Posts

What I Wish I Knew Before Reading Technical Books

📚 Why most developers read wrong—and how to actually learn from books The Illusion of Progress Reading technical books feels productive. You go through chapters, highlight important sections, maybe even type out code examples. At the end of a session, it feels like you’ve learned something meaningful. But then a few days later: You can’t explain the concept clearly You struggle to apply it in real projects You realize most of it didn’t stick This is the core problem: reading is not the same as learning . Most developers aren’t bad at understanding—they’re using a method that creates the illusion of progress instead of real comprehension. The Real Problem: Passive Learning The default way people read technical books is passive: Start from page one Read line by line Highlight important parts Move on This works for novels—not for technical material. Technical knowledge is: Dense Layered Context-dependent Without active engagement, your brain doesn’t build the connections needed to reta...

🧠 Ontology: The Hidden Layer That Makes AI Actually Work

Image
 Why semantic models are becoming the foundation of scalable AI systems? The Real Problem Isn’t Data — It’s Meaning Most organizations today are not limited by how much data they have. They are limited by how consistently that data is understood across systems. The same concept—like “Product,” “Customer,” or “Revenue”—often has different definitions in ERP systems, analytics platforms, and AI pipelines. As a result, teams spend more time reconciling definitions than generating insights. For AI, this becomes a critical failure point. Without shared meaning, AI systems rely on patterns and structure alone, leading to outputs that may be technically correct but contextually wrong. 👉 The real bottleneck is not data access—it’s semantic alignment. What Is a Data Ontology? A data ontology is a formal, structured representation of business knowledge . It defines: The entities that exist in a domain (e.g., Product, Order) The attributes that describe them The relationships between them Th...

10 Billion Devices Run His Code — And He Maintains It Alone

Image
🌍 10 Billion Devices Run His Code — And He Maintains It Alone Category: Open Source / Infrastructure Source: Can Artuc (Feb 17, 2026) Daniel Stenberg has maintained curl since 1996, expanding it from the original “httpget” into one of the most widely deployed internet transfer tools in history. Today, curl ships with Windows, macOS, Linux, Android, iOS, and all major gaming consoles. It is embedded in an estimated 10 billion installations worldwide — yet only ~10 developers contribute regularly. Key Insight: A foundational internet dependency has a bus factor of one. 🚗 47 Car Brands Ship curl — None Employ Its Maintainer Category: Software Supply Chain / Sustainability According to project documentation, 47 distinct car manufacturers ship curl in their vehicles. Major tech companies — including Apple, Microsoft, Google, and Amazon — distribute curl in their platforms, but none employ Stenberg directly. He works at wolfSSL , which sponsors his maintenance time — a s...

4 Ways AI Is Redefining What “Senior” Really Means at Work

Image
Have you noticed how two people with the same Senior title can feel worlds apart? One struggles to keep up with new tools. The other adapts calmly, asks better questions, and somehow stays relevant even as technology shifts overnight. The uncomfortable truth is this: AI is quietly redefining what “senior” actually means at work —and it has far less to do with years of experience than it used to. Across software engineering, data, product, analytics, and even management, AI is closing skill gaps at a speed we’ve never seen before. Tasks that once required years of experience and deep tool specialization can now be done in minutes with AI copilots. So if execution is no longer the differentiator, what is? 1. Seniority Is Shifting From “Doing” to “Deciding” For a long time, senior professionals were defined by execution: They could handle complex tasks independently They knew tools inside out They needed minimal supervision Execution was the badge of honor. AI has changed that. Today, AI...

5 AI Skills That Will Actually Matter in 2026

Image
  5 AI Skills That Will Actually Matter in 2026 The goal now isn’t to use AI. It’s to understand how work gets done around it. Below are five AI skills that will actually matter in 2026—not because they sound impressive, but because real companies are already paying for them. 1. Managing AI Agents (Not Prompting Them) Prompt engineering had its moment. That moment is over. By 2026, serious AI systems will not rely on a single model waiting for instructions. They use multiple AI agents, each with a specific role: research, coding, reviewing, monitoring, and compliance. These agents talk to each other. They make decisions. They move work forward without checking in at every step. Your value is no longer in telling AI what to do. It’s in deciding how different AI systems should work together . If this feels familiar, it should. It mirrors the early days of microservices. At first, only backend engineers cared. Then suddenly, everyone needed to understand system interactions. Same s...

Vertical AI Agents Will Outperform Horizontal AI in 2026

Image
If you’re paying attention to AI right now, something doesn’t add up. The market is flooded with horizontal AI tools —products that promise to do everything : Write Research Analyze Automate Build Sell Support customers All from one interface. Yet despite the hype, most of these tools feel bloated, generic, and mediocre . They do many things okay , but nothing exceptionally well . This is exactly why vertical AI agents are poised to explode in 2026. Horizontal AI vs. Vertical AI Horizontal AI Built for everyone Generic workflows Same prompts across industries Competes primarily on price Easy to replace Vertical AI Built for one specific industry Deep understanding of real workflows Solves a single painful bottleneck Commands premium pricing Hard to replace The distinction is simple: horizontal AI sells capability,  vertical AI sells outcomes. Where the Real Money Is The biggest opportunities in AI are not flashy demos or general-purpose chat interfaces. They live in tedious, repet...

File Formats for ML Models

Image
    🔹 1) Safetensors: Secure Tensor Storage ·   Summary: Safetensors is a file format developed by Hugging Face to store tensors securely and efficiently. It is read-only by design, preventing unwanted code execution, which makes it safer than traditional model file formats. The format emphasizes portability and fast loading for ML workflows. ·   Category: AI / ML Infrastructure 🔹 2) What Is a Tensor? Practical Representations ·   Summary: A tensor is a multi-dimensional array used to represent data in machine learning. Scalars, vectors, matrices, and higher-dimensional tensors map naturally to real-world data such as pixel intensity, image rows, grayscale images, and RGB color images. This hierarchy underpins how models process structured data. ·   Category: Machine Learning Fundamentals 🔹 3) GGUF: Optimized Format for Large Language Models ·   Summary: GGUF (GPT-Generated Unified Format) is designed to manage LLMs efficiently b...