Your production code is training AI models right now. Most engineers have no idea which tools are sending it where.   This is not about one company or one settings toggle. It is the default across the AI tooling stack.   Copilot sends code context to Azure for completions. Cursor uploads project files to cloud for indexing. LangChain traces log prompts, outputs, and API keys to LangSmith by default. Even CI/CD telemetry includes code snippets.   The pattern repeats everywhere: Opt everyone in. Bury the setting. Ship fast.   Engineers secure APIs with rate limits and auth tokens. Almost nobody audits what their own tools send upstream.   4 point audit every team should run this week:   1. List every AI tool touching your codebase. 2. Check each tool's telemetry and data sharing settings. 3. Run truffleHog on your repo: secrets in history leak first. 4. Add .cursorignore and .github/copilot to block sensitive paths.   The model is replaceable. Your proprietary code in someone else's training set is permanent. ___ ♻️ Repost if this helped a fellow engineer. 💬 Have you audited what data your AI coding tools send home?   Follow Mian Zubair for daily AI + system design breakdowns.   #SystemDesign #SoftwareEngineering #AIEngineering #DistributedSystems #BuildInPublic  

If you use AI daily, you're hemorrhaging context. Every time you close a chat window, your architecture debates and decisions vanish. I just published a technical teardown on MemPalace, an offline-first, RAG-killer memory architecture using a 30x compression dialect called AAAK. It achieved a 96.6% R@5 score on LongMemEval—running 100% locally. Huge props to the architects @bensig and @MillaJovovich for proving that the future of AI infrastructure doesn't need to be locked behind a cloud subscription. 📖 Read my full architectural teardown here: https://lnkd.in/drfXQHnf 💻 Star the repo: https://lnkd.in/dE7TR9s7 #AI #LLMs #OpenSource #LocalFirst #MachineLearning #Developers

Is the era of "all-you-can-eat" AI coding officially coming to an end? GitHub has announced a fundamental shift for Copilot, moving from its long-standing flat-rate subscription to a token-based consumption model starting June 2026. While the monthly fees remain nominally the same, they will now function as a pre-paid credit balance. This change is driven by the rise of "agentic" workflows—complex, multi-step autonomous tasks that consume significantly more compute power than simple autocomplete suggestions. For developers and enterprise leaders, this marks a transition from predictable seat-based expenses to variable operational costs. While basic code completions remain exempt, high-intensity tasks like architectural planning and deep-dive debugging will now require careful credit management. This shift reflects a broader market trend where AI is maturing from an experimental add-on into a metered utility, much like electricity or water. How will this change your team's approach to AI integration? Will "prompt optimization" become a financial necessity rather than just a technical skill? #GitHubCopilot #GenerativeAI #SoftwareDevelopment #TechTrends #CloudComputing Read more: https://lnkd.in/gbhtiATq

🚨 Your AI is writing code at lightning speed… But who’s checking it before it ships? That’s the real problem. Most teams today rely on AI to generate code faster than ever. But speed without control = silent bugs, broken logic, and risky commits. That’s where git-lrc by Hexmos changes the game 👇 👉 It hooks directly into your Git workflow 👉 Runs AI-powered reviews before every commit 👉 Catches things like: • Leaked credentials • Silent logic changes • Expensive cloud operations • Security risks Think of it as a “braking system” for AI-generated code 💡 No extra workflow 💡 No complex setup \(literally ~60 seconds\) 💡 Completely free, using your own API key And the best part? It builds a habit most developers skip: 👉 “Don’t commit code you haven’t reviewed.” In a world of AI-assisted development, reviewing isn’t optional anymore — it’s survival. If you’re building with AI tools like Copilot, Cursor, or Gemini… this is something you should definitely check out: 🔗 https://lnkd.in/gyK___6Z 💬 Curious — would you trust AI-generated code without a review step? #AI #SoftwareDevelopment #DevTools #Git #CodeReview #Programming #Developers #TechLeadership

🚀 Built an AI that understands any codebase in seconds One of the biggest problems in engineering teams: → “Understanding a new repo takes hours (sometimes days)” So I built CodeLens AI. 👉 Paste any GitHub repo 👉 Ask questions in plain English 👉 Get precise, context-aware answers with exact code references No more endless scrolling. No more guesswork. 💡 What makes it powerful: - Full repository ingestion + semantic search pipeline - Context-aware answers grounded in actual code (not hallucinations) - File-level traceability → know exactly where logic lives ⚡ Performance: - Handles multi-file repositories efficiently - Delivers answers in ~6–10 seconds per query 🧠 Why this matters: - Cuts code understanding time drastically - Speeds up developer onboarding - Strong use-case for dev tools / engineering teams 🛠 Tech Stack: Python • FastAPI • Docker AWS EC2 (deployed) • CI/CD with GitHub Actions ChromaDB • Sentence Transformers LLM orchestration via OpenRouter 🎥 Demo below 🔗 GitHub: https://lnkd.in/e2cZPCkY Still evolving towards production (better performance, hosted models, cleaner UX), but this version already feels powerful 🔥 #AI #LLM #DeveloperTools #AWS #DevOps #CI_CD #Startups #BuildInPublic #SoftwareEngineering

🚀 Built an AI-Powered Kubernetes Auto-Healing Agent Debugging failed pods in Kubernetes often involves manually checking logs, events, and configurations — which can be time-consuming and error-prone. To explore how AI can help here, I built a system that: 🔍 Scans the cluster for failing pods 🧠 Analyzes issues using a hybrid approach (rule-based + LLM) 📄 Uses pod status, events (describe), and logs for context ⚙️ Differentiates between infrastructure issues and application-level errors 🛠 Suggests (and in some cases applies) fixes intelligently 💡 Key scenarios handled: ImagePullBackOff (invalid image) CrashLoopBackOff (container crashes) Application errors (e.g., division by zero) Configuration issues (missing env variables) OOMKilled (memory limits exceeded) Pending pods (resource constraints) 🎯 What I focused on: Avoiding blind automation (no unnecessary restarts) Using deterministic rules for known failures Leveraging AI only when deeper reasoning is required Building a realistic DevOps/AIOps-style workflow 📽️ Demo video below 👇 🔗 GitHub: https://lnkd.in/gkJYi2SD This project helped me understand how AI can be integrated into Kubernetes troubleshooting and move towards more intelligent, context-aware operations. #Kubernetes #AI #DevOps #AIOps #Python #Cloud #Learning

🚀 This might be one of the most interesting AI projects trending right now — and it’s not coming from where you’d expect. Check out MemPalace 👉 https://lnkd.in/gNCCQDSQ Built by Milla Jovovich and Ben Sigman, this open-source project tackles a fundamental problem in AI systems: memory loss across sessions. 💡 What makes MemPalace different? 🧠 Uses the ancient “memory palace” technique — organizing data into spatial structures (rooms, halls, etc.) 📦 Stores everything (verbatim) instead of letting AI decide what to keep 🔍 Enables context-rich retrieval, not just keyword search 🔐 Runs locally — no APIs, no subscriptions, full data ownership ⚡ In just 24 hours, it crossed 10K+ GitHub stars, signaling massive developer curiosity. (CryptoMist) 🧠 Why this matters (especially for engineers & SREs): We’ve all felt this: Every AI session starts from scratch. Context is lost. Decisions are forgotten. MemPalace flips the model: 👉 Treat memory as a first-class system component, not an afterthought. This has huge implications for: Long-running AI agents Developer copilots Incident analysis & postmortems Knowledge retention in distributed teams ⚠️ Reality check: While benchmarks look impressive, real-world performance and scalability still need validation. 💭 My take: This is less about “another AI tool” and more about a shift in architecture thinking — from stateless prompts → persistent, structured memory systems. If this direction matures, it could redefine how we build: 👉 AI-native platforms 👉 Autonomous systems 👉 Engineering knowledge graphs Curious — would you plug something like this into your workflow or infra stack? #AI #OpenSource #LLM #Engineering #DevTools #SRE #MachineLearning #Innovation #memory

Future generations of developers are going to look at our StackOverflow search history and think we were absolute savages. This tweet is painfully accurate. For the last 15 years, the standard software engineering workflow was basically digital foraging. Think about the sheer trauma of debugging a cryptic AWS deployment error in 2019: 👉 Copy the error log. 👉 Paste it into Google. 👉 Scroll past 4 sponsored links and 3 SEO-stuffed medium articles that start with "What is Cloud Computing?" 👉 Open 15 different StackOverflow tabs. 👉 Finally find a GitHub issue from 2016 describing your exact problem... only to scroll to the bottom and see the original author closed it with: "nvm, figured it out." (With zero explanation). 😭 We were out here hunting and gathering code snippets with stone tools. Today? You dump the error, the config file, and the surrounding 50 lines of code into Cursor or Claude and say, "Why is this broken?" and it highlights the exact missing comma in 4 seconds. The era of "Googling for syntax" is over. The era of "AI Orchestration" is here. But here is the catch: You still need to know how to ask the right questions, provide the right context, and verify the AI isn't hallucinating. That is exactly why we built our new crash course at Prepflix. We got tired of the academic fluff and built a no-BS guide to the AI tools you actually need for a modern development role. We teach you how to stop searching like a digital caveman and start building like a 10x engineer. Ready to upgrade your workflow and stop digging through 100,000 blue links? 👇 Comment "AI WORKFLOW" below, and I will personally DM you the details to get access!

AI is breaking Github and forcing it to evolve. In Oct 2025, GitHub planned for 10× capacity. Four months later, that wasn’t enough. Now they’re designing for 30×. Why? “Agentic development workflows have accelerated sharply.” AI agents don’t behave like humans: • They write, review, and merge code nonstop • They hit every system at once, storage, merge queues, Actions, APIs • Workloads that used to be sequential are now parallel and constant The cracks are already showing: ↳ Apr 23: Merge queue regression corrupted 2,092 PRs ↳ Apr 27: Elasticsearch collapsed under bot-like load This isn’t just scale. It’s an architectural shift. GitHub is: • Moving critical paths out of its Ruby monolith into Go • Leaving custom data centers for Azure • Going multi-cloud The platform built for humans is being rebuilt for machines. If you’re building dev infrastructure, take note: AI doesn’t scale linearly. And systems designed for humans will keep breaking until they’re redesigned for agents.

🚀 𝐅𝐫𝐨𝐦 𝐒𝐭𝐚𝐭𝐢𝐜 𝐀𝐈 𝐭𝐨 𝐒𝐞𝐥𝐟-𝐔𝐩𝐝𝐚𝐭𝐢𝐧𝐠 𝐒𝐲𝐬𝐭𝐞𝐦𝐬: 𝐓𝐡𝐞 𝐍𝐞𝐱𝐭 𝐒𝐭𝐞𝐩 𝐢𝐧 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 While working on Agentic AI systems, one thing became very clear 👉 Most AI systems don’t fail because they don’t work… 👉 They fail because they don’t evolve with changing context I experienced this firsthand while building an AI workflow connecting: Jira → Code → GitHub The system worked well initially. But over time, a critical issue surfaced: 👉 The knowledge base became outdated as soon as the code changed. And that leads to real problems:  • Impact analysis becomes unreliable  • AI-generated suggestions lose accuracy  • Automation starts losing trust 🔁 𝗦𝗼 𝘄𝗵𝗮𝘁’𝘀 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝘀𝘁𝗲𝗽? Instead of static systems, we need: 👉 Self-updating AI systems that evolve with the codebase 𝗧𝗼 𝗮𝗱𝗱𝗿𝗲𝘀𝘀 𝘁𝗵𝗶𝘀, 𝗜 𝗲𝘅𝘁𝗲𝗻𝗱𝗲𝗱 𝘁𝗵𝗲 𝘀𝘆𝘀𝘁𝗲𝗺 𝘄𝗶𝘁𝗵: ✔ Real-time indexing using GitHub webhooks ✔ Incremental updates (no full reprocessing) ✔ Dependency-aware impact detection ✔ Scalable architecture using Docker, Qdrant, Redis & Celery 🧠 𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 This is where Agentic AI starts becoming truly practical: 👉 From “AI assisting developers” 👉 To “AI operating within development workflows” Modern AI systems are not just about generating outputs they need to adapt continuously as the environment changes. 🧩 𝗪𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 𝗲𝘅𝗽𝗹𝗼𝗿𝗲𝘀: ·      Keeping AI knowledge bases in sync with evolving systems ·      Real-time updates using GitHub webhooks ·      Incremental indexing vs full reprocessing ·      Practical challenges (Docker, Qdrant, Celery, ngrok, etc.) ·      Lessons learned from building a working system 🔗 𝗙𝘂𝗹𝗹 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 𝘄𝗶𝘁𝗵 𝗚𝗶𝘁𝗛𝘂𝗯 𝗹𝗶𝗻𝗸: https://lnkd.in/gvf7hzZ4 🔗 𝗣𝗿𝗲𝘃𝗶𝗼𝘂𝘀 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 𝗶𝗻 𝘁𝗵𝗶𝘀 𝘀𝗲𝗿𝗶𝗲𝘀: https://lnkd.in/g2J3zZSk #AgenticAI #AIEngineering #SoftwareEngineering #GenerativeAI #VectorDatabase #DeveloperProductivity #LLM #DevTools #RAG #ArtificialIntelligence #MachineLearning #DataScience #Automation #digitaltransformation #Qdrant #redis #celery #Prometheus