Actual Intelligence · AI 能力指南 Actual Intelligence · A Practical Guide to AI 2026

让 AI 在你的认知中
有形状

Give AI a
shape in your mind

大多数人对 AI 的感受还是一团雾。这份能力指南帮你把它拆清楚——知道它能做什么、在哪些工作里能帮你、怎么开始自己动手。 For most people, AI still feels like a fog. This guide helps you cut through it — what it can do, where it fits into your work, and how to start building with it.

1 Know

先看清 AI 在做什么

First, See What AI Actually Does

一张图给所有 AI 工具定位,一个核心视角看清选工具的逻辑,一把尺子筛选任何新工具。走完这节,你就有了一套稳定的判断。

One chart to place every AI tool, one core lens for choosing between them, one ruler to evaluate anything new. Finish this section and you have a stable way to judge.

AI 工具定位图
The AI Tool Map
行 = 工作领域。列 = 使用模式(Chat / Agent / Workflow / Vibe Coding)。重点不是记住每个工具,是练出"一眼看出它属于哪一格"的判断力。
Rows = domain of work. Columns = mode of use (Chat / Agent / Workflow / Vibe Coding). The point isn't memorizing every tool — it's training your eye to place any new one at a glance.
附录 · 四个列标题,逐个讲清 Appendix · The Four Columns, Defined
💬
Chat
你问,AI 答。
You ask, AI answers.
在 ChatGPT / Claude 里打"帮我写封道歉邮件"——它给你写出来。最常见的用法。
Type "write me an apology email" into ChatGPT / Claude. The most common use.
变体 · Copilot:AI 嵌在你已经在用的工具里(Notion / Slack / Figma),你干活时它在边上帮你。
Variant · Copilot: AI embedded inside a tool you're already using (Notion / Slack / Figma), assisting as you work.
🤖
Agent
你说终点,AI 自己开车。
You name the destination, AI drives.
你说"调研这家公司的业务模式和最新动态"——AI 自己上网、打开几十个页面、综合信息、给你结论。你不管中间怎么走。
Say "research this company's business model and recent news" — AI hits the web, opens dozens of pages, synthesizes, reports back. You don't manage the middle.
⚙️
Workflow
拖几个方块,AI 自动跑。
Drag a few boxes, AI runs on its own.
收到客户邮件 → AI 自动分类 → 重要的通知你,普通的自动回复。像搭乐高,不用写代码。代表:n8n、Zapier。
Email arrives → AI classifies → important ones ping you, routine ones auto-reply. Like LEGO, not code. Think n8n, Zapier.
Vibe Coding
你用说话造软件。
You build software by talking.
你说"做一个表,团队填周报,周五自动汇总发我邮箱"——AI 真的给你做出来。不用会编程。代表:Claude Artifacts、Cursor。
"Build me a form for weekly reports that auto-summarizes to my inbox Friday" — AI actually makes it. No coding needed. Think Claude Artifacts, Cursor.
领域↓ 模式→Domain↓ Mode→
💬
Chat
你说,AI回应
You speak, AI replies
🤖
Agent
你说目标,AI自主执行
You set the goal, AI runs it
⚙️
Workflow
拖拽搭自动化
Drag-drop automation
Vibe Coding
自然语言造东西
Build with plain language
🧠思考 & 文字Thinking & Text
ClaudeChatGPTGeminiPerplexityKimi
CoworkNotion AIManus
n8nMakeZapier
CursorClaude Code
🎨图像 & 设计Image & Design
MidjourneyDALL·EFluxIdeogramNanobanana
Canva AIFigma AIAdobe Firefly
ComfyUIn8n+API
Framer AICursor+Canvas
🎬视频 & 动态Video & Motion
KlingRunwaySeedanceSoraVeoPika
HeyGenDescriptRunway Act-One
RemotionDify+API
Cursor+Remotion
🎵音频 & 语音Audio & Voice
SunoUdioElevenLabsNotebookLM
DescriptAdobe Podcast
n8n+ElevenLabs
正在涌现中
Still emerging
📊数据 & 分析Data & Analytics
ChatGPT DataClaude ArtifactsJulius AI
CoworkRows AI
n8n+DBRetool AI
Cursor+PythonReplit
怎么读这张图:先定位两件事——你的工作属于哪个领域(行),你现在停在哪种用法(列)。下一步不一定是往右挪一格——Workflow 和 Vibe Coding 是两条并行路径,很多人会直接从 Chat/Agent 跨到 Vibe Coding。 How to read this chart: Start by locating two things — which domain your work sits in (rows), and which mode you're currently using (columns). The next step isn't always one cell to the right — Workflow and Vibe Coding are parallel paths, and many people leap straight from Chat/Agent into Vibe Coding.
💡 选工具的核心视角:一个只会打字的 AI,和一个能读你 Slack、查你 Drive、帮你发邮件的 AI,是两个完全不同的东西。不要只看"聊天有多聪明"——优先看它能连接什么。MCP 生态越丰富的平台,AI 的实际价值越高。就像选手机看 App Store,平台是载体,连接才是上限。 💡 The lens that matters: An AI that only types and an AI that reads your Slack, searches your Drive, and sends your emails are two completely different things. Don't just ask "how smart is the chat?" — ask what it can connect to. The richer the MCP ecosystem, the higher the AI's real ceiling. Like choosing a phone for its App Store — the device is just the vessel.
一把尺子
THE RULER
遇到任何新 AI 工具,问自己五件事
Five Things to Ask of Any New AI Tool

不是给你"要不要买"的答案——是给你一把能反复使用的尺子。每条都配一对信号:过关的样子,警惕的样子。

Not a "should you buy it" verdict — a ruler you can use again and again. Each one has two signals: what passing looks like, what caution looks like.

  1. 01
    它在图里落在哪一格?
    Where does it land on the chart?
    过关PASS 它落进的是你还空着的那一格——真正补位。It lands in a cell you haven't covered yet — genuinely filling a gap.
    警惕FLAG 那一格你早有趁手的老工具,这个只是换了张皮。You already have a solid tool in that cell; this one's just reskinned.
  2. 02
    它能嵌进你已有的工作流吗?
    Can it slot into your existing workflow?
    过关PASS API、MCP、Webhook——能对话,能被调用。Has API, MCP, webhooks — it can speak and be called.
    警惕FLAG 只能关在自己的围墙里用,导出也困难——花园再漂亮也是花园。Lives inside its own walls, hard to export — pretty garden, still a cage.
  3. 03
    从安装到出效果,要多久?
    From install to first real value — how long?
    过关PASS 10 分钟内看到价值。真该用的工具都应该这样。Under 10 minutes to something valuable. Real tools meet you where you are.
    警惕FLAG 要学 3 天教程才能跑起来——沉没成本会把你留在不该留的地方。Takes 3-day tutorials to run — sunk cost keeps you where you shouldn't be.
  4. 04
    它的护城河在哪?
    Where is its moat?
    过关PASS 独家数据、生态、垂直深度——大厂明年也做不出来。Has proprietary data, ecosystem, or vertical depth — the incumbents can't ship this next year.
    警惕FLAG 功能明天 ChatGPT/Claude 原生就能做——GPT-Wrapper 的命运。What it does is something ChatGPT/Claude will natively ship tomorrow — the GPT-wrapper fate.
  5. 05
    你的数据去了哪里?
    Where does your data go?
    过关PASS API 或企业版——数据合同隔离,不进训练池。API or enterprise tier — contractually isolated, not in any training pool.
    警惕FLAG 免费版 = 你是训练材料。处理机密信息时,这条是硬红线。Free tier = you are training material. For anything confidential, this is a hard line.
基础概念 · 了解的可以跳过 Foundations · Skip if you know this
跳到 Use → Skip to Use →
💬 大语言模型(LLM)— Chat 的引擎
核心:根据前文,预测下一个最合理的词。它读过海量文字,学会了语言的模式。所以它能写、能翻译、能分析——但它不是在"思考",是在做极其高级的模式匹配。
→ 这就是为什么它偶尔"一本正经地胡说八道"——它追求的是"听起来合理",不是"事实正确"。
🎨 图像生成 — Midjourney / DALL·E / Flux
扩散模型:从一张全是噪点的图开始,一步步"去噪"变成清晰图片。你的 prompt 就是去噪的方向。
→ 类比:雕塑家面前的大理石,你的 prompt 是设计图。
🎬 视频生成 — Kling / Sora / Seedance
图像扩散的升级版——不只生成一帧,而是同时生成时间维度上连贯的多帧。需要理解物理规律和运动连续性。
→ 目前:静态/慢动作很惊艳,复杂运动和多物体交互仍是弱项,但进步飞快。
💻 AI Coding — Cursor / Claude Code
本质还是 LLM——因为代码也是语言,而且比人话更有规律,AI 学起来更容易。AI Coding 工具在此基础上加了文件系统访问、终端运行、编辑器集成。
→ Vibe Coding:你是产品经理,AI 是程序员。你描述需求+验收,它写代码+改 bug。
🤖 Agent — 不是新模型,是新用法
Chat = 你说一句它答一句。Agent = 你说目标,它自己规划步骤、调用工具、执行到底
→ "Agentic"是形容词:描述"能自主规划+执行"的系统。不是新技术,是设计模式。
💬 Large Language Models (LLM) — the engine behind Chat
Core: given what came before, predict the most plausible next word. It has read enormous amounts of text and learned the patterns of language.
→ This is why it sometimes "confidently makes things up." It optimizes for sounding plausible, not being correct.
🎨 Image generation — Midjourney / DALL·E / Flux
Powered by diffusion models: starts from pure noise and "denoises" step by step into a clear image. Your prompt is the direction.
→ A sculptor facing a block of marble. Your prompt is the design sketch.
🎬 Video generation — Kling / Sora / Seedance
An upgrade of image diffusion — generates multiple frames coherent across time. Requires a sense of physics and motion.
→ Stills are stunning. Complex motion is still weak — but improving fast.
💻 AI Coding — Cursor / Claude Code
Still an LLM — because code is also a language, and one with more regular patterns. AI Coding tools add filesystem access, terminal, and editor integration.
→ Vibe Coding: you're the PM, the AI is the engineer.
🤖 Agent — not a new model, a new way of using one
Chat = ask and answer. Agent = state a goal, it plans steps, calls tools, executes end-to-end.
→ "Agentic" is an adjective — a design pattern, not a new technology.
你天天会听到的几个词。
Prompt
跟 AI 沟通的方式。不是魔法咒语。核心:说清楚你是谁、要什么、背景、格式。
Token & Context Window
Token ≈ 1 英文词 ≈ 0.7 中文字。Context Window = AI 一次能看多少内容。
AI 的工作桌面——桌子越大能摊越多文件。
LLM / 大语言模型
AI 的引擎。Claude / GPT / Gemini 都是 LLM。日常说"模型",指的就是它。
Hallucination / 幻觉
AI 一本正经地胡说。看着可信,但事实是错的。重要事项必须自己核对。
用 AI 的几种姿势——直接聊、嵌入工具、还是接进系统。
API vs 消费者版
消费者版(claude.ai)= 直接用,月费。API = 程序员接口,按量付费,可嵌入自己系统。
消费者版 = 去餐厅吃。API = 叫大厨来你家做菜。
Copilot
AI 嵌入你正在用的工具里,实时辅助。你主导,它建议。
副驾驶——你开车,它帮你导航和看盲区。
System Prompt / 系统提示
给 AI 设定的"人设"和规则,藏在对话最前面。决定它怎么回应你。
Temperature / 温度
调节输出的随机性。低 = 稳定可预测;高 = 有创意有变化。
让 AI 真正"干活"的关键词。
Agent / 智能体
能自己拆任务、调工具、多步执行的 AI。不只回答,还会"做"。
从顾问升级成实习生——给个目标,它自己跑流程。
RAG / 检索增强
先翻资料 → 再让 AI 基于资料回答。让 AI 用上你私有的知识库。
开卷考试——先翻书再答题。
Tool Use / 工具调用
AI 主动调用外部工具完成任务:搜索、计算、查数据库、发邮件。
MCP / Model Context Protocol
AI 连接外部工具和数据源的统一标准。装上 MCP,Claude 就能读你的 Notion、操作浏览器、查数据库。
USB-C——一个口接所有设备。
Multimodal / 多模态
AI 同时处理文字 + 图片 + 音频 + 视频。不只看文字了。
Reasoning / 推理 (CoT)
让 AI 在回答前先"想"一遍——分步骤、列假设、自检。复杂问题准确率显著提升。
造 AI 产品的人在聊的。
Fine-tuning / 微调
用专属数据再训练一次模型,让它更懂你的领域。成本高,多数场景 RAG 就够。
Harness Engineering / 工程外壳
模型只是引擎。让它在真实场景里好用的是周边那一层——工具、上下文、提示词、评估、权限。Claude Code 这类 agent 工具,本质就是给模型套一个 harness。
F1 引擎再强,没车架、没刹车也跑不起来。Harness 就是车身。
The terms you'll hear every single day.
Prompt
How you talk to an AI. Not a magic incantation. Core: say who you are, what you want, the context, the format.
Token & Context Window
A token ≈ one English word ≈ 0.7 Chinese characters. Context Window = how much the AI reads at once.
The AI's desktop — bigger desk, more files spread out.
LLM / Large Language Model
The engine of AI. Claude / GPT / Gemini are all LLMs. When people say "the model", this is what they mean.
Hallucination
The AI confidently making things up. Sounds credible, but the facts are wrong. Always verify what matters.
Different ways to use AI — chat directly, embed it, or wire it into a system.
API vs Consumer tier
Consumer tier (claude.ai) = use directly, pay monthly. API = developer interface, pay-as-you-go, embeddable.
Consumer = going to a restaurant. API = hiring the chef to cook in your kitchen.
Copilot
AI embedded inside the tool you're using, assisting in real time. You drive, it suggests.
A co-pilot — you drive, it handles navigation and blind spots.
System Prompt
The "persona" and rules set up before the conversation starts. Shapes how the AI responds to you.
Temperature
Controls output randomness. Low = stable and predictable; high = creative and varied.
The terms that show up when AI starts actually doing work.
Agent
AI that breaks down tasks, calls tools, and runs multi-step workflows on its own. Doesn't just answer — it does.
From consultant to intern — give it a goal, it runs the workflow.
RAG / Retrieval-Augmented
Retrieve relevant docs first, then let the AI answer based on them. Lets AI use your private knowledge base.
Open-book exam — flip through, then answer.
Tool Use / Function Calling
The AI actively calls external tools to get things done — search, compute, query databases, send email.
MCP / Model Context Protocol
An open standard for connecting AI to external tools and data. With MCP, Claude can read your Notion, drive a browser, query a database.
USB-C — one port for every device.
Multimodal
AI handling text + images + audio + video at the same time. Not just text anymore.
Reasoning (Chain-of-Thought)
The AI "thinks" before answering — breaks down steps, lists assumptions, self-checks. Big accuracy gains on hard problems.
What the people building AI products are actually talking about.
Fine-tuning
Re-train a model on your own data so it knows your domain. Expensive — for most use cases RAG is enough.
Harness Engineering
The model is just the engine. What makes it useful in production is the surrounding scaffolding — tools, context, prompts, evals, permissions. Agent tools like Claude Code are essentially harnesses around the model.
An F1 engine with no chassis or brakes goes nowhere. The harness is the car.
Chat / 对话场景
① Prompt 质量 — 影响力最大
你给的指令越清楚,输出越好。这是你唯一完全可控的变量。从40分到80分,往往只需要把一句话改成一段话。
② 模型推理能力 — 天花板
Claude Opus > Sonnet > Haiku。简单任务差别不大;复杂任务模型差异决定了结果能不能用
③ 上下文 — 你喂了多少信息
AI 只能基于你给的信息工作。Context Window = 一次能喂多少,主流 100K-200K tokens ≈ 一本书。
④ 迭代次数
2-3 轮迭代 > 花 20 分钟写"完美 prompt"
Agent 额外变量
可用工具 + 规划能力 + 纠错能力
Agent 的能力上限 = 它能调用什么工具。规划+纠错是最薄弱也进步最快的方向。
创意生成
模型版本 > Prompt精度 > 参考图 > 参数
描述越具体越好。参考图比纯文字更精准。
Chat / Conversation
① Prompt quality — biggest lever
Clearer instruction = better output. This is the only variable you fully control.
② Model reasoning — the ceiling
Claude Opus > Sonnet > Haiku. For complex tasks, the model difference is the difference between usable and not.
③ Context — how much you feed it
Context Window = how much it can take in. Mainstream: 100K–200K tokens ≈ a full book.
④ Iteration
2–3 iterations beats 20 minutes of writing the "perfect prompt."
Agent — extra variables
Available tools + Planning + Self-correction
An Agent's ceiling = the tools it can call. Planning + reflection is the weakest and fastest-improving area.
Creative generation
Model version > prompt precision > reference images > parameters
The more specific the description, the better. Reference images beat words alone.
☁️ 云端 vs 🏠 本地
云端:大多数 AI 工具的运行方式。不需要好电脑,随时可用。数据经过别人的服务器。
本地:模型下载到自己的机器上跑,数据不出门。需要好硬件(GPU),通常不如云端最新。
🔓 开源 vs 🔒 闭源
开源(Llama, Stable Diffusion, Mistral):免费、可定制、可本地部署。需要技术能力。
闭源(Claude, GPT-4, Gemini):性能最强、持续更新。供应商锁定+定价权在对方。很多公司同时用两种。
🛡️ 数据安全三问
① 数据会被用来训练吗?免费版通常会,API版不会,企业版有合同保护。
② 存储在哪?大多数服务器在美国。受 GDPR/数据安全法约束的业务要确认合规。
③ 员工怎么管?最大风险是员工把机密粘贴进免费 ChatGPT。一页 A4 纸的使用准则就够
☁️ Cloud vs 🏠 Local
Cloud: how most AI tools run. No powerful machine needed, always available. Data passes through someone else's servers.
Local: download and run on your machine. Data never leaves. Requires a GPU, usually behind the frontier.
🔓 Open source vs 🔒 Closed source
Open source (Llama, Stable Diffusion, Mistral): free, customizable. Requires technical skill.
Closed source (Claude, GPT-4, Gemini): strongest performance, constantly updated. Vendor lock-in. Many companies use both.
🛡️ Three data-security questions
① Training? Free tiers usually yes, API tiers no, enterprise contractually protected.
② Where stored? Most servers in the US. Check compliance for GDPR-bound businesses.
③ Employee management? Biggest risk: staff pasting confidential material into free ChatGPT. A single A4 page of usage rules is enough.

2 Use

合上指南,动手试一遍

Close the Guide. Try Something.

两种进入方式——从你的职业切入,或者从你要做的事切入。再往下,是七件具体的事,每件都给你三个档位。

Two ways in — start from what you do, or from what you want done. Below that, seven concrete moves, each at three depths.

A从职业走进去Enter by profession
你的工作,AI 能从哪里插手?
Where can AI slot into your work?
输入你的职业——30 秒看到它在你工作里的三个入口:认识、使用、构建。
Type your profession — see three entry points in your work in 30 seconds: know, use, build.
医生Doctor 教师Teacher 律师Lawyer 教练Coach 会计Finance 销售Sales HR 记者Media
B从目标走进去Enter by goal
你想做的事,AI 有哪条路?
What path does AI open for your goal?
说出你要做的事——回你三个档位:今天就能做 · 认真搞 · 系统化。
Say what you want done — get three depths back: do today · get serious · systematize.
提高员工效率Team efficiency 客户提案Client proposals 批量内容Batch content 数据报告Data reports 图片视频Images & video 自动化Automate 造个小工具Build a tool 学新领域Learn fast
具体场景PLAYBOOKS
七件具体的事,三个档位
Seven Concrete Moves, at Three Depths
每条都可以抄。绿色:今天就能做的最短路径;蓝色:认真搞的组合;橙色:系统化的长线。
Each one is copyable. Green: the shortest path you can take today; Blue: the serious combination; Orange: the systematic long game.
示例 Prompt我在给 [品牌名] 做 [项目类型] 提案。行业 [行业],目标 [目标]。请:
1. 分析3个行业优秀案例
2. 提炼3个策略方向
3. 每个方向一句话洞察+一句话执行路径
4. 推荐你最看好的方向并说明为什么
⚡ 今天就能做
Claude + 手动搜索
直接在 Claude 对话里完成:粘贴客户背景 → 让它出策略方向 → 让它写每页文案。
Claude
10分钟上手 · 免费起步
🔧 认真搞一下
Perplexity研究 + Claude分析 + Notion整理
Perplexity 搜实时数据和竞品 → Claude 做深度分析和文案 → Notion 结构化管理。
PerplexityClaudeNotion
1天上手 · $20-40/月
🚀 系统化
Cowork Agent + 模板化流水线
用 Cowork 建提案 Agent:自己搜集资料、生成分析、套模板出完整提案文档。
CoworkClaude CodeNotion API
1周搭建 · 长期复利
Sample PromptI'm pitching [project type] for [brand]. Industry: [industry]. Goal: [goal]. Please:
1. Analyze 3 strong industry case studies
2. Distill 3 strategic directions
3. For each: one-line insight + one-line execution path
4. Recommend your favorite and why
⚡ Do it today
Claude + manual search
All inside a Claude conversation: paste client background → strategic directions → write each page.
Claude
10 min to start · free
🔧 Get serious
Perplexity + Claude + Notion
Perplexity for real-time data → Claude for deep analysis and copy → Notion to structure everything.
PerplexityClaudeNotion
1 day · $20–40/mo
🚀 Systematize
Cowork Agent + templated pipeline
Build a proposal Agent: it gathers research, runs analysis, outputs a full proposal doc.
CoworkClaude CodeNotion API
1 week · compounds
示例 Prompt附上我写的文章。改写成4版:
1. 小红书帖:500字内,口语化,钩子标题
2. LinkedIn帖:英文,专业,300词
3. Newsletter摘要:3段
4. 60秒短视频脚本:口播,前3秒悬念
⚡ 今天就能做
Claude 一次改写
把原文粘贴给 Claude,一次性让它出多个平台版本。
Claude
5分钟 · 免费
🔧 认真搞一下
Claude改写 + Midjourney配图
Claude 出文案,Midjourney/Nanobanana 生成配图,Notion 内容日历跟踪。
ClaudeMidjourneyNotion
半天上手 · $30/月
🚀 系统化
自动化流水线
n8n 搭可视化流水线 或 Cursor Vibe Coding 造定制发布工具。"你只管写原文"。
n8nCursorClaude API
1-2周搭建 · 完全自动
Sample PromptHere's an article I wrote. Rewrite it 4 ways:
1. XHS post: under 500 chars, conversational, hook headline
2. LinkedIn post: English, professional, 300 words
3. Newsletter summary: 3 paragraphs
4. 60-sec video script: voiceover, hook first 3 seconds
⚡ Do it today
One-shot Claude rewrite
Paste original, ask for all platform versions in one go.
Claude
5 min · free
🔧 Get serious
Claude + Midjourney visuals
Claude for copy, Midjourney for visuals, Notion content calendar.
ClaudeMidjourneyNotion
Half day · $30/mo
🚀 Systematize
Automated pipeline
n8n visual pipeline or Cursor Vibe Coding custom tool. "You just write the original."
n8nCursorClaude API
1–2 weeks · fully automatic
示例 Prompt我是 [行业] 的 [职位]。评估是否 [决策]。请:
1. 列5个关键维度
2. 每个给 Pro/Con(要数据支撑)
3. 综合推荐,信心1-10,最大不确定因素
⚡ 今天就能做
Claude 直接问
把决策背景详细描述给 Claude,让它搭框架、列 Pro/Con。
Claude
10分钟 · 免费
🔧 认真搞一下
Perplexity搜数据 + Claude深度分析
Perplexity 搜最新市场数据 → 喂给 Claude 做结构化分析 → NotebookLM 多文档综合。
PerplexityClaudeNotebookLM
1-2小时 · 更扎实
🚀 系统化
定制决策 Agent + 数据接入
连接公司内部数据+外部市场数据的分析 Agent。每周自动推行业变化简报。
Cowork+MCP内部数据源
2-4周 · 长期战略工具
Sample PromptI'm a [role] in [industry]. Evaluating whether to [decision]. Please:
1. List 5 critical dimensions
2. Pro/Con for each, backed by data
3. Final recommendation, confidence 1–10, biggest unknown
⚡ Do it today
Just ask Claude
Describe the decision in detail, ask it to build a framework and list Pros/Cons.
Claude
10 min · free
🔧 Get serious
Perplexity + Claude + NotebookLM
Perplexity for data → Claude for structured analysis → NotebookLM for multi-doc synthesis.
PerplexityClaudeNotebookLM
1–2 hours · solid
🚀 Systematize
Custom decision Agent + data
Analysis Agent connected to internal + external data. Weekly auto-briefings on industry shifts.
Cowork+MCPdata sources
2–4 weeks · strategic
⚡ 今天就能做
录音 + Claude 手动处理
手机录音 → Otter/Granola 转文字 → 粘贴给 Claude 提炼要点和写邮件。
OtterClaude
5分钟/次
🔧 认真搞一下
Granola 自动总结 + Notion
Granola 边开会边自动生成 AI 总结 → 导入 Notion → 分配 action items。
GranolaNotion
1天配好 · $10/月
🚀 系统化
全自动:录音→总结→任务→提醒
Granola 转写 → n8n 自动调 Claude → Asana/Notion 创建任务 → Slack 提醒。零手动。
Granolan8nClaude APISlack
2周搭建 · 每次会议省20分钟
⚡ Do it today
Recording + Claude (manual)
Record → transcribe with Otter/Granola → paste into Claude to extract takeaways.
OtterClaude
5 min per meeting
🔧 Get serious
Granola auto-summary + Notion
Granola auto-summarizes in real time → import to Notion → assign action items.
GranolaNotion
1 day · $10/mo
🚀 Systematize
Full auto: recording → tasks → reminders
Granola → n8n → Claude → auto-create tasks → Slack reminders. Zero manual.
Granolan8nClaude APISlack
2 weeks · saves 20 min/meeting
⚡ 今天就能做
Midjourney / Nanobanana
写描述 → 生成图片。先用 Claude 帮翻英文 prompt 效果更好。
MidjourneyNanobanana
5分钟/张 · $10-30/月
🔧 认真搞一下
图片+视频组合
Midjourney 出静态图 → Kling/Seedance 变动态 → Canva 排版。
MidjourneyKlingSeedanceCanva
1小时/套素材
🚀 系统化
ComfyUI 批量流水线
固定品牌风格 → 批量输入 → 自动生成一致风格系列图。适合电商、MCN。
ComfyUIStable Diffusionn8n
1-2周 · 批量复利
⚡ Do it today
Midjourney / Nanobanana
Write a description → generate. Claude helps translate to polished English prompts.
MidjourneyNanobanana
5 min/image · $10–30/mo
🔧 Get serious
Image + video combo
Midjourney stills → Kling/Seedance to animate → Canva layout.
MidjourneyKlingSeedanceCanva
1 hour per set
🚀 Systematize
ComfyUI batch pipeline
Lock brand style → batch content → auto-generate consistent series.
ComfyUIStable Diffusionn8n
1–2 weeks · scales
⚡ 今天就能做
Claude Artifacts
在 Claude 对话里直接说"帮我做一个XX",秒出可预览的组件。
Claude
5分钟 · 免费
🔧 认真搞一下
Cursor 本地开发
多页面网站、有数据库的应用。AI 帮你写代码,你负责描述需求和验收。
CursorVercel
1-3天 · $20/月
🚀 系统化
Cowork + Claude Code 全栈
造有用户系统、数据库、API 的完整产品。然后 Vercel/Railway 部署。
CoworkClaude CodeVercel
1-2周 · 真产品
⚡ Do it today
Claude Artifacts
Say "build me a [thing]" — instant previewable component.
Claude
5 min · free
🔧 Get serious
Cursor local dev
Multi-page sites, apps with a database. AI writes code, you review.
CursorVercel
1–3 days · $20/mo
🚀 Systematize
Cowork + Claude Code full stack
Real product with users, database, API. Deploy to Vercel/Railway.
CoworkClaude CodeVercel
1–2 weeks · real product
⚡ 今天就能做
Claude 问答
"我需要快速了解 [领域],我的背景是 [背景]。给我知识地图+5个核心概念+5个显得懂行的问题。"
Claude
30分钟 · 免费
🔧 认真搞一下
Perplexity + Claude + NotebookLM
Perplexity 搜最新报告 → Claude 搭知识框架 → NotebookLM 生成播客式摘要。
PerplexityClaudeNotebookLM
2-3小时 · 扎实
🚀 系统化
个人知识库 Agent
连接 Notion + Drive 的知识 Agent:自动收集 → 每周推简报 → 随时可问。
Cowork+MCPNotionn8n
1周搭建 · 持续复利
⚡ Do it today
Just ask Claude
"I need to understand [field] fast. Background: [X]. Give me a knowledge map + 5 core concepts + 5 smart-sounding questions."
Claude
30 min · free
🔧 Get serious
Perplexity + Claude + NotebookLM
Perplexity for reports → Claude builds framework → NotebookLM for podcast-style summary.
PerplexityClaudeNotebookLM
2–3 hours · solid
🚀 Systematize
Knowledge-base Agent
Agent connected to Notion + Drive: auto-collects → weekly briefings → ask anything.
Cowork+MCPNotionn8n
1 week · compounds

3 Build

把 AI 嵌进你的工作里

Build AI Into the Work Itself

Use 是个人的提效工具。Build 是让 AI 从"有人在用"变成"这就是我们的工作方式"。这道槛,大多数团队还没有主动跨过。先知道你在哪,再知道下一步往哪走。

Use makes individuals more efficient. Build is what turns "some people use AI" into "this is how we work." Most teams haven't crossed that threshold yet. Know where you are first — then find your next move.

你的组织在哪个阶段? Where is your organization?
Level 1
个人探索
Exploration
几个人在试 ChatGPT / Claude,没有统一做法,没人知道别人在用什么。AI 是有趣的补充,不是工作方式。
A few people trying ChatGPT / Claude. No shared playbook. No one knows what anyone else is using. AI is an interesting supplement, not a way of working.
你的下一步
Your next move
选 1 个高频痛点跑通,开一次分享会。
Pick 1 pain point. Prove it. Run a team share-out.
Level 2
场景落地
Adoption
3–5 个场景在用 AI,部分人形成习惯,但各搞各的,没有共同语言,好做法没有沉淀。
3–5 use cases active. Some habits formed. Everyone does it differently. No shared language, no institutional knowledge built up.
你的下一步
Your next move
建 Prompt 库,跑一条 Workflow,让好做法可以复制。
Build a Prompt library. Run one Workflow. Make good practice repeatable.
Level 3
流程改造
Integration
AI 接入业务系统,有自动化流水线,有人负责,ROI 可以被量化。AI 不再只是辅助——它是流程本身。
AI wired into systems. Automation pipelines live. Someone owns it. ROI measurable. AI isn't just support anymore — it's part of the process itself.
你的下一步
Your next move
量化 ROI,找到你们的数据优势,构建别人复制不了的能力。
Quantify ROI. Find your data advantage. Build capability others can't copy.
Level 4
AI-Native
AI-Native
AI 融入核心业务和产品,有战略,有评估体系。越用越好——数据积累变成护城河,差距每天都在扩大。
AI is core to business and product. Strategy and evaluation in place. Compounds over time — data becomes a moat, the gap widens every day.
你在做的事
What you're doing
数据飞轮转起来,用差异化竞争,持续迭代。
Data flywheel spinning. Competing on differentiation. Compounding the lead.
接下来NEXT
知道站在哪了——怎么往前走?
You know where you stand. Now how do you move?
四个工具,帮你把想法变成行动。不用按顺序——跳到你现在最需要的那个。
Four tools to turn intent into action. No fixed order — jump to what you need most right now.
好的 AI 场景 = 重复性高 + 判断标准清晰 + 出错成本可控
⚡ Quick Win
今天就能省时间
邮件处理 · 会议纪要 · 内容初稿 · 数据报表 · 文档摘要
1天上手 · 节省30-60%时间
🔧 中期改造
1-4周配置,ROI清晰
客服AI响应 · 内部知识库问答 · 销售线索评分 · 多平台内容分发
需要一些配置 · 可衡量
🚀 竞争壁垒
拉开差距的长期布局
个性化推荐 · 供应链预测 · 定制Agent · AI-native产品功能
需要数据+技术投入
不适合 AI 的(至少现在)
高度人际判断(招聘终面)· 强创造性核心(品牌创意方向)· 合规高压区(医疗诊断、法律意见)· 出错=灾难(财务审批)
A good AI use case = highly repetitive + clear criteria + manageable error cost
⚡ Quick Win
Save time today
Email · meeting notes · first drafts · data reports · doc summaries
1 day to start · 30–60% saved
🔧 Mid-term
1–4 weeks, clear ROI
CS AI · internal KB Q&A · lead scoring · multi-platform distribution
Some setup · measurable
🚀 Moat
Long-term plays
Personalization · supply-chain forecasting · custom Agents · AI-native features
Data + tech investment
Where AI does NOT belong (yet)
High-stakes interpersonal judgment · core creative direction · high-compliance zones · errors = disaster
路径 1
现有工具的 AI 增强
Notion AI、Slack AI、Figma AI——你正在用的工具已经内嵌了 AI。零成本零迁移。
零成本启动
路径 2
MCP 连接 + Skills 扩展
Claude/Cowork 通过 MCP 直接读写你的 Slack、Notion、Drive。一个地方操作所有系统。
10分钟安装一个
路径 3a
Workflow 自动化(拖拽式)
n8n/Make/Zapier 搭流水线。不写代码。上限 = 平台支持什么
适合标准连接
路径 3b
Vibe Coding 定制
Cursor/Cowork + 自然语言造专属工具。更灵活。上限 = 你的想象力
适合定制逻辑
3a 和 3b 怎么选?
不互斥。趋势:Vibe Coding 门槛在快速下降。标准操作用 Workflow 更省事。复杂/定制需求,Vibe Coding 往往更快。
Path 1
AI inside existing tools
Notion AI, Slack AI, Figma AI — already built in. Zero cost, zero migration.
Zero-cost start
Path 2
MCP connectors + Skills
Claude/Cowork reads and writes your Slack, Notion, Drive via MCP. One place for all systems.
10 min to install
Path 3a
Workflow automation (drag-drop)
n8n/Make/Zapier pipelines. No code. Ceiling = platform support.
Standard connections
Path 3b
Vibe Coding (custom)
Cursor/Cowork + natural language. More flexible. Ceiling = imagination.
Custom logic
3a vs 3b?
Not mutually exclusive. Trend: Vibe Coding bar is dropping fast. Standard connections → Workflow. Anything custom → Vibe Coding is often faster.
① 从业务场景开始
说"我们每天200封邮件要人工分类",不说"我想做NLP"。让技术团队选方案,你描述问题。
② 给可衡量的标准
"准确率85%以上可先上线" "响应从4小时降到30分钟"。
③ 给真实数据样本
100封真实邮件 + 期望的分类结果 > 1000字需求文档。
④ 分阶段,别 All-in
2周 pilot 验证核心假设。最贵的不是试错,是花3个月做没人用的系统。
① Start with the business scenario
Say "we manually classify 200 emails a day," not "I want NLP." You describe the problem; they pick the solution.
② Give measurable standards
"85%+ accuracy is good enough to ship." "Response from 4 hours to 30 minutes."
③ Give real data samples
100 real emails + expected classifications > a 1,000-word spec.
④ Phase it
2-week pilot validates the core assumption. The most expensive thing isn't trial and error; it's 3 months building a system nobody uses.
⏱️ 时间节省
每月多少人·小时?AI省多少?人·小时×时薪=最直接ROI。
📈 质量一致性
出错率降低多少?响应速度提升多少?24/7也是价值。
💰 全成本
不只API费——学习成本+流程改造+数据准备+维护都算进去。
🔮 战略卡位
短期ROI一般,但不做12个月后被拉开差距?这条单独评。
Quick Rule
2周内能验证核心假设 + 成功后每月省10小时以上 = 现在就试。最大成本不是试错,是等待。
⏱️ Time saved
Person-hours per month? How much does AI save? Person-hours × rate = most direct ROI.
📈 Quality
Error rate drop? Speed improvement? 24/7 availability is value too.
💰 Full cost
Not just API fees — learning curve + process redesign + data prep + maintenance.
🔮 Strategic
Short-term ROI meh, but not doing it = left behind in 12 months? Score separately.
Quick Rule
Validate in 2 weeks + saves 10+ hours/month = try it now. Biggest cost isn't trial and error, it's waiting.
Actual Intelligence A Guide by Actual Fiction

想把这份认知变成真正的能力?

Want to turn this into real capacity?

这份指南只是张地图。如果你想把它落到团队里——更清晰的判断、合适的工具、属于你们自己的用法。

This guide is just a map. If you want to ground it in your team — clearer judgment, the right tools, a way of working that's actually yours — let's talk.

© Actual Fiction 2026