Agents in One Example

Tell a normal chatbot "research three competitors and put the results in a spreadsheet," and it gives you a method or a template. Say the same thing to an agent, and it runs the whole chain itself:

1. Planning — split the goal into search, filtering, comparison, and tabulation;
2. Acting — query search engines, visit web pages, run code;
3. Observing — if information is missing, search again with different keywords;
4. Delivering — produce the spreadsheet, perhaps even email it to you.

You supply only the goal; it figures out the path.

The Four Core Components of an AI Agent

| Component | Role | Analogy | |

Path 1: AI-Assisted Content Creation

Use LLMs for ideas, scripts, and outlines, plus AI image and video tools, on blogs, YouTube, TikTok, or newsletters.

1. Pick a niche you know or will study for years;
2. Let AI draft; you own the point of view, fact-checking, and voice;
3. Revenue comes from ad share, sponsorships, affiliate sales, and paid products.

Fully AI-generated bulk accounts look identical and get throttled — your voice and trust are the moat.

Path 2: Freelance Services Supercharged by AI

Take orders on Fiverr, Upwork, Taobao, or Xianyu and use AI to multiply delivery speed:

• Writing: articles, product descriptions, résumé polishing;
• Design: AI images plus manual retouching for logos, posters, and hero shots;
• Translation: AI draft plus human proofreading, ideally in one niche;
• Coding: tools like Cursor for small websites, scripts, and automation jobs.

Your edge is never "knowing AI" — it is understanding the client and delivering quality.

Path 3: Digital Products, Templates, and Courses

Build once, sell repeatedly: prompt packs, Notion or Excel templates, résumé templates, AI art packs, or short niche courses. Sell through Gumroad, Etsy, Xiaohongshu shops, or Taobao. Validate demand with free content first.

Path 4: AI Operations and Consulting

Many small businesses want AI but cannot use it. Build customer-support knowledge bases (RAG), train staff, set up automation workflows, or run AI-content accounts for clients — typically a one-time setup fee plus a monthly retainer. Go deep in one industry, build a flagship case, then replicate.

Path 5: Private Communities Plus AI Efficiency

If you already run a community or own a customer list, use AI to raise service density: automated Q&A, personalized pushes, batch-generated posts, and support scripts. Serve more clients in the same hours; revenue grows through order value and repeat purchases, not recruiting.

Path 6: E-Commerce Creative Assets

Sellers constantly need product photos with new backgrounds, AI model shots, detail-page copy, short-video scripts, multilingual listings, and chat scripts. Take long-term store outsourcing or sell standardized asset packs. Check platform rules and copyright carefully.

Path 7: Build Vertical AI Tools

If you can code, or use low-code and AI coding tools, build small products that kill a specific pain: contract clause reviewers, title generators, or industry support bots, monetized by subscription or per-use fees. Find a narrow, deep problem people will keep paying to solve.

The Seven Paths Compared

| Path | Barrier to start | Core skill | Time to first cash | |

Why Prompts Matter So Much

An LLM works by predicting the next token from the context it is given. The clearer that context, the less room the model has to guess. Writing a prompt is like briefing a brilliant new hire who knows nothing about your business: you have to explain the background, the goal, the requirements, and the standards.

1. Assign a Role

Giving the model a specific persona activates relevant knowledge and sets the tone:

You are a cross-border e-commerce copywriter with 10 years of experience. Write three punchy English ad headlines for a portable coffee mug.

2. Give Examples (Few-Shot)

Instead of describing the requirement in abstract terms, provide one to three "input to output" examples. The model will imitate the pattern and tone automatically. This is called few-shot prompting, and it works especially well for classification, rewriting, and fixed-template generation. Stating the rule before the examples works even better.

3. Ask for Step-by-Step Reasoning

For reasoning, calculations, or complex decisions, add "think step by step — first show your analysis, then give the conclusion." This noticeably reduces mistakes. For long documents, ask for an outline first and expand only after you approve it.

4. Specify the Output Format

State the exact structure you want and skip the reformatting work:

• Output a Markdown table comparing the three products;
• Return JSON with the fixed fields title, summary, and tags;
• First a 100-word summary, then three actionable recommendations.

5. Set Clear Constraints

| Constraint type | Example | |

How an LLM Works: Next-Token Prediction

Think of an LLM as a world-class "sentence completion" engine that has read enormous amounts of text. When you send a prompt, it does not look up an answer in a database. Instead, it calculates, token by token, what is most likely to come next.

• A token is a unit of text — a word, part of a word, or a punctuation mark.
• The model generates one token, appends it to the input, and repeats until the answer is complete.
• It samples from a probability distribution rather than picking one "correct" word, which is why the same question can be answered differently each time.

The Transformer: Understanding Context

Introduced by Google in 2017, the Transformer architecture is the foundation of every modern LLM. Its key ingredient is self-attention: when processing each word, the model weighs how strongly it relates to every other word in the text.

Simple analogy: in "The dog chased the cat because it was afraid," attention helps the model link "it" to the cat, not the dog. The longer the context, the more this ability matters.

How LLMs Are Trained: Three Stages

| Stage | What happens | Analogy | |

Why RAG Is Needed

A general-purpose LLM is like a widely read consultant who has never seen your company's internal documents:

1. Its knowledge has a cutoff date, so it misses the latest policies and product updates;
2. It has never read your support tickets, contracts, wikis, or notes;
3. When unsure, it tends to invent an answer instead of saying "I don't know."

Retraining the model on all your documents? Expensive, slow to update, and risky for data security. RAG offers a much lighter path.

How RAG Works

Think of RAG as an open-book exam: instead of memorizing everything, the model is allowed to look things up before answering.

| Step | Name | What happens | |

What Is ChatGPT: OpenAI's Flagship Conversational AI

ChatGPT is the conversational AI launched by OpenAI in late 2022, the product that brought generative AI into the mainstream. It runs on the web, as a desktop app, and on mobile, and it combines text chat, voice calls, image recognition, image generation, and file handling. Its positioning is simple: a universal AI assistant for everyone.

Core Capabilities of ChatGPT

• Q&A and tutoring: explain concepts, build study plans, or practice a language with a tutor that is always available.
• Writing and content creation: draft articles, scripts, emails, and marketing copy, then revise them iteratively.
• Data and file analysis: upload spreadsheets or documents to clean data, extract insights, and produce charts.
• Multimodal and agents: voice conversations, image generation, and custom GPTs that handle repeatable workflows.
• Integrations: memory, web browsing, and third-party tool connections, with new features often arriving here first.

Pricing Model: Free Tier and Paid Subscriptions

ChatGPT follows a freemium model. The free tier gives access to a capable base model plus limited access to flagship models. Paid plans (such as Plus, Pro, and Team) add higher flagship-model quotas, priority response, and the full premium feature set. Exact pricing changes over time, so check OpenAI's official site.

| Dimension | Free Plan | Paid Subscription | |

What Is Claude: Anthropic's Safety-First AI Assistant

Claude is built by Anthropic (anthropic.com), the US AI company known for treating safety and reliability as core product features. It is available through the claude.ai website plus desktop and mobile apps, and through APIs for businesses. Rather than chasing the longest feature checklist, Anthropic focuses on long-context understanding, steady output, and controllability.

Core Capabilities of Claude

• Very large context window: you can drop in a book-length manuscript, multiple PDFs, or a full set of project documents and discuss them without the model "forgetting" the early pages.
• Long-form writing and revision: natural prose with a strong sense of structure, suited to reports, proposals, fiction, screenplays, and in-depth articles.
• Coding and Claude Code: especially strong on multi-file edits, refactoring, debugging, and reading unfamiliar codebases; its official coding tool, Claude Code, works directly in your terminal and repository.
• Projects and Artifacts: save long-term project knowledge and instructions for reuse, and preview generated pages, code, and documents on a dedicated canvas.

Pricing Model: Free Tier and Paid Subscriptions

Claude offers a free tier at claude.ai with usage limits. Paid subscriptions, such as Pro and Team, raise those limits and unlock more advanced capabilities, and enterprise plans are available for organizations. Check anthropic.com for current pricing.

| Dimension | Free Plan | Paid Subscription | |

What Is DeepSeek: The Open-Source Chinese LLM

DeepSeek is developed by DeepSeek (Shenzhen DeepSeek Artificial Intelligence), a Chinese company that drew worldwide attention with its open-weight, low-cost strategy. It offers a web app and mobile app, a public API, and downloadable model weights for local or private deployment. The lineup includes general chat models and reasoning-focused models, covering both everyday questions and heavy thinking tasks.

Core Capabilities of DeepSeek

• Chinese fluency: it handles Chinese context, idioms, and local knowledge naturally, which makes it comfortable for formal writing, marketing copy, and summaries.
• Reasoning: the reasoning models visibly think step by step on math, logic, and complex analysis, so the chain of thought is transparent.
• Coding assistance: code generation, completion, debugging, and source-code explanation are solid enough to serve as a free programming partner.
• Open weights and API: weights are public for private deployment, and the API is priced aggressively for batch workloads and product integration.

Pricing Model: Free Web App Plus a Low-Cost API

The web and mobile apps are free after registration. The API is metered by usage and sits among the cheapest in its class. The open-weight models can be downloaded and self-hosted, where your main cost is your own infrastructure. Check deepseek.com for current pricing.

| Dimension | Free Web / Mobile App | API or Self-Hosted Open Weights | |

Why it matters

OneNote has long been one of the best note apps: Microsoft-backed storage, all-device sync, and password-protected sections that most modern note apps dropped. But it never offered Markdown. HydroMark closes that gap without making you abandon OneNote.

Feature tour

Headings and lists

All six # levels render as native OneNote text, including level skips. Ordered, unordered, and deeply nested lists keep clean indentation and render in place.

Task lists, quotes and code

- [ ] becomes a real OneNote to-do checkbox whose state round-trips back to Markdown. > renders as a styled quote block with nested indentation, and fenced code blocks get syntax highlighting for dozens of languages.

Links, tables and text styles

Standard Markdown links become native OneNote hyperlinks; GFM tables with alignment render as proper OneNote tables, not images; and bold, italic, strikethrough and other inline styles all work.

Math, highlights and boxes

Inline $...$ and block $$...$$ formulas render as native OneNote equations — editable and scalable. ==highlight== marks key text, and INFO / WARNING / IMPORTANT / ERROR callout boxes make important notes stand out.

Mermaid and ZenUML diagrams

This is the headline feature. Write Mermaid — flowcharts, sequence, class, state, gantt charts, mindmaps, git graphs — and get a rendered diagram embedded in your page. Press Ctrl + Space on a rendered diagram to jump back to its Markdown source.

Slash command palette

Type / anywhere to open a searchable command palette that instantly inserts headings, lists, code blocks, boxes, formulas, tables or diagrams — the fastest way to write structured notes.

Pricing

HydroMark offers a 14-day full-feature free trial, then splits into a Basic tier with the core Markdown features and a Lifetime tier with all advanced features (LaTeX, diagrams, highlight boxes, themes).

Who should use it

• OneNote power users who want Markdown without leaving their notebooks.
• Students and academics who need editable LaTeX and structured notes.
• Developers and researchers who want Mermaid diagrams embedded inline.
• Anyone who values password-protected notes plus a modern writing experience.

Common questions

Do I have to leave OneNote to write Markdown?

No — everything stays inside OneNote. You write Markdown and watch it render in the same page.

Are diagrams editable after rendering?

You can jump back to the Markdown source with Ctrl + Space, edit, and re-render.

Which OneNote version does it support?

HydroMark targets the Windows desktop OneNote application.

Conclusion

HydroMark is the add-in that finally gives OneNote native-feeling Markdown with live preview, native equations, and complete Mermaid diagrams — at a lower price than the category. If you have ever wished OneNote had Markdown, this is worth the 14-day trial. Visit HydroMark to try it.

Quick comparison

| Feature | OneMark | HydroMark | |

What Is TraeWork: TRAE's AI Office Platform

TraeWork runs on desktop (macOS and Windows), the web, and mobile. Its core loop is "delegate a task in natural language, the AI breaks it down and delivers, and you review and refine." The TRAE family also includes TraeCode, its AI IDE, so the two products cover office workflows and software development together.

Core Capabilities of TraeWork

• Work mode: aimed at product managers, marketers, and analysts, with automated slides, deep research, document writing, and data analysis.
• Code mode: agent-driven code generation, debugging, review, and testing, plus an IDE-style development experience.
• Design mode: visual and design tasks finished inside the same tool.
• Multi-device sync: desktop, web, and mobile share tasks, so you check progress on your phone and review on a computer.
• Natural-language driven: requests are decomposed automatically, with no expert prompt engineering required.

Pricing Model: Credits with a Free Start

TraeWork lets you start for free, with a credits bundle for newly registered users. Different modes and models consume credits, and once those run out you can top up or subscribe under the platform's current rules. Check trae.ai or trae.cn for the latest details.

| Dimension | Free Start | Paid / Subscription | |

Who Is Agibot?

Agibot is a Chinese robotics company co-founded in early 2023 by Peng Zhihui, better known by his online name "Zhihuijun," an engineer who became famous for elaborate DIY tech videos. The company's mission is straightforward:

• Build humanoid robots powered by general embodied AI — software that gives large models a physical body that can move and manipulate objects.
• Go beyond lab demos by emphasizing mass production, manufacturing engineering, and commercial deployment.
• Keep raising capital and expanding production capacity and partnerships.

Why Humanoid Robots Are Leaving the Lab

For years, humanoid robots were caught in the same trap: capable machines were far too expensive, while affordable ones could barely do useful work. Two changes are now breaking that loop.

Foundation Models Meet Factory Floors

• Smarter brains: multimodal models let robots understand spoken instructions, recognize objects, and generalize across tasks instead of following rigid scripts.
• Cheaper bodies: domestically produced actuators, sensors, and dexterous hands keep driving hardware costs down.
• Repeatable manufacturing: leading companies now use automotive-grade supply chains and dedicated production lines rather than hand-building a handful of prototypes.

How the Main Players Compare

| Company / Product | Positioning | Strengths | Who Should Watch It | |

Who the Four Contenders Are

China's large-model market has shifted from comparing parameter counts to competing on experience, price, and ecosystem. Four products dominate the conversation:

• DeepSeek, from DeepSeek AI, broke out with open-weight models, strong reasoning, and aggressive cost efficiency.
• Doubao, from ByteDance, is the consumer-facing assistant available at doubao.com and through its app, backed by ByteDance's content ecosystem.
• Tongyi Qwen, from Alibaba, pairs the widely adopted open-source Qwen model family with the Tongyi app and office-oriented tools.
• Yuanbao, from Tencent, lives at yuanbao.tencent.com and taps into the WeChat ecosystem, with a focus on deep search.

How to Choose at a Glance

| Product | Company | Positioning | Key Strengths | Best For | |

What Gemini Is

Gemini is Google's flagship family of multimodal large language models. "Multimodal" means it does not just read text — it can understand images, audio, video, and code together, and produce text or images in return. Google has been pushing forward on three clear fronts:

• Multimodal input: from reading images to understanding video and live screens.
• Long context: handling much longer documents, codebases, and meeting transcripts in one go.
• Deep integration: connecting Gemini with Google Search, the Gemini app, and Workspace apps such as Gmail, Docs, Sheets, and Meet.

Where the Latest Progress Is Going

Multimodal: From Chatbot to Screen-Aware Assistant

• Upload images, documents, audio clips, or video and ask questions about them.
• In supported scenarios, Gemini can use camera or screen input to explain what it sees in real time, help solve problems, or guide an action.
• Image and voice generation keep improving, making conversations feel closer to face-to-face help.

Long Context: Stop Chopping Up Your Documents

A larger context window means you can hand over a full report, an entire contract, or a big chunk of source code and ask for summaries, differences, or reasoning across the whole thing. That reduces the familiar problem of models ignoring parts of fragmented, shortened inputs.

Search and Workspace Become the Interface

| Capability | What It Looks Like | Typical Use | |

This isn't sci-fi: in the AI era, a single day can hold a year's worth of creation. My "500 years" is the name I give to this time-leverage — using it to learn, to build, and to turn ideas into income.

This blog is the running record of my journey with AI:

• 🤖 Time leverage: handing the repetitive over to agents & workflows
• 💰 Turning time into money: real paths from skills × AI
• 🚀 Era by era: every new model is a new epoch

Once it took a decade to sharpen a single sword; now, with AI, we forge a hundred in a night.

No shortcuts here — just the real story of putting AI's leverage to work, little by little.

This page is an demo that disables the following features:

• Navbar
• Sidebar
• Breadcrumb
• Page information
• Contributors
• Edit link
• Update time
• Prev/Next link
• Comment
• Footer
• Back to top button

• Navbar
• Sidebar
• Footer

Also each page can contain:

• BreadCrumb
• Title and information
• TOC (Table of Contents)
• Meta information including update time and contributors
• Comments

The theme also has the following elements:

• Darkmode button
• Back to top button
• Print button

You can customize them in theme options and page frontmatter.

You should create and write Markdown files, so that VuePress can convert them to different pages according to file structure.

Markdown Introduction

If you are a new learner and don't know how to write Markdown, please read Markdown Intro and Markdown Demo.

Markdown Config

VuePress introduce configuration for each markdown page using Frontmatter.

Markdown Extension

The Markdown content in VuePress will be parsed by markdown-it, which supports syntax extensions via markdown-it plugins.

VuePress Enhancement

To enrich document writing, VuePress has extended Markdown syntax.

For these extensions, please read Markdown extensions in VuePress.

Theme Enhancement

By using VuePress plugins, the theme extends more Markdown syntax and provides richer writing functions.

Code Tabs

::: code-tabs#shell

@tab pnpm

pnpm add -D vuepress-theme-hope

@tab yarn

yarn add -D vuepress-theme-hope

@tab:active npm

npm i -D vuepress-theme-hope

:::

• View Detail

Tabs

::: tabs#fruit

@tab apple

Apple

@tab banana

Banana

@tab orange

Orange

:::

• View Detail

Footnote

This text has footnote[^first].

[^first]: This is footnote content

• View Detail

Include files

• View Detail

TeX

$$ \frac {\partial^r} {\partial \omega^r} \left(\frac {y^{\omega}} {\omega}\right) = \left(\frac {y^{\omega}} {\omega}\right) \left{(\log y)^r + \sum_{i=1}^r \frac {(-1)^i r \cdots (r-i+1) (\log y)^{r-i}} {\omega^i} \right} $$

• View Detail

Tasklist

• [x] Plan A
• [ ] Plan B

View Detail

Image Enhancement

Support setting color scheme and size.

• View Detail

Superscript and Subscript

19^th^ H~2~O

• View Detail

Component

• View Detail

Hint box and GFM alerts

Safely use {{ variable }} in Markdown.

const a = 1;

• GitHub Alert
• Hint boxes

Align

::: center

I am center

:::

::: right

I am right align

:::

• View Detail

Attrs

A word{#word} having id.

• View Detail

Mark

You can mark ==important words== .

• View Detail

Spoiler

VuePress Theme Hope is !!powerful!!.

• View Detail

Stylize

Donate Mr.HydrolinX a cup of coffee. Recommended

• View Detail

Reveal.js

• View Detail

Chart.js

• View Detail

ECharts

• View Detail

Flowchart

• View Detail

MarkMap

• View Detail

Mermaid

• View Detail

PlantUML

@startuml Alice -> Bob: Authentication Request

alt successful case

Bob -> Alice: Authentication Accepted

else some kind of failure

Bob -> Alice: Authentication Failure
group My own label
Alice -> Log : Log attack start
    loop 1000 times
        Alice -> Bob: DNS Attack
    end
Alice -> Log : Log attack end
end

else Another type of failure

Bob -> Alice: Please repeat

end @enduml

• View Detail

Code Demo

• View Detail

Playground

• View Detail

Kotlin Playground

• View Detail

Sandpack Playground

• View Detail

Vue Playground

• View Detail

Page Title

The first H1 title in Markdown will be regarded as page title.

You can also set title in Markdown's Frontmatter: