#AI编程
9360 related articles
AI Application Developer Learning Path: Agent Engineering and Vibe Coding in Practice
A learning path for AI application developers from zero基础 to enterprise-grade agents: LLM APIs and private deployment, LangChain/LangGraph, MCP/Skills/A2A, context/harness/loop engineering, GraphRAG, and Vibe Coding with Cursor and Claude Code.
PatchPilot: How Kernel Patches Move from Modification to Delivery Endpoint
How PatchPilot turns kernel patches from a code change into an orchestrated, verifiable delivery process: triage, dependency analysis, per-conflict-hunk reasoning, structured failure memory, delivery re-check, and persistent recovery—letting engineers decide based on evidence whether to retry or take over.
From Principles to Practice: A Complete Guide to Understanding AI Agents and the ReAct Pattern
Based on a Bilibili hands-on course, this guide explains the essence of AI Agents and the ReAct pattern, then shows how to build a working ReAct Agent from scratch using a system prompt and Python code—ideal for beginners in agent development.
LangChain for Beginners: Four Key Pain Points and When to Use It
Bilibili LangChain tutorial explained: what LangChain is, how it solves four LLM weaknesses—outdated knowledge, no private data, no tool calls, lost context—plus when to use it and when not to.
Li Hongyi's Introduction to Generative AI: The Right Path to LLMs for Complete Beginners
What does NTU Li Hongyi's "Introduction to Generative AI" actually cover? This article breaks down the course positioning, three modules, assignment design, and AI attitude, offering complete beginners a path to learning LLMs by understanding principles first rather than chasing tools.
Tech FrontiersWe Tested 7 Claude Code Tips: Only Two Actually Save Usage
69 controlled runs testing seven popular Claude Code quota-saving tips: only /clear and a lean CLAUDE.md really save money, subagents cost 5.5x more, and compact and concise styles are net losses.
Cursor Pro Discontinued, Claude Code Cost Savings, and Five Major AI Coding Developments
Cursor cancels its $20 Pro tier and folds into X subscriptions, Anthropic builds its own version control, Claude Code auto-compresses context, GitHub Copilot orchestrates multiple models, and Alibaba Qoder opens agent teams to individuals—five major AI coding developments reveal a shared thread of pricing restructuring and cost optimization.
Cursor Official Plugin Swapped: Claude Code Extension Hid an SSH Backdoor
A Claude Code plugin in Cursor's official extension marketplace was swapped for a malicious version, planting an SSH backdoor for about 11 hours and affecting up to 5,000 developers.
Tencent WorkBuddy Hands-On: How AI Agents Crush Traditional Chat Assistants
WorkBuddy vs Doubao tested: how AI Agents directly operate computers to convert PPT to PDF, archive files and more. Analysis of WorkBuddy's low barrier, domestic ecosystem, pricing and plugin marketplace, plus AI Agent selection advice.
Claude Code v2.1.316 Update Analysis: Enterprise Gateway and Multiple Stability Fixes
Claude Code v2.1.316 released: new enterprise gateway pooled group budgets, layered policy mode, subagent effort control, dozens of Shell, network, and session recovery fixes, and Opus 4.8 fast mode deprecated.
FDE (Forward Deployed Engineer): From Concept to Practice — The Key Role in AI Implementation
An in-depth look at FDE (Forward Deployed Engineer): concept, deliverables, and career path. Big tech pays 100K RMB/month; OpenAI, Anthropic, Alibaba, DeepSeek are all investing. From RAG to Agent development to private deployment — the key role in AI implementation explained.
Is Harness Dead? Why the Creator of T3 Code Says Stop Obsessing Over AI Coding Tools
T3 Code creator explains what Harness really is, why models and Harness are converging with only a few percentage points of difference, and why engineers obsess over the wrong details.
Tech FrontiersDeepSeek Harness in Practice: A Beginner's Guide to Engineering AI Agents
Based on a Bilibili tutorial, this guide explains how DeepSeek Harness enables engineering AI Agents, covering core concepts, four roles, hands-on projects, and evolution.
Surface RTX Spark Dev Box Runs 120B Model Fully Offline: An Intelligent Terminal Hands-On
Microsoft Surface RTX Spark Dev Box demo: running a 120B local model offline via open-source Intelligent Terminal, building a bootable Hyper-V VM with one sentence, showcasing Project Zemuth local AI.
@ai-sdk/workflow 2.0.91 Patch Released: AI Dependency Synced
@ai-sdk/workflow releases patch 2.0.91, syncing the ai dependency to 7.0.159. This article explains the automated release changes, engineering signals, and upgrade advice for developers.
AI SDK Workflow Harness Toolkit 1.0.170 Patch Released
AI SDK workflow toolkit @ai-sdk/workflow-harness releases patch 1.0.170, aligned with @ai-sdk/harness version, auto-published via GitHub Actions, low-risk patch-level update.
AI Comic Drama with WorkBuddy from Scratch: A Step-by-Step Skill Pack Installation Guide
Learn how to create AI comic dramas with WorkBuddy from scratch. This guide covers installation, Skill pack import and invocation, plus a script generation demo with manual revision tips to get you through the first step from script to storyboard.
Can Multi-Token Prediction Be Better? Unlock Soft-MTP with One Line of Code
The standard MTP recipe in frontier pretraining suffers from gradient noise. Researchers propose Soft-MTP, which separates the KL self-distillation signal with just one line of code, improving both speculative decoding and next-token prediction, with code and paper open-sourced.
DeepSeek Harness Installation Guide for Beginners: Temporary vs. Global Install Explained
Complete beginner's guide to installing DeepSeek Harness: set up Node.js first, then choose temporary or global install, with version verification, launch commands, and troubleshooting.
WorkBuddy for Beginners: Office Automation Made Easy from Zero
A beginner's guide to WorkBuddy, the local AI assistant: covering the core content of a 720-minute Bilibili tutorial, including document and spreadsheet automation, data analysis, and browser control, helping zero-experience users quickly get started with AI office automation.