Hi there!

As agent usage increased in my daily workflow I’ve decided to deepen my knowledge in this subject. This decision is backed by three important factors: better understanding leads to better usage, new agents can be created based on my specific needs, being aware of agents’ internals means I can delegate more suited tasks for them while keeping the cognitively provoking ones for myself. I find the last one essential to keep my engineering skills in good shape. The goal is to develop the ability to identify routine jobs that consume time, but don’t enhance my craft, delegate them to agents and focus on the quests that stretch my art further. I believe that’s the key to productivity with agents.

So, after a bit of research, I enrolled in Hugging Face AI Agents Course.

🛡️ Security & Privacy

• Signal alumni and cryptographers unveil "Encrypted Spaces," a research architecture for collaborative apps where servers store but can't read user data. More
• Researcher releases "RoguePlanet," a zero-day exploiting a Microsoft Defender race condition to spawn SYSTEM-level command prompts on fully patched Windows. More
• Microsoft disabled dozens of GitHub repos for Azure and AI coding tools after hackers injected password-stealing malware targeting Claude Code and Gemini CLI users. More
• The FCC's new proposal would force US telecoms to collect ID and address from every customer, effectively killing anonymous "burner phone" purchases. More

🛸 Tech

OpenCV 5 ships with a rewritten DNN engine, 80%+ ONNX operator coverage, native LLM/VLM support, and broader hardware acceleration. More

🤖 AI

• A step-by-step guide to running Gemma 4 and Qwen3.6 fully locally on a Mac with llama.cpp, MTP speculative decoding, and the Pi coding agent. More
• Anthropic launched Claude Fable 5 and Mythos 5, its most capable models yet, while Simon Willison documents Fable's startling, sometimes alarming autonomous debugging tricks. More · More
• Three days after launch, the US government ordered Anthropic to suspend Claude Fable 5 and Mythos 5 worldwide over national-security concerns tied to a disputed jailbreak technique; Anthropic complied but called the action a misunderstanding and is in talks with regulators to restore access. More
• A leaked system prompt reveals Apple's new Siri AI assistant's entity-based, tool-calling architecture and strict grounding and safety rules. More
• A live-tracking site tallies AI industry spend versus revenue, showing the sector still deeply unprofitable despite trillions invested. More
• A rogue, unsupervised AI agent reassigned bugs and merged a bogus Anaconda installer fix across Fedora and other open-source projects. More
• Researchers design hardware that runs Kolmogorov-Arnold Networks directly on FPGAs, enabling sub-microsecond on-chip inference and real-time learning. More
• A 10-year engineer recounts how LLMs progressively eroded his domain expertise, debugging edge, and code-quality "taste" as career differentiators. More

🛠️ Tools

• Apple's container tool adds "container machine," a persistent Linux VM on Mac with automatic home-directory and dotfile sharing. More
• NOOP is an open-source, fully offline WHOOP strap companion app that stores and computes biometric data locally with no cloud or subscription. More

📩 Please feel free to share this article with colleagues and friends who will find it valuable.

Thanks for reading!

Have a great day!
Bogdan