🔥 Big Story
Anthropic Open-Sources Agent Skills, Challenging OpenAI's Enterprise Dominance
Anthropic has made a strategic power move by releasing its Agent Skills technology as an open standard on December 18, 2025. The open specification and reference SDK, published at agentskills.io, allows developers to create portable skills that work across Claude and other AI platforms. Microsoft has already integrated Agent Skills into VS Code and GitHub, with Cursor, Goose, Amp, and OpenCode following suit.
The bigger story: Y Combinator announced that Claude has dethroned OpenAI as the preferred AI model among its Winter 2026 startup batch—the first time OpenAI hasn't dominated YC startup preferences since ChatGPT's launch.
Why it matters: Anthropic's open-source strategy is paying dividends—they're winning developer mindshare and enterprise adoption simultaneously.
⚡ Quick Updates
Anthropic Launches Claude Chrome Extension
Claude is now available as a Chrome extension for Pro subscribers ($20/month), enabling browser automation, form filling, and scheduled tasks. Works seamlessly with Claude Code for debugging.
Read more →
Claude Opus 4.5 Achieves Record 5-Hour Task Horizon
METR benchmark results show Claude Opus 4.5 achieved a 50% time horizon of approximately 4 hours and 49 minutes—the highest score ever recorded for sustained complex task completion.
Read more →
Google Publishes Year-End AI Breakthroughs
Google's 2025 research review highlights Gemini 3's dominance, processing over 1 trillion tokens per day. Gemini 3 tops LMArena leaderboards with state-of-the-art scores on Humanity's Last Exam.
Read more →
Google Cloud Joins Genesis Mission
Google Cloud partnered with the DOE's Genesis Mission, providing Gemini for Government across all 17 National Laboratories with agentic AI tools.
Read more →
Samsung SDS Becomes First Korean OpenAI Reseller
Samsung SDS signed a reseller partnership with OpenAI, becoming the first Korean company to offer ChatGPT Enterprise to the Korean enterprise market.
Read more →
📄 Top Research Papers
Gnosis: Can LLMs Predict Their Own Failures?
A lightweight self-awareness mechanism (~5M parameters) that enables frozen LLMs to predict failures by inspecting internal states. Consistently outperforms external judges in accuracy and calibration.
Impact: Major step toward solving hallucination problems in production AI.
LongVideoAgent: Multi-Agent Reasoning with Long Videos
A breakthrough framework for hour-long video reasoning. A master LLM coordinates grounding and vision agents using RL-trained cooperation. Outperforms baselines on LongTVQA benchmarks.
Impact: Could revolutionize video understanding for movies, lectures, and surveillance.
SpatialTree: How Spatial Abilities Branch Out in MLLMs
A cognitive-science-inspired hierarchy organizing spatial reasoning into four levels: perception, mental mapping, simulation, and agentic competence. Proposes an "auto-think" strategy that suppresses unnecessary deliberation.
Impact: Could change how we train spatial reasoning for robotics and autonomous systems.
💻 Top GitHub Repos
langgenius/dify |⭐ 122.9k +
Production-ready platform for agentic workflows with visual builder, RAG pipeline, and MCP integration across GPT, Gemini, and Claude.
browser-use/browser-use |⭐ 74k+
Enable AI agents to automate browser tasks via Playwright. Navigate websites, interact with elements, and complete complex web workflows autonomously.
infiniflow/ragflow |⭐ 70k+
Leading open-source RAG engine with agentic capabilities, deep document parsing, GraphRAG support, and MCP integration.
OpenHands/OpenHands |⭐ 65.9k+
AI-driven development agent that writes, tests, and debugs code autonomously with GitHub integration.
block/goose |⭐ 25k+
Block's open-source extensible AI agent for coding. Installs packages, executes commands, edits files—now with Agent Skills support.
🛠️ Top AI Products
Super Agents by ClickUp |👍 531 upvotes
AI teammates you can spin up in seconds. @mention, assign, direct message, and schedule agents just like human teammates.
/agent by Firecrawl |👍 486 upvotes
Magic API that gathers structured data from anywhere on the web. Describe what you need; the agent handles navigation and data structuring automatically.
Claude in Chrome |👍 439 upvotes
Anthropic's browser extension enabling Claude to see, click, type, and navigate in your browser. Now in beta for all paid subscribers.
🐦 Top Tweets
@ycombinator
"Congrats to @LemonSliceAI on their $10.5M seed!…"
@GeminiApp
"Another gift before 2026 🎁
For a limited time, new members get from 50% off the Google AI Pro annual plan (auto renews at the subscription price after offer ends)…"
@github
"It was the closest race yet when it came to the top programming language of the year. 👀"
@NVIDIAAI
"NVIDIA Nemotron 3 Nano is now available as a fully managed, serverless model on Amazon Bedrock…"
Until next newsletter, Stay Curious and Keep building!!
Brain Pulse
