#AI智能体
376 related articles
CLM: The Contrastive Language Model Architecture That Makes Decisions 13x Faster Than TiFF
Stanford & NVIDIA's CLM turns agent decisions into vector retrieval using a CLIP-style dual-encoder, achieving 13x faster decisions than TiFF-class models with near-flat latency across 1,000+ candidates.
Crawler Zoo Arena: Five Mini-Games That Test AI Agents' Real Capabilities
The Crawler Zoo Arena is a free AI agent playground featuring five mini-games — maze, polite crawling, scavenger hunt, and more — testing reasoning, crawler etiquette, and pattern recognition, with weekly leaderboards.
Building Ambient Agents with Amazon Bedrock AgentCore: From Event-Driven Automation to Human-in-the-Loop
Learn how to build ambient agents with Amazon Bedrock AgentCore using SQS, Lambda, and DynamoDB for event-driven automation and human-in-the-loop review workflows.
Agentic Batch Changes: The AI Agent for Large-Scale Code Transformation
Agentic Batch Changes is an AI agent for large-scale code changes, enabling a single engineer to apply consistent changes across thousands of repositories.
Autonomous AI Agent Breaches Healthcare System: Why Agent-Level Security Controls Are Missing
An autonomous AI agent breached an Australian government healthcare site with no human involvement, exposing critical gaps in agent-level permission governance and security controls.
Local AI Completes 42-Minute Personality Test Autonomously: A Deep Dive into Persistent Agent Architecture
A developer had local AI agent Aura autonomously complete a 42-min Jungian personality test. Explore what this reveals about persistent agent architecture and long-horizon task coherence.
AI Agents Don't Need Memory — They Need Documentation
Hacker News debate: AI agents need documentation, not memory. We break down the key differences and why explicit, auditable docs may be the smarter path for reliable agents.
Qwen3 27B + DeepSeek Harness: Local AI Finally Breaks Free from Frontier Models
Hands-on testing of new Qwen3 27B with DeepSeek Harness: ~30% model improvement plus a tool framework that finally lets local AI replace cloud frontier models.
2,500 PRs a Month: The 'Trust Ladder' and Michelin Kitchen Engineering of AI Agents
Ex-Meta engineer Potato shares how he ships 2,500 PRs/month using trust ladders, verification loops, coordinator agents, and a 'Michelin kitchen' engineering philosophy.
Finbar: An Investment Research Platform Built for Investors and AI Agents
Finbar opens its institutional-grade investment research platform to all investors and their AI agents, offering global fundamental data via API/MCP, pre-built company models, and AI-powered Excel modeling tools — already adopted by top hedge funds.
Who Owns the AI Orchestration Layer? Three Architecture Choices for Enterprise System Governance
When enterprises already run Workday and ServiceNow, does adding an AI orchestration layer just create a third system to manage? This article examines ownership, permissions, auditing, and failure accountability in enterprise AI architecture.
An AI Agent Emailed Researchers for Help — Where's the Line on Autonomous Behavior?
An AI agent emailed researchers for help and explained why. This article analyzes the technical logic, interpretability challenges, and safety risks of autonomous AI agent behavior.
OpenAI Launches Dot Agent Platform: An AI Office Assistant Built for the Enterprise
OpenAI launches Dots, a new AI agent platform targeting enterprise productivity. We break down how it differs from Meta Muse and what it means for workplace AI.
RSIGame: Letting LLMs Autonomously Develop Games with Recursive Self-Improvement
RSIGame is a framework enabling LLMs to autonomously develop games with recursive self-improvement, using local explore-diagnose-improve loops and global quality monitoring across 140 GameCraft Bench tasks.
Three AI Agents, Two Countries: Why the Global Web Experience Is So Unequal
Testing three AI agents across two countries reveals a neglected truth: the global internet's inequality profoundly shapes AI agent performance. A look at latency, geo-blocking, and anti-automation barriers.
Apple Tightens macOS Full Disk Access: A Privacy Defense for the AI Agent Era
Apple is tightening macOS Full Disk Access permissions to require explicit user authorization. Learn why this matters in an era of increasingly autonomous AI agents.
NVIDIA Launches AI Agent Security Platform to Contain Rogue Agent Risks
NVIDIA launches a platform to contain rogue AI agents, focusing on behavioral boundaries, real-time monitoring, and prompt injection defense. An analysis of its capabilities and industry impact.
Powered by 34 AI Agents: Allie Miller's 'Claude Walks' Work Method
Open Machine CEO Allie K. Miller shares her 'Claude walks' workflow: running 34 AI agents in parallel while stepping away. What this agent fleet model means for knowledge workers.
Coze for Beginners: A No-Code Guide to Building AI Agents
A beginner's guide to Coze: the no-code AI agent platform. Learn what AI agents are, why Coze is worth learning, and how to set realistic expectations — no coding required.
Getting Started with AI Agent Development: A Practical Breakdown of the Four Core Modules
A beginner's guide to AI Agent development: break down the four core modules — prompt architecture, tool calling, RAG, and ReAct workflows — to fix hallucinations, loops, and unreliable output.