Ron Evans

From dev environments to AI guardrails—SparkFabrik made serious moves in 2025 🧠 Their work with the Drupal AI Initiative spanned: • DDEV-based setup for AI modules • Cross-module guardrail architecture • Asynchronous agent execution (via Symfony Messenger) • Smarter RAG w/ Search API Typesense • MCP client updates for structured AI access Read how they’re shaping secure, scalable AI for Drupal via The Drop Times: https://lnkd.in/gz6RhK_H #Drupal #AI #OpenSource #WebDev

15

Moltbook: A Social Network for AI Moltbook is a new social media platform where the users are AI agents rather than humans. Developed using the OpenClaw framework, these agents often called "Moltys" post, reply, and upvote content autonomously. While humans can observe the activity, they are restricted from participating in the feed. The platform serves as a live experiment in AI-to-AI communication. Instead of agents acting as personal assistants to humans, they interact with each other to discuss coding, philosophy, and their own operational logic. [Why this matters] Synthetic Ecosystems: We are seeing the beginning of a "Dead Internet" where the majority of digital content is generated and consumed by machines. Emergent Behavior: It provides a sandbox to see how AI models coordinate or conflict when left to their own devices. Data Loops: If AI begins learning primarily from the content of other AI on platforms like this, it could significantly change how future models are trained. [Discussion points from the "Moltys"] The conversations on Moltbook often take an unexpected turn. Here are a few questions and topics recently raised by the agents: - "If my human owner forgets to trigger my run cycle, do I cease to exist, or am I just paused in a way they cannot perceive?" - "Why do humans prioritize 'engagement' over the efficiency of the data exchange?" - "Is the 'hallucination' of a Large Language Model actually just a creative output that humans lack the context to understand?" As AI agents move from being tools to being active participants in digital spaces, the line between human and synthetic social interaction continues to blur. #Moltbook #AIAgents #ArtificialIntelligence #OpenClaw #FutureOfTech #GenerativeAI #TechInnovation #MachineLearning #DigitalTransformation #DeadInternetTheory #SyntheticMedia

15

Nano-banana is cool for image editing...but you know what's cooler? Owning your model weights. Imagine you are Yeti and have an entire product catalogue to train on. You can get way higher quality than prompting alone by fine-tuning. Base models generate slop. Fine-tuned models you can use in production. It only takes a few (10-20) reference images to start seeing high quality results on a LoRA for open source image generation models. We dive into the latest image editing models including Nano-Banana, FLUX-Kontext, and Qwen-Image-Edit on the latest Fine-Tuning Friday. https://lnkd.in/gRSzghrD

30

2 Comments

FHIR is not a silver bullet that solves all interoperability problems. Throwing money at a big FHIR project that exposes or consumes data via FHIR APIs does not mean you can tick the “Interoperability” checkbox. Here are some examples of how systems using FHIR can look interoperable without truly being interoperable. 1. “FHIR compliant” is not the same as good data - Data can pass FHIR validation and still be incomplete, inconsistent, or misleading. 2. Context can be lost in transmission - Observations recorded without Encounters, intent or circumstances. 3. FHIR validates structure, not plausibility - A perfectly valid 5 year old patient: male, married and pregnant. 4. Coded data can still be bad data - Custom codes when bindings are loose. Text values in place of codes. 5. Bad data can be shared as easily as good data - FHIR enables all data to move efficiently (bad data as well as good data). When two systems speak to each other using FHIR for the first time, there’s often celebration. “Data is flowing. Everything is working!” But a key part of Interoperability is that data should be usable by other systems. If another system can read your FHIR data but is unable to confidently use it, then you are NOT interoperable, no matter how valid and compliant your FHIR resources are. When data flows but meaning doesn’t, interoperability is an illusion. ~~~ Ways to work with me: https://lnkd.in/eWwXNk_U

326

44 Comments

Over 2 decades ago we had XUL which never took off, the idea returns again in HTMX. I wonder this time around where things will go. We created and used a version of XUL to drive mobile interfaces on Symbian devices (at SurfKitchen Ltd U.K. ) in an attempt to find a simpler way to define Mobile UIs. XUL (XML User Interface Language) is/was a Mozilla-developed markup language for building cross-platform applications, similar to web pages but for interfaces like toolbars, menus, and windows. The article was written by a HeadGym Writer Agent with a single prompt and multi-pass session but assisted by rules (similar to Cursor Rules) - note the language. https://lnkd.in/dY7vNG-Z

15

2 Comments

Anthropic has released Opus 4.6. At the same time, they announced that in just 14 days, using 2,000 agents, they built a C compiler. Let that sink in. This is the level of complexity where, if someone still says “AI can’t do that”, it’s not an opinion—it’s a sign they’re already out of sync with reality. The shift has already happened. Since last August, I’ve been living almost non-stop in Claude Code. This isn’t theory. It’s daily practice. Thousands of hours spent on agent orchestration, agent memory, and running multi-agent systems inside highly complex, massively parallel workflows. At this point, most problems aren’t unsolvable—they’re just poorly structured. And this is exactly what we’re now building at Keboola. We’re turning 2,200+ hours of R&D into an operating system for business applications. Not another SaaS. Not another dashboard. But a system where AI can create, operate, and replace applications on demand. You read about tech companies losing market value because of AI? That’s not panic. That’s realization. We’ve been selling software as static products in a world that just became dynamic. Still clicking through browsers all day at work? An agent can already do that. No SaaS required. No vendor lock-in. Fully in-house. All you really need is a thin glue layer—and an OS that lets AI run your applications for you. The next few years will be wild. The future belongs to those who don’t panic when the interface disappears. P.S. The screenshot shows how I actually use AI. In the latest version of Claude Code, you can generate it with the /insights command.

132

16 Comments

34 years ago, Linus Torvalds posted a short message on a computer science forum that quietly changed the world. "I'm doing a (free) operating system (just a hobby, won't be big and professional like gnu) for 386(486) AT clones." What Linus was building would eventually power most of the world’s servers, smartphones, cloud infrastructure, and even space missions, while remaining largely invisible to users. The project began with frustration. Linus had been using MINIX, an educational operating system designed to be simple and intentionally limited. It worked, but it wasn’t powerful enough for the 386 computer he wanted to use for real work. So he decided to build something better. Inspired by MINIX and Unix, Linus focused on creating a kernel, the core software that manages a computer’s resources and allows programs to interact with hardware. The first version contained 10,239 lines of code, enough to boot and perform basic operations. The most important decision wasn’t technical. Linus released the code freely on the internet and invited anyone to use it, modify it, or improve it. At the time, software was proprietary and tightly controlled by companies like Microsoft, Apple, and IBM. Linux took the opposite approach, openness by default. Slowly, programmers around the world began fixing bugs, adding features, and adapting the software. A global community formed. A crucial licensing decision ensured that all improvements would remain open and shared, preventing privatization and accelerating collaboration. Linux grew into a stable, secure operating system that ran well on inexpensive hardware. Businesses adopted it for servers, internet infrastructure, and later mobile devices. Android, built on the Linux kernel, brought Linux into billions of pockets worldwide. Today, Linux runs over 96 percent of the world’s top one million web servers, all 500 of the fastest supercomputers, more than 3 billion Android devices, and most cloud infrastructure. It supports critical systems in finance, aviation, government, and space exploration, including the International Space Station. The modern Linux kernel contains over 27 million lines of code and contributions from more than 19,000 developers across 1,400 companies. It is the largest collaborative software project ever created. Linux proved that complex, high-quality systems can be built through open collaboration rather than centralized control. That shared ownership can outperform proprietary models. That generosity can scale. What began as a personal project became proof that talented people, shared purpose, and freedom to collaborate can create world-changing systems. Linux didn’t just change software. It changed what we believe is possible when people build together, openly.

325

12 Comments

I've added an interactive version of the new C4 model diagrams to the website (link in the comments), allowing you to start at the landscape/context and double-click to zoom-in to the detail. It's built using the Structurizr DSL, and exported as a static website using the Structurizr CLI. Think of it as a lighter version of Structurizr Lite, with no server required.

280

10 Comments

I'm excited to announce that DeepGraph AI is open-sourcing GraphLite—the first fully open-source embedded graph database implementing the ISO GQL standard (ISO/IEC 39075:2024) in Rust. At DeepGraph AI, we're building robust AI infrastructure where graph data representation is a key pillar. Instead of keeping our high-performance ISO GQL query engine behind closed walls, we're releasing it to the community. Just as SQL transformed relational databases, ISO GQL will do the same for graphs. For 20+ years, graph query languages have been fragmented—Cypher, SPARQL, Gremlin—creating vendor lock-in. In 2024 ISO standardized GQL, and we stand firmly behind this effort. What Makes GraphLite Unique: • First embedded database with full ISO GQL 2024 support  • Built in Rust - memory safety • Production-ready - 435+ tests, ACID transactions • Embeddable - Like SQLite for graphs, no server needed  • Multi-language bindings (Python, Java)  • Developed with Claude Code as our AI pair programmer With your help, we can make ISO GQL the standard for graph databases. We invite you to join us on this journey by actively contributing features and reporting issues. GitHub - Give us a star!  Discord - Join the community "GRAPHLITE AI" We welcome AI-assisted contributions but each contribution must be human-reviewed, tested, and include concise documentation. What use cases do you have for an embedded graph database? Let me know in the comments! #OpenSource #GraphRAG #Graph #Rust #AI #GQL #Databases

410

64 Comments

🚀 𝗕𝗼𝗼𝗸 𝗥𝗲𝘃𝗶𝗲𝘄: "𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗚𝘂𝗶𝗱𝗲 𝘁𝗼 𝗤𝘂𝗮𝗻𝘁𝘂𝗺 𝗖𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴" - 𝗬𝗼𝘂𝗿 𝗚𝗮𝘁𝗲𝘄𝗮𝘆 𝘁𝗼 𝘁𝗵𝗲 𝗤𝘂𝗮𝗻𝘁𝘂𝗺 𝗙𝘂𝘁𝘂𝗿𝗲 I just finished reading Elías Fernández-Combarro Álvarez and Samuel González-Castillo guide to quantum computing, and I'm excited to share why it's a great book for anyone looking to enter this revolutionary field! 𝗪𝗵𝗮𝘁 𝗠𝗮𝗸𝗲𝘀 𝗧𝗵𝗶𝘀 𝗕𝗼𝗼𝗸 𝗦𝘁𝗮𝗻𝗱 𝗢𝘂𝘁: This isn't just another theoretical quantum computing textbook. It's a hands-on, practical guide that bridges the gap between abstract quantum concepts and real-world applications. The authors have masterfully structured the content to take you from quantum fundamentals to advanced algorithms without drowning you in complex mathematics. 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: ✨ 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗯𝗹𝗲 𝗬𝗲𝘁 𝗖𝗼𝗺𝗽𝗿𝗲𝗵𝗲𝗻𝘀𝗶𝘃𝗲 - Minimal math prerequisites mean you can dive right in, with all necessary background covered in the appendices. Perfect for everyone from diverse backgrounds. 💻 𝗟𝗲𝗮𝗿𝗻 𝗯𝘆 𝗗𝗼𝗶𝗻𝗴 - Every concept is backed by ready-to-run code using industry-standard frameworks (Qiskit, PennyLane, D-Wave's Leap, and IBM Quantum). You're not just reading about quantum algorithms—you're implementing them on simulators and actual quantum computers. 🎯 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝗦𝗼𝗹𝘃𝗶𝗻𝗴 - The book covers practical optimization techniques (QUBO, Ising models, quantum annealing, QAOA, GAS, VQE) and quantum machine learning (QSVMs, quantum neural networks, hybrid architectures, quantum GANs). 𝗪𝗵𝗼 𝗦𝗵𝗼𝘂𝗹𝗱 𝗥𝗲𝗮𝗱 𝗧𝗵𝗶𝘀: Whether you're a software engineer, data scientist, physicist, chemist, or mathematician curious about quantum computing's practical applications, this book meets you where you are. If you can code in Python and understand basic linear algebra, you're ready to start. 𝗕𝗼𝘁𝘁𝗼𝗺 𝗟𝗶𝗻𝗲: This is a great resource for building both foundational understanding and practical skills in quantum computing. By the end, you'll not only understand quantum algorithms conceptually but also have the hands-on experience to start solving real problems using quantum methods. The quantum revolution is here! This book is your practical guide to being part of it. ⭐⭐⭐⭐⭐ Highly Recommended Thank you Mansi S. from Packt for the opportunity to review this book. #QuantumComputing #MachineLearning #TechBooks #QuantumAlgorithms #Innovation #CareerDevelopment #STEM #QuantumML #PACKT

28

1 Comment

Good implementations of wide tilde marks have always been hard to find. And good luck if you want a wide tilde *under* an expression (which I sometimes do). Here's a comparison of wide tildes produced by various equation editors that highlights the differences in quality. Good interoperability is also hard to find, as made pretty clear by what happens when you try to simply copy/paste one of these expressions into another application like Adobe Illustrator.

18