We asked AI to fix one slow workflow. We got 19 agents, 3 councils, and a 5-stage journey ending in "Ascend." The actual fix was three index cards: fix the process, name an owner, start small. They're now in a glass case labeled "what we should have done." Instead — Agentic
SYSTEMS
-
Ego Agent: AI runtime outperforming headless setups for seamless automation
By
–
4/ Why @ego_agent outperforms headless setups: → Shares your active Chrome state to skip auth loops.
→ Semantic snapshots read deeply nested iframes without̀ hallucinating. A TRUE first-class runtime for AI, not an afterthought. → https://
lite.ego.app macOS. 100% FREE. -
Direct vs. indirect leakage in large model training on GSM8k
By
–
You mean direct & provable "leakage"? Because the large model writing the solution likely was trained on GSM8k — accidentally or on purpose — so there must be some indirect assumptions being made all along (also a form of leakage).
-
New NVIDIA Cosmos 3 Model Provides Pretrained Foundation for Physical AI
By
–
Trained on billions of samples across modalities, the model provides developers with a powerful pretrained foundation for building physical AI systems with less data and lower training costs. Read more in our technical blog: https://
developer.nvidia.com/blog/develop-p
hysical-ai-reasoning-world-and-action-models-with-nvidia-cosmos-3/
… -
LLMs as bandaids: need for a sounder AI foundation
By
–
Current AI is bandaids all the way down; LLMs can’t play nicely with basic tools like databases and knowledge graphs, and you never know what you are going to get. It’s long time to face facts: LLMs have their uses, but we need a sounder foundation for AI.
-
Solving a multi-agent management problem with Google DeepMind
By
–
It's been almost 1 and a half years that I can't solve a multi-agent cohort management problem and now I just came across a library from @GoogleDeepMind released 2 months ago that offers an ultra elegant approach. Damn, this company is really the boss.
-
Second Brain & Financial App: File System + Agentic AI is Enough
By
–
Second brain, Obsidian; financial app has its own frontend. In the end, the system only requires a file system (folders on your computer) and an agentic AI (Codex, Claude Code, etc.)
-
Karpathy’s brain ingestion paradigm: A scalable pattern for various applications
By
–
Karpathy's second brain ingestion paradigm, where you have a folder to dump your raw data into and then an agent processes and structures it to add it to a wiki or second brain, is a scalable pattern for so many other applications. In my case, for example, I already had a
-
AI Agents: Orchestrating Multiple LLMs for Enhanced Capabilities
By
–
AI Agents ≠ single LLM They’re systems of models • General LLMs → reasoning
• Domain LLMs → expertise
• RAG → real-time data
• Tools → actions & automation
• Open-source → control & privacy Real power = orchestration of multiple LLMs Via Giuliano Liguori -
From Apprenticeship to Agentic Orchestration: Digital Workers as the Primary Engine
By
–
This represents a transition from slow apprenticeship to agentic orchestration where digital workers become the primary engine.