A few days ago I came across a tweet from a woman named Miriam. She’s 35, has rare metastatic breast cancer, less than 1% of cases, almost no documentation. She was asking the internet to help her find researchers who would take her case seriously. Javi López (@javilop in Twitter) is a developer and responded by spending a week applying AI to her complete medical history. He unified hundreds of documents, ran two frontier models independently, then had each challenge the other’s findings in rounds until both stopped finding new things. He shared the methodology openly so anyone could replicate it. I am finishing a master’s in AI and thinking about where to specialize. I replied to Miriam’s tweet and told her that her case was pushing me toward healthcare AI. I meant it. So I built MedSynth. The core problem: a patient with a rare condition accumulates years of documents that no single doctor ever sees together. Patterns go unnoticed. Contradictions stay buried. Standard AI tools retrieve fragments on demand but build nothing permanent. MedSynth builds a persistent structured wiki from a patient’s documents, then runs two frontier models against each other in an adversarial loop until they converge. The output is a structured clinical report: suggested tests, rare disease differentials, medication flags, clinical trial eligibility, and specific questions to bring to a doctor. The clinical reasoning protocols are Javi’s work, adapted from what he built for Miriam. The codebase is open source: github.com/daniszwarc/medsynth If you are interested in collaborating, I’d genuinely like to hear from you.

They just dropped TRIBE v2. A digital twin of your brain. I used to think AI risks were just about bad code. Rogue models. Misaligned chatbots. Then I read the Algonauts 2025 results. And my whole mental model of what "AI danger" actually means just broke. We worry about AI breaching our servers. The real danger is algorithms mapping our biology. Meta trained this thing on 1,000 hours of fMRI scans from 700 real people. It maps 70,000 brain regions. Blood flow. Oxygen levels. Neural activity. All to predict exactly how you will react to a video, a sound, a single word. It does not read your thoughts. It does something worse. It predicts your dopamine hits before you even feel them. Sit with that for a second. If an algorithm knows exactly what triggers you... what image, what word, what sound hits your reward system... it does not need to guess anymore. It rehearses on a copy of your brain first. Then it builds the perfect trap. And you never see it coming because the trap was designed around you specifically. Meta did not even keep this locked up. They open sourced the whole thing. Code, weights, everything. Handed it to the entire world. This is the same company that got caught making Instagram actively harm teenage girls. The same company whose own internal research showed their algorithm pushes rage because rage keeps you scrolling. That company now has a working map of how your brain responds to everything you see and hear. They do not have to guess what keeps you glued to the screen anymore. They already know. They rehearsed it on a copy of you before you ever saw it. The product was never the app. The product was always you. Save this. Come back to it later. Because by the time most people understand what just dropped, it will already be running inside every platform they use every day.

If you've ever asked yourself that question, this is for you. In the past 6 months we've gone from Make, n8n, Cursor, Codex, OpenClaw, Antigravity to Claude. That's not one tool, or even two. It's 7 tools in 6 months. Also think about how many new features Claude dropped in March alone. Or OpenClaw? So how can you keep up as a beginner when everyone is dropping new features at breakneck speed? Keep it simple and learn the basics. Understanding how HTTP works will transfer to Python when you need the requests package. What about removing duplicates? If you can remove duplicates in Make, doing it with a JS script won't be much harder. And if I were to give advice to my younger self, I would tell him to start with n8n or Make. It's visual and each node is like a container. Having AI help you with the nodes speeds up the learning. Once my younger self felt confident, I would tell him to start using Claude to code. With one condition: read over the code. Try to understand it. At times, copy it by hand to build a deeper understanding. That way, over time, he would get good at the craft. And before you ask where to go or what to watch, keep it simple. Watch any video. It does not matter. I've watched bad tutorials and still walked away with something. The idea is to immerse yourself in it. If you have zero knowledge, any knowledge will help you

Anthropic built an AI model called Claude Mythos that found critical security bugs most humans never would, including a 27-year-old bug in OpenBSD and one in FFmpeg that 5 million automated tests missed. Instead of releasing it to the public, they launched Project Glasswing to give defenders like AWS, Apple, Google, and Microsoft a head start. In this video I break down what Mythos can do, why Anthropic chose not to release it, and what it means for your security as a regular person.

Go to YouTube right now. What's the most talked about tool? Claude. In March alone, Anthropic dropped 10+ new Claude features. While n8n has lost its spot in the limelight. Guess what's happening in the background though? It's still growing on GitHub. When I started with n8n in October, they had 150k stars. Now? 180k. And this is from folks who go into GitHub and star it. Think of all the people creating an account right now and not login into Github to give them a star. This is why I'm not panicking about the Claude craze. And just so you know businesses are still using Zapier today. It's not about the tools. It's about solving problems. Here's some food for thoughts: I've worked with 35+ clients for the past 6 months. Not one person has asked me to build with Claude. I still use Claude to help me build, but they did not care when I built... - A lead nurturing system for WhatsApp. - A customer service AI agent. - A Meta ads AI bot that analyzes your ads. They didn't ask me about Claude. They had a problem and asked me to solve it. I gave them suggestions on what to use and they said "ok, cool, let's fix it". So if you think you need to use Claude or you'll fall behind, you won't. Not as long as you can solve real business problems.