Tech notes & insights · AI × Full-stack × DevOps · 265+ hands-on writeups
Explore the latest tech trends
先工具、再 MCP、最後 Agent：用 Hermes Agent 打造全地端 AI 助理
我先做了聲音、圖片、影片模型的 API，再把它們包成 MCP，最後才讓 Hermes Agent 去調用。結果意外發現，跑在地端 Qwen 35B 上的 Agent，已經能自己準備素材、訓練另一個機器人的 LoRA，還能生圖、生影片、用聲音跟大家打招呼。這篇分享整套全地端架構、為什麼順序是「先工具、再 MCP、最後 Agent」，以及地端模型目前夠用和不夠用的地方。
Categories
Browse articles by category
The world's leading VPN, independently audited
NordVPN
Best current discount via this site's link
- Encrypts public Wi-Fi traffic
- Watch home streaming abroad
- One account protects 10 devices
This page contains affiliate links: buying through them may earn me a small commission; clicking the link or entering the promo code usually gets you a better price.
Podcast
Mark's Tech Insights — Daily AI News
Daily curated AI and tech trends. Catch the latest developments via audio summaries — covering AI applications, software architecture, DevOps, and engineering practice.
More Posts
Continue exploring more tech content
Building Super Mermaid with Claude Fable: beautifying and exporting VS Code Mermaid diagrams
Notes on building Super Mermaid, a VS Code extension I made with Claude. It addresses two gaps in VS Code's built-in Mermaid support—hard-to-read default styles and a broken export workflow—with automatic coloring and one-click high-resolution export.
Stop Using Regex! Build World-Class International Phone Number Validation with Google libphonenumber 套件
Struggling with the complex rules and regular expressions (Regex) for international phone number validation? This article shows how to use Google's libphonenumber library and easily implement it with React-phone-number-input (frontend) and libphonenumber-csharp (backend). Say goodbye to maintenance nightmares and provide an excellent user experience.
Training Your Own AI Voice: Hardware Requirements, Open-Source Model Comparison, and LoRA Fine-Tuning
My experience training AI voice: hardware differences for training vs. inference, comparing open-source TTS models like Qwen3-TTS and CosyVoice, LoRA fine-tuning implementation, and handling Chinese pronunciation.
AI Product Thinking: The Dual Engine Driving Internal Efficiency and External Growth
In the AI era, merely adopting technology is insufficient. This article explores how to leverage "AI product thinking"—from enhancing internal software engineering efficiency and quality to optimizing external user experience and creating business value—to fully unleash AI's true potential and build winning products.
From Project to Product: The Structural Dilemma of Taiwanese Software?
This article delves into the challenges facing Taiwan's software industry, analyzing core problems such as market size, dependence on government projects, talent drain, and development processes. Under the impact of the AI era, how can Taiwan's software industry transform to find new paths forward and a competitive edge?
Vibe Coding Tips: AI Is an Amplifier — Don't Let It Magnify Your Tech Debt
Vibe Coding looks productive, but with AI assistance it can amplify flaws and leave systems hard to maintain. This article explores why solid software engineering thinking still matters, and shares concrete engineering tips for avoiding isolated information silos and building genuinely robust applications.
AI Works Hard, But It Has No Idea What You're Doing: Practical Reflections from Multi-Agent to Vibe Coding
Introduction: The more I use AI, the more I feel it doesn't really understand what you're doing. This isn't me being pessimistic, but an honest takeaway after getting my hands dirty for a while. When you give it a complex task, the quality of its output is inherently capped by its limited context. It's fast and diligent, but it doesn't "understand" what you're doing. In this post, I want to discuss three things: the essence of Multi-Agent, the hidden aspects of Vibe Coding...
When AI Becomes the #1 Reason for Layoffs: A Calm Observation from an Engineer
In March 2026, AI surpassed "operational restructuring" and "market adjustments" for the first time to become the leading reason for corporate layoffs, accounting for 25% of all announced cuts — about 15,341 positions. As an engineer who uses AI tools daily, I take a calm look at what the data actually says, what's hype, and what it means for our profession.
Three Major Software Development Trends for 2026: The Decline of Manual Coding, Enterprise Adoption Hurdles, and the Reshaping of AI Development and Full-Stack Roles
By 2026, AI-assisted development will be the norm. While code is being produced faster, software quality isn't keeping up. This post covers three phenomena I've observed: the decline of handwritten code, the structural reasons why traditional enterprises are struggling to adopt AI, and the trend of full-stack roles being reshaped.
Multi-Agent in Practice: Why I Chose the Supervisor Pattern to Coordinate AI Agents
An in-depth comparison of the two dominant Multi-Agent architectures (Network vs Supervisor), illustrated with a LangGraph-powered podcast auto-generation pipeline. Covers Supervisor design thinking, feedback loops, and fault-tolerance strategies.
Claude Code Advanced Techniques — GitHub Web Development, Hook Knowledge Base, Remote Control, and Common MCPs
A roundup of practical Claude Code techniques — Memory MCP for cross-session memory, Schedule for automation, Headless Remote Control, GitHub Actions integration, recommended MCPs, and security scanning — to make AI Pair Programming more effective.
From Copilot to Cursor to Claude: Auto-Distilling Knowledge Docs During Development with Claude Code Hooks
A real-world comparison of Claude, Cursor, and GitHub Copilot, plus notes on Gemini 3.1 image generation, the rise of CLI tools, an SQL reasoning failure case, and a Claude Code hook-driven knowledge distillation workflow.
Subscribe to Newsletter
Subscribe to get new posts delivered instantly — never miss a tech share.
By submitting, you agree to receive emails. You can anytime.