#AI编程
9457 related articles
Claude Code + Codex Multi-Agent System Tested: The Real Problem Behind a 98% Cache Rate
Developer shares 7-day test of a Claude Code + Codex multi-agent system: 98% cache rate, 354:1 input/output ratio, but stuck on orchestration and output share.
Cursor SDK New Feature: Real-Time Steering of Running AI Agents
Cursor SDK adds run.steer(), letting developers inject messages in real time while an AI agent runs to adjust its direction. Subagents move to the background on steer.
Go Tool Unspill: Scanning AI Coding Assistants' History for Credential Leaks
Introducing Unspill, a Go CLI tool that locally scans Copilot CLI, Claude Code, and Codex histories for leaked API keys and credentials, with redacted reports and no network requests.
AI Coding Assistant Fixed the Bug, Then "Lost Its Memory" in the Next Session
A Reddit developer complains that an AI coding assistant fixed a bug but completely forgot the fix in the next session, revealing the pain of missing session memory in AI Agents, analyzing causes, impacts, industry solutions, and practical advice.
Make the Most of Existing Logs: Fine-Tune a Small Model to Surpass Frontier Large Models
A Twitter developer shares: filter successful cases from existing runtime logs, fine-tune a 9B small model to beat frontier large models on specific tasks. This article analyzes the principles, implementation steps, and the smithtune tool, and explores data flywheel strategies in AI application development.
Meta Muse Wonder Lands on Cursor: 46.1% on CursorBench
Meta Superintelligence Labs' Muse Wonder launches on Cursor, scoring 46.1% on CursorBench at default effort — the highest in the Muse series. Input $4/M tokens, output $20/M tokens.
Cursor Opens Up System Prompt Customization: The Next Step in AI Coding Tool Controllability
Cursor now lets users replace the system prompt: custom prompts override defaults while Rules, Skills, and tool Schemas still load. Rolled out account by account.
The Essential Difference Between Chatbots and AI Agents: Crossing from Conversation to Execution
What's the difference between a chatbot and an AI agent? Based on the MCP series on the Learn Tech channel, this article uses examples like booking flights and summarizing notes to explain: a chatbot only gives answers, while an agent actually completes tasks through a think-act-observe loop. It breaks down the four core components—model, tools, memory, and goal—as well as MCP's role as a bridge, and discusses agents' applicable boundaries and permission risks.
Free GitHub Repo: 342 Hours of AI Courses, from Math to MCP Servers
A free GitHub repo offers 342 hours of AI courses: 523 lessons across 20 phases, from math fundamentals to agent building and MCP servers. Nearly 100,000 learners have started.
Claude Code Quota Draining Fast? Context Window Management Tips Save 35%
Claude Code quota draining fast? This article breaks down the root causes of context window bloat, sharing practical tips like cleaning by task milestones, disabling idle connectors, and streamlining CLAUDE.md. In real tests, the same 20 tasks cut token consumption by about 35%, reducing API costs from $5.94 to $3.91.
NVIDIA's 30,000 Developers Use Cursor, Code Commit Volume Triples
Cursor engineer Chris Diaz breaks down five organizational gaps blocking enterprise AI coding scale—strategy, risk tiering, governance, metrics, and operating model. NVIDIA's 30,000 developers use Cursor daily, tripling code commits while defect rates stay flat. The key: rewrite the SDLC, not add tools.
Ponytail: Giving AI Coding Agents a Senior Engineer's Mindset
Ponytail is a free GitHub tool that injects a senior engineer's mindset into AI coding agents, making them ask whether code is even necessary first. According to TamilDeck, it can compress 200 lines of AI-generated code to 30, cutting token costs, with support for Cursor rules and Claude Code skill integration—though security review still requires human oversight.
Complete Guide to Connecting Cursor AI to the GitHub MCP Server
A step-by-step guide to connecting Cursor AI to the GitHub MCP server: create a personal access token, configure mcp.json, store the token in the ENV block, and verify with a PR query.
WorkBuddy Full Tutorial: Build an AI Office Automation Agent from Scratch
A complete WorkBuddy tutorial: installation, three working modes, multimodal picture book video practice, expert team agent building, and workflow configuration for low-cost AI office automation.
8 AI Accounts, 7 Failed: An Honest Review of Using AI to Get Work Done
Only 1 of 8 AI-run Bilibili accounts succeeded. We break down the real difference between "performing AI" and "using AI to get work done."
DeepSeek Harness Step-by-Step Installation Guide: Core Principles and Building Automated AI Applications
Based on Bilibili creator Pan Ge's DeepSeek Harness tutorial, this guide covers Node.js installation, npm global directory and cache configuration, environment variables, and the core principles of Harness with Agents.
WorkBuddy Beginner's Guide: A Full Breakdown of Desktop AI Agents and MCP Configuration
WorkBuddy desktop AI agent beginner's guide: its core capabilities, differences from CodeX, three major misconceptions, and the MCP configuration and Expert Center learning path to help office workers transform from AI question-askers into AI managers.
AI Tool Boundaries and Automation Practices for Researchers in the Agent Era
From webcoding to research automation, Agent tools are reshaping research workflows. This article outlines the core differences between chatbots and coding Agents, analyzes multi-Agent division of labor and tool categories, and explores AI's capability boundaries in research—researchers still need to master problem formulation, experimental design, and result evaluation.
Complete Guide to Switching AI Models in Cursor: In-Conversation Switching and Default Settings
How to switch AI models in Cursor: use the selector at the bottom of the chat panel to switch temporarily, change the default model in Customize, and troubleshoot missing models caused by subscription plans.
I Questioned the AI Agent, and It Changed My System Design
A real account of collaborating with the built-in AI Agent Spaceship: by questioning why the index router doesn't mount decision models, the Agent responded and changed the system design. Exploring the collaborative value and usage tips of AI Agents in architectural decisions.