Writing
Pruning Your AI Workflow: A TikTok Summarizer Case Computer science calls it pruning. Building a TikTok video summarizer end to end, and using pruning to cut an AI workflow to the bone. 
Nine Principles of Prompt Engineering Nine principles distilled from four prompt engineering whitepapers, turning prompts from one-off conversations into reusable assets. Nano Banana: Gemini Image Naming Decoded Explaining the version numbering logic of Google's Gemini Image series (Nano Banana), and why Nano Banana 2 didn't surpass Nano Banana Pro. Lately
2026.03Lately I've been thinking about the cost of AI coding: the small mistakes agents slip into a codebase, where Skills end and MCP begins, and how to prune a workflow down to the steps that matter.
- Agent
- Tool Usage
- LLM
- Networking
- Programming
- AI
- Prompt
- Mental Models
- Thinking
Seven Tips for Using Claude Code From Plan mode to Sub-fractal Architecture, from MCP retrieval to Agent collaboration—seven battle-tested tips to double your Claude Code efficiency. 
Five Insights from Altman on AI and Entrepreneurship From Altman's live interview: the future of software engineers, GTM bottlenecks, moats for AI products, and the skills that matter most. 
Why Skill Quality Varies So Much Between People Writing a Skill starts with planning. How technical research changes a scraper Skill's stability, speed, and token cost by an order of magnitude. 
Restoring Proxy Speed to Gigabit: A Day of Debugging From picking a Clash distribution to tuning Tun stack mode, tracing a proxy slowdown down to DNS resolution passing through gVisor. X's Official Open-Source Recommendation Algorithm X open-sourced x-algorithm, exposing its recommendation core. A look at the positive and negative weight parameters and the Grok-driven ranking. Budget Typeless DIY: 1/10 the Cost, 85% the Effect An ASR post-processing prompt built on GLM-4.7: strips filler words aggressively and fixes mixed Chinese-English terminology. 
Stop Memorizing Skill Docs — Start Building A practical route into Skill: set the goal, understand the concepts, then learn by building something that actually saves you time. Agency: Dan on the Key Skill for the Next Decade An exploration of Dan's article on why 'agency' is the most critical skill in the AI era, and how to reflect, iterate, and act like an Agent. 
Skill = Agent + Workflow: The Logic Behind It Reviewing Anthropic's "Building effective agents" and understanding why Skill is the fusion of Agent and Workflow. The Boundary Between MCP and Skill MCP provides remote service capability; Skill is a local Agent SOP orchestrator. Knowing the boundary tells you which one to reach for. 
Four Design Patterns Every Powerful AI Agent Needs Four design patterns behind capable AI Agents: Orchestrator-Workers, Pipeline, Circuit Breaker, and State Machine — what each solves. 
Four Elements of an AI Agent: From LLM to Agent LLM + Memory + Planner + Tool Use = Agent. The components, the ReAct pattern, and why one model behaves so differently across products. 
An Underrated Gem: OpenCode + oh-my-opencode A free, open-source AI coding combo that combines all the strengths of Claude Code and AmpCode, even surpassing them in certain aspects. PACER: From Hoarding Knowledge to Digesting It A breakdown of Justin Sung's PACER system: five kinds of knowledge seen through neuroscience, and the learning strategy each one needs. 
I Never Come Up with Prompts for Image Generation A complete workflow for generating images with Gemini: plan first, generate specialized prompts, then iterate on the results. Inside 'Life K-Line': BaZi and Decade Cycles A look at the BaZi theory behind the 'Life K-Line' site, its decade cycle system, and the crypto promotion mixed into its readings. Why TUN Mode Bypasses Geo-Restrictions From OS-level proxies to TUN virtual interfaces: what differs between proxy mode and TUN mode, and why TUN hides itself better. 2026 (16)
Pruning Your AI Workflow: A TikTok Summarizer Case Computer science calls it pruning. Building a TikTok video summarizer end to end, and using pruning to cut an AI workflow to the bone. Nine Principles of Prompt Engineering Nine principles distilled from four prompt engineering whitepapers, turning prompts from one-off conversations into reusable assets. Nano Banana: Gemini Image Naming Decoded Explaining the version numbering logic of Google's Gemini Image series (Nano Banana), and why Nano Banana 2 didn't surpass Nano Banana Pro. Seven Tips for Using Claude Code From Plan mode to Sub-fractal Architecture, from MCP retrieval to Agent collaboration—seven battle-tested tips to double your Claude Code efficiency. Five Insights from Altman on AI and Entrepreneurship From Altman's live interview: the future of software engineers, GTM bottlenecks, moats for AI products, and the skills that matter most. Why Skill Quality Varies So Much Between People Writing a Skill starts with planning. How technical research changes a scraper Skill's stability, speed, and token cost by an order of magnitude. Restoring Proxy Speed to Gigabit: A Day of Debugging From picking a Clash distribution to tuning Tun stack mode, tracing a proxy slowdown down to DNS resolution passing through gVisor. X's Official Open-Source Recommendation Algorithm X open-sourced x-algorithm, exposing its recommendation core. A look at the positive and negative weight parameters and the Grok-driven ranking. Budget Typeless DIY: 1/10 the Cost, 85% the Effect An ASR post-processing prompt built on GLM-4.7: strips filler words aggressively and fixes mixed Chinese-English terminology. Stop Memorizing Skill Docs — Start Building A practical route into Skill: set the goal, understand the concepts, then learn by building something that actually saves you time. Agency: Dan on the Key Skill for the Next Decade An exploration of Dan's article on why 'agency' is the most critical skill in the AI era, and how to reflect, iterate, and act like an Agent. Skill = Agent + Workflow: The Logic Behind It Reviewing Anthropic's "Building effective agents" and understanding why Skill is the fusion of Agent and Workflow. The Boundary Between MCP and Skill MCP provides remote service capability; Skill is a local Agent SOP orchestrator. Knowing the boundary tells you which one to reach for. Four Design Patterns Every Powerful AI Agent Needs Four design patterns behind capable AI Agents: Orchestrator-Workers, Pipeline, Circuit Breaker, and State Machine — what each solves. Four Elements of an AI Agent: From LLM to Agent LLM + Memory + Planner + Tool Use = Agent. The components, the ReAct pattern, and why one model behaves so differently across products. An Underrated Gem: OpenCode + oh-my-opencode A free, open-source AI coding combo that combines all the strengths of Claude Code and AmpCode, even surpassing them in certain aspects.
2025 (4)
PACER: From Hoarding Knowledge to Digesting It A breakdown of Justin Sung's PACER system: five kinds of knowledge seen through neuroscience, and the learning strategy each one needs. I Never Come Up with Prompts for Image Generation A complete workflow for generating images with Gemini: plan first, generate specialized prompts, then iterate on the results. Inside 'Life K-Line': BaZi and Decade Cycles A look at the BaZi theory behind the 'Life K-Line' site, its decade cycle system, and the crypto promotion mixed into its readings. Why TUN Mode Bypasses Geo-Restrictions From OS-level proxies to TUN virtual interfaces: what differs between proxy mode and TUN mode, and why TUN hides itself better.