration. Students and early‑career folks should identify 10–30 key relationships (guides, align, partners, network), nurture them intentionally, and practice while stakes are low. For teams and universities, treat mentoring and relational skills as core competencies: senior engineers teach to sharpen clarity, and programs embed networking and onboarding fundamentals into the curriculum.</p> <p><a href="https://people-work.io/blog/junior-hiring-crisis/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Instagram Orders Full Return to Office in 2026 ending WFH</strong> </h3> <p>Instagram will require US employees with assigned desks to work in-office five days a week starting February 2, 2026, according to an internal memo from Adam Mosseri. The memo frames the change as necessary to “evolve,” notes that 2026 “is going to be tough,” and offers limited flexibility to work remotely “when you need to,” leaving judgment calls to staff. This marks a shift from Meta’s broader three‑days‑in policy and runs counter to the hybrid norms many tech workers have settled into since the pandemic.<br> What does this signal for tech culture and productivity? Mosseri also outlined moves to “be more nimble,” including canceling recurring meetings every six months unless deemed essential, favoring product prototypes over decks, and speeding up decision‑making. The implications: expect tighter in‑person coordination, fewer standing meetings, and pressure to show faster shipping. The open question is whether a strict RTO improves execution or sparks pushback in a market where talent still values flexibility.</p> <p><a href="https://www.engadget.com/social-media/instagram-mandates-total-return-to-office-for-employees-in-2026-212738225.html" rel="noopener noreferrer">Link</a></p> <h3> <strong>Failed Software Projects Are Strategy Problems, Not Coding Problems</strong> </h3> <p>Why do so many software projects crater even when the tech stack looks fine on paper? This article argues the root cause is almost always strategic, not tactical. A Delphi-based acoustics tool succeeded globally despite dated code because the team had clear goals, strong domain expertise, and solid operations. In contrast, Australia’s Bureau of Meteorology spent $96M rebuilding its site on Drupal while leaving core needs—like TLS everywhere—unfixed, revealing fuzzy objectives, weak in‑house capability, and poor vendor choices.<br> The takeaway for leaders and engineers: define what success looks like, understand your constraints, and build or hire the right domain expertise before writing code. A practical lens from Clausewitz helps—identify your “centre of gravity,” the key barrier that prevents the desired end state, then design projects to remove it. With strategic clarity, tactical missteps get corrected quickly; without it, even competent developers can’t save a misaligned plan.</p> <p><a href="https://deadsimpletech.com/blog/failed_software_projects" rel="noopener noreferrer">Link</a></p> aitrends securityalert cloudcomputing techculture Madhu Sudhan Subedi Sun, 30 Nov 2025 14:53:45 +0000 https://dev.to/weekly/weekly-48-2025-ai-enterprise-knowledge-and-the-future-of-engineering-eg8 https://dev.to/weekly/weekly-48-2025-ai-enterprise-knowledge-and-the-future-of-engineering-eg8 <p>🎙️ <strong><a href="https://madhusudhansubedi.com.np/weekly/37" rel="noopener noreferrer">Click here to listen on Madhu Sudhan Subedi Tech Weekly →</a></strong> </p> <h3> <strong>Stack Overflow Introduces Stack Internal</strong> </h3> <p>Stack Overflow has reintroduced its enterprise knowledge platform as Stack Internal, marking its next step toward becoming the most trusted source for technologists. Designed for modern engineering teams, Stack Internal blends human insight with AI automation to keep enterprise knowledge accurate, secure, and always up to date.<br> The platform shifts the burden from entirely human effort to a human + AI partnership, helping developers focus on building products while reducing risk and improving productivity. New features include Knowledge Ingestion, which automatically pulls high-value information from tools like Microsoft Teams and Confluence, then uses AI scoring plus human review to produce verified, structured knowledge.<br> Another major addition is the MCP Server, a secure layer that connects tools like GitHub Copilot and ChatGPT to an organization’s validated knowledge, reducing hallucinations and improving AI accuracy. Stack Internal also now integrates with Microsoft 365 Copilot for trusted, bi-directional knowledge access.<br> Stack Overflow says this evolution creates a smarter, continuously learning ecosystem that supports real innovation across the enterprise.</p> <p><a href="https://www.devopsdigest.com/stack-overflow-introduces-stack-internal?utm_source=tldrdevops" rel="noopener noreferrer">Link</a></p> <h3> <strong>Linus Torvalds: Vibe coding is fine, but not for production</strong> </h3> <p>Linux creator Linus Torvalds recently shared his thoughts on AI in software development. He says he’s “fairly positive” about vibe coding, but only as a way for beginners to get into computing — not for production code, where maintenance would be a nightmare.<br> Torvalds explained that his role in the Linux kernel has shifted over the years. Once known for saying “no” to risky ideas, he now sometimes has to push maintainers to adopt changes like Rust, which is finally becoming a real part of the kernel.<br> On AI, he’s not worried about Nvidia’s dominance, and even credits the AI boom for making Nvidia a better Linux player. The real problem, he says, is AI crawlers disrupting kernel.org and generating fake bug reports.<br> Torvalds believes AI is just another tool — like compilers — that boosts productivity without replacing programmers. And if you disagree, you can email him… just don’t expect a reply.</p> <p><a href="https://www.theregister.com/2025/11/18/linus_torvalds_vibe_coding/?ref=dailydev" rel="noopener noreferrer">Link</a></p> <h3> <strong>LLM01:2025 Prompt Injection - OWASP Gen AI Security Project</strong> </h3> <p>Prompt Injection is a vulnerability where attackers manipulate an AI model just by crafting deceptive prompts, sometimes even hiding them in web pages, documents, or images the model reads. The result can be data leaks, biased or incorrect answers, or even unauthorized actions in connected systems. There are two main flavors: direct injection, where a user tells the model to ignore its rules, and indirect injection, where hidden instructions in external content silently hijack the response. To reduce risk, teams are tightening system prompts, validating outputs in code, isolating untrusted content, and adding human approval for high‑impact actions. Prompt injection is quickly becoming one of the key security topics for any serious AI deployment.</p> <p><a href="https://genai.owasp.org/llmrisk/llm01-prompt-injection/" rel="noopener noreferrer">Link</a></p> <h3> <strong>Cracking X’s Algorithm: How to Actually Grow Your Audience</strong> </h3> <p>What does it take to get noticed on X (formerly Twitter)? With the platform’s recommendation code now public, the rules are clearer: posts that spark real conversation, drive actions beyond likes, and show consistent engagement within a specific community perform best. Warming up by interacting with others, replying quickly to early comments, and varying your post formats all boost your relevance.<br> Posting too often or jumping between unrelated topics can actually hurt your reach. X groups users by interests, so sticking to a clear niche helps the algorithm understand and recommend you. Quick “social proof” from mutuals can unlock much wider visibility, while spammy or negative signals—like blocks or mutes—push your content down.<br> For creators and brands, the message is simple: build genuine connections, keep content focused and engaging, and remember that replies and bookmarks matter just as much as likes. The algorithm rewards real community participation, not gimmicks.<br> <a href="https://supabird.io/articles/how-to-grow-on-x-what-we-learned-from-their-algorithm-reveal" rel="noopener noreferrer">https://supabird.io/articles/how-to-grow-on-x-what-we-learned-from-their-algorithm-reveal</a></p> <h3> <strong>The Future of Engineering: Top Trends Shaping the Industry in 2025</strong> </h3> <p>Engineering in 2025 is being reshaped by the convergence of automation, AI, and robotics across every industry. Engineers are no longer just designing parts; they’re architecting fully automated systems that can sense, decide, and act on their own. Collaborative robots are moving out of cages and working side by side with humans on factory floors, while AI-driven simulation and digital twins let teams test thousands of design variations before building anything in the real world. At the same time, there’s a huge push for sustainable engineering: smarter energy use, less waste, and circular design from day one. For anyone in engineering, the big question this year is simple: how fast can you learn to design not just products, but intelligent, automated systems end to end?</p> <p><a href="https://www.automate.org/news/the-future-of-engineering-top-trends-shaping-the-industry-in-2025" rel="noopener noreferrer">Link</a></p> promptinjection enterpriseknowledge stackoverflow xalgorithm