# Start a 1-Person Business with Claude (FULL COURSE)
**作者**: Ronin
**日期**: 2026-06-28T18:03:20.000Z
**来源**: [https://x.com/DeRonin_/status/2076690611399176506](https://x.com/DeRonin_/status/2076690611399176506)
---
I run a $40k MRR agency with zero employees
No developers. No sales team. No ops manager. No agency partners taking 30%
Just Claude Code, a folder of skills organised like a real company, and a client pipeline that fills itself while I sleep
This is NOT another "AI will make you rich" article. I'm going to show you every system, every prompt, every price and every mistake, in order, from zero
If you need money next week, close this tab. If you're ready to build something real over the next 3-6 months, this is the only article you need
After reading and implementing this, you'll have:
1. A complete AI operating system for your business (Claude Code + a skills folder backed up in the cloud)
2. Two client machines running in parallel (freelance platforms + cold email on autopilot)
3. A sales process that closes without being salesy (including the trick where Claude listens to your sales calls)
4. Three service tiers you can deliver completely solo: websites, automations, full AI systems
5. A build framework that lets you ship things you've never built before
6. A pricing model that keeps clients paying for years (the 5x rule)
[ Let's build it ] ↓↓↓
## Why a 1-Person Business? (And Why 2026 Is the Cheat Code Year)
Quick math on where the world actually is with AI:
Roughly 84% of the planet has never used AI once. Not one prompt
Around 16% touched a free chatbot and stopped there
About 0.3% pay $20/mo for a real model
And a rounding error of humanity knows tools like Claude Code exist
You're reading an article about running a company on Claude skills
You're not late. You're early by default
Here's why that matters for money:
Billions of people and tens of millions of businesses will NEVER learn these tools
They're focused on their craft: roofing, dentistry, law, e-commerce
To them, what you'll build by the end of this article looks like magic
And magic is billable
One more unlock most people miss:
3 years ago only the US, UK, Canada and Australia were realistic markets for AI services
Right now Europe, LATAM and Asia are waking up at the same time
If you speak Swedish, Portuguese, Greek, Polish, anything non-English: that market is basically EMPTY
Same service, near-zero competition, way higher reply rates
## The Honest Part (Because I Promised No Guru Energy)
Every new business follows the same curve:
Day 1: you're hyped. "I can automate everything, clients everywhere, I'm rich"
Week 3: reality. The first client is hard. Skills take time. Nobody answers your messages
That's the bottom of the curve, and 95% of people quit exactly there
They call it a scam and move on to the next shiny thing
The 5% who push through the flat months win everything
My first client took months of daily outreach and paid a few hundred dollars
If that sentence just killed your motivation, this business is not for you
If it didn't, remember one rule for the rest of this article:
GET GOOD BEFORE YOU GET RICH
People pay you for the time you invested into skills they don't have. No skills = no invoice
## The Business Model (3 Levels, You Start at Level 1)
Before any tools, here's the full map of what you're actually selling
## Level 1: AI-Built Websites
Roughly 30% of US small businesses have no website at all. Most guides tell you to chase them
Don't. A business that survived 15 years without a website doesn't want one. That's the hardest sell, not the easiest
The real buyers are the other 70%: businesses with BAD websites
They already paid for one once, so they're proven buyers. The pain is visible: slow, broken on mobile, invisible on Google, looks like 2011
Plus two bonus segments almost nobody targets:
- brand-new businesses (fresh LLC registrations are public data, they need a site NOW and have no vendor yet)
- businesses running ads into a bad site (easiest pitch alive: "you're paying for clicks that bounce")
Easiest to deliver, fastest first dollar, and every rebuild starts with proof: their current site IS the before picture
Example on how you can get build easily one (with no specific design skills):
Typical ticket: ~$500 first sale, $3-10k once you have proof
## Level 2: Automations and Agents
Invoicing, lead follow-up, onboarding, review collection, support
This is where recurring revenue starts: things that run monthly get billed monthly
The niches where this is genuinely bleeding money:
- home services (roofers, plumbers, HVAC): one job is worth $2-15k
and the owner is on a roof when the lead calls. missed-call textback + follow-up is a printing press here
- dentists and clinics: every no-show costs $200-500. reminders, recalls and reactivation sell themselves
- real estate: a lead answered in 5 minutes converts many times better than one answered in an hour. speed-to-lead automation is the product
- law firms: intake chaos + clients ghosting paperwork. document chasing alone is worth the retainer
- solar and contractors: graveyards of old quotes. one reactivation campaign = five figures from zero ad spend
- e-commerce, gyms, salons: support inboxes, win-backs, abandoned carts
and here's the cheat code:
it's the same 4 automations in every niche. instant follow-up, review collection, reminders, dead-lead reactivation
build them once, skill-ify them, deploy per niche forever
Typical ticket: $1-5k setup + $200-500/mo
## Level 3: Full AI Systems
5-25 automations bundled behind one dashboard the client logs into
The client sees executions, hours saved, money saved. It's a product, not a gig
Typical ticket: $10-20k projects or $2-5k/mo retainers
The Rules of This Model
Everyone starts at level 1, no exceptions. Websites build skill, trust and testimonials at once
Levels compound: the client who bought a $500 website buys a $2k/mo system 8 months later
And you're not selling AI. Nobody pays for "an agent"
BAD: "we implement AI agent solutions"
GOOD: "you'll never miss a lead again, and your invoices send themselves"
[ PLAYBOOK BEGINS ] ↓↓↓
## Your AI Operating System (The 20 Minute Setup)
Your entire company will live in ONE folder
Claude Code is the operating system, skills are the employees, the folder is the office
Setup, step by step:
1. Install VS Code
You won't write code. Claude will. VS Code is just the garage
2. Install Claude Code from the terminal
Log in with a subscription (Pro or Max), not an API key. Pro is enough to start
3. Create your company folder and name it like your agency
This folder is your business. Everything you ever build lives here
Here's the exact structure you're gonna build:
```
/your-agency
├── CLAUDE.md (who you are, what the business does, how to behave)
├── skills/
│ ├── engineering/ (superpowers, context7, mcp-builder, webapp-testing)
│ ├── design/ (frontend-design, taste, brand-guidelines)
│ ├── sales/ (proposal-writer, campaign-builder, call-prep)
│ ├── delivery/ (build-premium-website, mini-automation, new-client-system)
│ └── operations/ (invoicing, onboarding, reporting)
├── clients/
│ └── [one folder per client: context, transcripts, builds]
├── templates/
│ └── [website templates, contract template, proposal skeleton]
└── content/
└── [proof posts, looms, screenshots for marketing]
```
Go create this structure RIGHT NOW. Seriously, pause reading and do it
Empty folders are fine today. They fill up over this article
4. Install your first skills
Skills are portable instruction sets that turn Claude into a specialist. I shared my full org chart of them before (42+ skills organised like departments), grab what fits
Install flow, copy-paste this into Claude Code:
```
I downloaded a folder of Claude skills to my Downloads folder
Research the official documentation for installing Claude skills
in a project, then install them into this folder
After installing, list every skill you now have access to
```
Restart the session after installing, skills load on start
5. Back it up
Create a PRIVATE GitHub repo named after your agency, then tell Claude:
```
push this folder to [repo url], don't push any secrets
```
If your laptop dies tomorrow, your company shouldn't die with it
The Core Principle (Remember This One)
The IP of a one-person business is not code
It's your SOPs, your processes and your skills
Every time you build something twice, turn it into a skill
By month 6 your skills folder is worth more than your client list
OS is running. Now we need the thing every beginner skips: customers
[ Client Machine #1 ] ↓↓↓
## Personal Brand + Lead Magnets (Clients Come to You)
Most guides send beginners to freelance platforms
I won't. A fresh profile with zero reviews competes against 6-year-old accounts, and you pay for every application. Brutal place to get offer #1
Here's the smarter machine: build proof in public and let clients DM you
The Proof-First Principle
Nobody hires a stranger with no track record
But here's the cheat: you don't need clients to build a track record
You need BUILDS. And with Claude you can produce a portfolio in one weekend that most freelancers take a year to collect
Step 1: Build Before Anyone Asks
Pick ONE niche (say, roofers or dentists)
Then build 3-5 real demos for businesses in that niche:
- a premium website for a real local business (yes, without asking)
- a missed-call textback automation with a 60 second demo video
- a review-collection flow showing the before/after math
fake clients, real builds. the work is identical to paid work, only the invoice is missing
Step 2: Post Every Build (The Content Engine)
every build becomes 3-5 posts:
- the 60 second screen recording: "built this website for a local roofing company in 10 minutes with Claude"
- the before/after shot of their current site vs your rebuild
- the numbers post: "this automation answers calls the owner misses. roofers lose $2-15k per missed call"
- the how-I-did-it breakdown for the builder audience
post on X and LinkedIn, daily. one platform for reach, one for buyers
you're not posting content. you're posting EVIDENCE
Step 3: The Lead Magnet (Turn Views into DMs)
a lead magnet is something so useful people hand over contact
info to get it. yours are free because Claude built them:
- the niche template pack: "5 website templates for [niche], free, DM me TEMPLATE"
- the audit: "comment AUDIT and I'll record a 2 minute video on what your website is losing you"
- the calculator: "this tool shows how much revenue a slow site costs you. link below"
the mechanic that matters: make them COMMENT or DM to get it
every comment = the algorithm pushes your post harder
every DM = a warm conversation you didn't cold-start
Example on how you can sell your product with this mechanics:
> **Miko@Mho_23**: [原文链接](https://x.com/Mho_23/status/2071293386724982947)
>
> I just cracked the code on AI animation ads
> every one of these cost me 12 cents and under 20 seconds to make
> Gemini Omni + Claude Code is f*cking wild.
> i built a system that turns any product into an end-to-end AI animated ad in ANY style you want.. claymation, pixar-style
>
> 
His tweet got 765 interested people in this system
For example 300 of 765 will read the docs he sent in DMs
In this doc, he literally leaves a lot of sales magnets to "BOOK A CALL" or if you want to get ready such as system with no efforts, "BUY IT HERE"
His system costs $247 (and let's just imagine the most minimum conversion 1%). That's 3 buyers at $247 = $741 from only one tweet and it scales...
In total, from such as conversion loops, he has already attracted 683 members who got interested to buy this system for $247 (IN TOTAL: $168,701)
Proof from Whop:
Step 4: The Smart Sell (The Demo-First Close)
here's the move that closes the loop, and it's the whole philosophy of this article in one play:
1. find a business in your niche with a bad website
2. rebuild their homepage with Claude (10 minutes, you have the skill)
3. deploy it on a free subdomain
4. send the owner a short loom: "I rebuilt your homepage, here's yours vs this. if you want it, it's yours, and I'll connect your domain today"
5. post the before/after publicly (blur the name if they haven't answered)
you're not asking for work. you're handing them the finished product with the price tag as the only step left
rejection rate is irrelevant: every "no" still produced
a portfolio piece and 3 posts
Step 5: First 3 Clients = Testimonial Price
your first 3 clients don't pay full price, they pay in proof:
"I'll build it for [cheap/free]. in exchange: a testimonial, a case study with real numbers, and 2 referrals if you're happy"
after 3 case studies with numbers, the personal brand machine
runs itself: post proof, collect DMs, close on calls
and unlike a freelance profile, this asset is YOURS. every post compounds, nobody charges you per application, and the audience you build becomes distribution for everything you ever sell after
Personal brand is your daily grind. now the machine that works while you sleep
[ Client Machine #2 ] ↓↓↓
## Cold Email (The Background Engine)
Personal Brand = active daily work, an hour or two every day
Cold email = passive pipeline, set up once, feeds you leads in the background
Run BOTH from day one. This pair is the highest-leverage combo that exists for a beginner
The Stack:
```
Instantly sending + inbox management ~$10-15 per domain
Apollo lead database + filtering free to browse
lead scraper exporting Apollo lists ~$0.005 per lead
MillionVerifier cleaning the list cheap per batch
Claude + MCP writing and managing campaigns your subscription
```
Deliverability Crash Course
Skip this and you burn your domains:
- NEVER send cold email from your main domain. One bad campaign kills it
- Buy pre-warmed domains inside the sending tool (~$10-15 per domain, 5 inboxes each)
- Each inbox sends ~20 emails/day safely, so one domain = ~100 sends/day
- Warming your own domain takes ~30 days, pre-warmed skips the wait
- Set domain forwarding to your real website so curious prospects land somewhere real
- Wait for 100% health score before sending anything
Building the List
The formula:
[niche] x [country] x [1-50 headcount] x [owner/founder titles] x [verified emails only]
- Filter in Apollo: location, industry keywords, titles (CEO, Founder, Owner), verified emails
- Headcount is the most underrated filter: 1-50 employees means the owner reads their own inbox
- Export through a scraper: 1,500 leads for ~$25 instead of Apollo pricing
- Clean the list through a verifier: expect ~60-65% "good", send ONLY to good
- Sending to dead emails = bounces = blacklisted domain = start over
And remember the language arbitrage: Swedish, Portuguese, Greek, Polish inboxes are EMPTY while US inboxes are a warzone. Same offer, translated, multiples of the reply rate (check local cold-email laws first, seriously)
The MCP Moment
Now the part that makes this a Claude article
Connect your sending tool's MCP to Claude (Claude settings → Connectors → add custom connector → paste the API URL)
Suddenly Claude can create campaigns, write sequences, pull replies and manage follow-ups on its own, ~38 tools worth of control
Then encode your campaign structure into a skill. The skill interviews YOU before writing:
```
1. What are you selling and to whom? (one paragraph)
2. What's the #1 pain this niche feels daily?
3. What tangible asset does the email offer?
(a demo site, a lost-revenue calculator, an audit,
this becomes the subject line)
4. What should the campaign be called?
```
Answer, and Claude drafts the full sequence with follow-ups and creates it in the tool through the MCP. You review, save, launch
Here's the first-email framing that works:
```
subject: [tangible asset for their business]
I was going to call, but figured I'd write first
I think [company] is losing [jobs/patients/orders] because
[specific observable problem: site doesn't rank, no reviews,
slow follow-up]
I made you [the asset] so you can see what I mean: [link]
Worth a 15 minute call?
```
The Numbers
So you know what "working" looks like:
A good campaign: ~10,000 sends, 5%+ reply rate. The industry norm is ~0.5%
Follow-ups produce most of the replies. Never send one-and-done
Rule of thumb: 100 sends/day = one interested lead every other day
One interested lead every other day = a full calendar within a month
And the offer rule that drives all of it: never email "we do AI solutions". Email a tangible artifact. "I built your company a demo site, here it is" flips the whole dynamic, they've seen the product before the call even starts
Leads are replying. Now the part everyone fumbles: the call
[ The Sales System ] ↓↓↓
## How to Close Clients (The Doctor Approach)
Beginners lose deals in three places: before the call, during the call, after the call
Here's all three, fixed
Before the Call
Rule 1: NEVER give price before the call
Price in DMs = judged on price alone, they can't see value yet
Price after a demo = compared against the value they just saw
Show $10k of value, then say $2k, and their brain does the math for you
If they push for price over text: "depends on scope, that's exactly what the 15 minutes is for"
Rule 2: confirmations + reminders on every channel
People forget they booked, feel embarrassed, and ghost. Kill it with cadence:
```
instantly: "confirmed, [day] at [time]. calendar invite sent"
every 3 days: "still on for [day]! preparing something for you"
1 day before: "quick reminder about tomorrow. I built a demo
based on your actual business, excited to show it"
1 hour before: "see you at [time]"
5 min before: "just wrapping another call, see you in 5"
```
The demo mention is deliberate. Skipping a meeting where someone built something FOR YOU feels rude. Guilt is a show-up strategy
Rule 3: never sell in DMs
Limited info in text, full value live
During the Call
A doctor doesn't open with "buy this prescription"
They ask where it hurts, diagnose, THEN prescribe
You talk 20%, they talk 80%, and your 20% is mostly questions
Why it works: when the CLIENT says the pain out loud, they convince themselves. You can't out-argue a conclusion they reached on their own
The question ladder:
```
1. "how do new customers usually find you right now?"
2. "what happens to a lead that calls after hours?"
3. "what do you think a customer does when they compare
your 5 reviews to your competitor's 150?"
4. "how much is one new [job/patient/client] worth to you?"
5. "what have you already tried to fix this?"
```
Question 3 is the pattern: make them state the loss themselves
Question 4 makes them price the solution before they've heard your price
The call skeleton:
1. Pain questions (the ladder above)
2. "I might have something for this, let's see if it even makes sense"
3. Show the demo, screen share, under 10 minutes
4. State the price. ONE number, clean
5. Then SHUT UP. The silence after price is where deals close. Don't fill it
6. Handle objections (they're usually questions in disguise: "what if it doesn't work?"), answer, close
And the ethics line: if they don't need it, tell them. A business with 500 reviews doesn't need your review system. One honest "you don't need this" generates more referrals than any close
Closing Mechanics
- Always book the next call ON the current call. Calendar = commitment. "Let me send the invite right now while we're here"
- Always take a commitment on the call
- The $1 Stripe trick: create a $1/year recurring payment link. "Let's put a card on file, it's $1 and I'm refunding it". Later you charge the real amount with zero friction
And the conversion weapon, the Proof of Concept close:
Payment upfront + satisfaction guarantee
"You pay to start, I build the entire thing, you see the finished product, and if you're not happy: full refund, no questions"
Removes 100% of their risk, signals total confidence
If your work is good you will basically never refund, and your close rate jumps
After the Call
- Overdeliver on the first project. Your reputation IS the business at this stage
- Set recurring bi-weekly review calls with every client. Agencies average 20-30% monthly churn, review calls take you to ~5%. 30% churn means clients stay 3 months, 5% means years
- Upsell ON those calls, never cold: website client → review automation → AI phone agent → ads
- Referrals, 2 rules: be SPECIFIC ("do you know any plumbers or electricians?" beats "know anyone?", the brain can't search "anyone") and give 20% commission on referred revenue
The Claude Trick I Promised
Connect an AI notetaker MCP to Claude (Fathom or similar). It records and transcribes every call
After the call, one prompt:
```
Check the transcript from my [time] call with [client]
Extract: their business context, every pain point mentioned,
what we agreed to build, timeline and budget signals
Write a full project spec into clients/[client-name]/
and propose the build plan
```
Claude reads the meeting, writes the spec, and can START BUILDING before you've had lunch
Zero context loss between what the client said and what gets built
Client closed, money committed. Now we deliver, and this is where Claude turns you into a team
[ Delivery Level 1 ] ↓↓↓
## Websites (Your First Product)
~10 million US small businesses have no website. They will never learn Claude Code, they're busy roofing and filling teeth. That gap is the job
What the market charges:
```
template freelancer $1-3k
boutique studio $3-10k
proper agency $10-30k
mid-tier agency $30-80k
top agency $80-250k+
your realistic FIRST sale ~$500
your target in year one $3-10k
```
The $500 first website is not underselling. It's buying a testimonial, a portfolio piece and a future upsell path in one transaction
The Build
Use a premium-website skill (build one or grab one from the skill repos)
/build-premium-website, then answer the interview: industry, brand vibe, colors, services
Production React + Tailwind site on localhost in under 10 minutes
Then iterate with screenshots: paste what looks wrong, describe it in plain words, Claude fixes it
What Makes It Sellable (Not Just Pretty)
- Bold CTA above the fold with a click-to-call number
- Social proof section: reviews, before/after photos
- Contact form wired to their CRM or email
- Fast load, animations in code
- Mobile checked (inspect → toggle device view)
Most designers build pretty sites that don't convert. You build converting sites. That's the pitch
Ship It, Free
1. Push the code to a GitHub repo
2. Vercel: add new project, paste the repo, deploy
3. Live in ~30 seconds on a free subdomain
4. Custom domain later through any registrar
Hosting cost at this stage: $0
The Portfolio Play
Build 5 template sites for 5 different industries THIS WEEK. Fake businesses are fine
One portfolio site that showcases all 5
Prospects pick instead of imagine, and picking closes faster than imagining
These templates are also your Upwork portfolio and your first proof posts
Then, after your second real website: encode the process into a skill
Delivery collapses from days to an afternoon
You just created your first employee
Websites open doors. Automations keep the money coming every month
[ Delivery Level 2 ] ↓↓↓
## Automations (Where Recurring Money Starts)
The default advice is drag-and-drop builders. Skip them
Use a code-based open source automation runner instead, here's why:
- Claude writes code better than it clicks UI
- Code is versionable, testable, and lives in your GitHub
- Open source = host anywhere, even on the client's own servers
- "Runs on your servers" is a phrase enterprise buyers pay extra for
For third-party access use an OAuth broker (Composio style) so Claude never touches raw Google credentials. You run one script, click two links, connections go active
One Complete Build: The Invoicing Automation
Form in → PDF invoice out → auto-sent via Gmail → archived to Drive
The prompt pattern:
```
I want to automate my invoicing process
Flow: I fill a simple web form, it builds the invoice,
exports a PDF, emails it to the client via Gmail,
and archives a copy to Google Drive
Use two skills:
1. the composio skill for all third-party authentication
2. the trigger.dev skill to build and host the task
Auto-generate invoice numbers by date. Keep all secrets
in a .env file, nothing hardcoded
Plan mode first: ask me your clarifying questions before building
```
Memorize that pattern, it repeats for everything: outcome + skills to use + constraints + plan mode
Claude asks clarifying questions (fields, trigger, email style), you answer, approve the plan, it builds
Then wire it up:
```
runner API key (from the automation platform)
auth broker key (from the OAuth service)
Anthropic API key (for AI-generated email text)
Drive folder ID (from the folder URL)
business details (logo URL, bank details, currency, net terms)
```
The Debug Loop
Something always breaks on the first test. The loop:
"check the logs, did my test work?"
Claude reads the run logs, finds the issue, fixes it, you rerun
Repeat until green. You never open a stack trace yourself
This loop is 80% of what "being technical" means now
Deploy the task to production, deploy the form to Vercel, done
One SECURITY WARNING: an exposed endpoint wired to your API key = a stranger burning your credits at 3am. Auth on anything public is not optional (the full pattern is in the next section)
The Biggest Idea in This Article
BUILD ONCE. SKILL-IFY IT. DELIVER FOREVER
After any successful build:
```
Create a skill from what we built this session
Make it GENERIC: any mini automation with a form front end,
a background task back end, and OAuth-brokered integrations,
not invoicing-specific
The skill must interview the user about their use case first,
and must research before building
Make it fully standalone: no references to this project's paths
Verify it's standalone before finishing
```
Then double check: "is this skill standalone? does it reference the invoice app anywhere?"
Your skills folder = your product catalog = the actual IP of the business
Every skill is an employee that works free, forever
Single automations sell for hundreds. Bundled systems sell for thousands
[ Delivery Level 3 ] ↓↓↓
## Full AI Systems (The $10k+ Tier)
A full AI system = 5-25 automations behind ONE dashboard the client logs into
Most agencies hand over spaghetti flows the client never opens
You hand over a product: executions this month, hours saved, tickets answered
The dashboard is what makes a retainer feel like software instead of an invoice
The stack:
```
front end Next.js + Tailwind + shadcn components
auth magic links via email, no passwords,
restricted to the client's email domain
back end code-based automation runner (open source)
database MongoDB (built-in vector search for RAG)
hosting Vercel (front) + runner cloud (back),
or the client's own servers for enterprise
```
The Build Order That Works
1. Scaffold the structure FIRST with zero design
Dashboard page, sidebar of automations, settings, login
The rule: it's much harder to fix bad to good than to build good from the start, but structure comes before beauty
2. Design makeover pass with a frontend skill
"Give this app a makeover, [style] vibe, light and dark mode", then iterate with screenshots
3. Install automation #1
The invoicing build you already have, rebranded inside the dashboard
4. Onboarding automation, built in minutes
Contract PDF generated with the client's details + calendar link email + onboarding survey
One trigger, three emails, new client onboarded while you sleep
5. The flagship: the AI support inbox
What it does:
- Watches a shared support inbox every 10 minutes
- RAG over the client's knowledge base (they paste docs, PDFs, URLs)
- Auto-answers the ~60% of tickets that are repetitive
- Every answer ships with a confidence score + citations
- Low-confidence tickets escalate to a human queue WITH a pre-drafted reply waiting
- Analytics tab: tickets answered, escalation rate, hours saved
The Economics That Sell It
```
a support rep costs $2-4k/mo
a 10-person team cut to 5 saves $15k/mo
you charge $2-5k/mo
client still saves $10k+/mo
nobody churns from that math
```
And the same system resells to your next client with just a new knowledge base. Second sale = almost pure margin
Ship It + Skill-ify It
Private GitHub repo, Vercel deploy, set every env variable in the hosting dashboard, and update the auth URL from localhost to the live domain (everyone forgets this exactly once)
Then the final skill-ification: encode the whole scaffold as a "new-client-system" skill that asks for client name, email domain and brand colors
Every future client gets a live dashboard on DAY ONE, in minutes
This is the moment it stops being freelancing and becomes a company
And if you're thinking "I could never build this", the next section deletes that excuse
[ The Meta-Skill ] ↓↓↓
## How to Build ANYTHING With Zero Experience (The 4-Step Framework)
1. BUILD PLAN (the design spec)
What you're building + the tech stack + the design and UX
The stack determines cost, speed, scalability and how hard it is to extend later
Go back and forth with Claude for literally HOURS on a complex build
This document is where the project is won or lost
2. IMPLEMENTATION PLAN
The how: the build plan broken into steps with checkboxes
A complex build is ~40 steps: file structure, database, auth, features, in order
Forcing sequence forces good decisions
3. BUILD
Feed steps one at a time: "start with step 1"
Approve between steps, don't dump the whole plan at once
Use sub-agent execution for speed on big builds
4. TEST AND REFINE
You WILL hit bugs. Paste every error to Claude verbatim
Then the design makeover pass at the end
The Ratio Rule
80% of your time goes into steps 1 and 2
20% into step 3
Everyone does the opposite, then wonders why the AI "built the wrong thing"
Give me six hours to chop down a tree and I'll spend four sharpening the axe
The Zero-Experience Cheat: Two-Model Verification
One AI proposes the stack and the plan
A second session audits it: "here's a proposed architecture, is this actually the right choice? what breaks at scale?"
Consultant + auditor. They catch each other's blind spots
This is how you make senior-engineer decisions with zero background
And when you don't know the answer to a clarifying question, "what would you recommend?" is always allowed
Your kickoff prompt for anything:
```
I want to build [thing]
First: a build plan. Ask me questions about how it should
work, then write the design spec as a .md file
Recommend the tech stack and justify it: cost to run,
speed, scalability
Don't write any code until I approve the spec
```
The framework is the skill. The stack is a detail
You can build anything. Now let's make sure you charge like it
[ Pricing ] ↓↓↓
## How to Price AI Services (The 5x Rule)
The golden rule of everything:
GIVE EVERY CLIENT A 5X RETURN ON WHAT THEY PAY YOU
Save them $10k/mo, charge $2k/mo
Why it works:
- The client re-does the math every month and re-decides to stay. Make the math undeniable
- Low churn compounds into referrals
- Enough referrals = negative churn: the agency grows without outreach
Calculating the Value
Path 1, payroll saved:
```
support team: 10 reps at $3k/mo
your system cuts it to 5
value created: $15k/mo
your price at 5x: $3k/mo
client keeps: $12k/mo, forever
```
Path 2, new revenue:
Price on GROSS PROFIT of NEW revenue only, never existing
System generates $100k/mo new revenue at ~$40k gross profit → your price ~$8k/mo
The Three Pricing Structures
1. Result-based (% of gross profit, ~20%)
Highest ceiling, scales infinitely with the client
Requires trackability: a real CRM, clean attribution
One result-based client can produce six figures of LTV where a retainer would have made $12k
Rule: retainer if you're average, result-based if you're great
When value is untrackable (a support widget), fall back to per-unit pricing (~$1 per conversation)
2. Upfront + recurring (the default)
Example: $2k upfront + $500/mo service fee
The upfront creates commitment. Clients who pay nothing ghost
The recurring is justified as support, monitoring, fixes, improvements
ALWAYS pair with the satisfaction guarantee
Ten clients on this = $5k/mo baseline before any new sales
3. Usage tiers (whale insurance)
Like software: 100 calls / 1,000 calls / 5,000 calls / custom above
Protects you from the $500/mo client who turns out to do 10,000 calls
Simplicity Rules
A confused person doesn't buy
ONE price, not a menu of line items
Bundle software costs into your price 99% of the time ($600 flat beats "$500 + $100 in tools")
Exception: enterprise that wants to own and host everything, they pay their own tooling and a much higher fee
And the rule above all pricing rules:
BAD: "an AI agent with RAG over your knowledge base and confidence-scored escalation"
GOOD: "your support inbox answers itself and you save $10k a month"
Nobody buys AI. Nobody even buys automation. They buy more money, more time, and fewer employees doing boring work
One more engine and the whole loop closes
[ The Flywheel ] ↓↓↓
## Show Your Work (The Compounding Loop)
Every client build is a content asset you already paid for
The loom you recorded for the proposal: content
The dashboard screenshot: content
The before/after of their website: content
The support inbox hitting 63% auto-answered this month: content
The loop:
build for a client → post the proof → proof pulls inbound leads → inbound = no connects, no sending costs, no cold anything → better clients → better builds → better proof → repeat
What to post: "built this in 10 minutes with Claude" screen recordings, before/after shots, real numbers, and the org chart of your skills folder (meta, but it works, ask me how I know)
I run this exact posting system across 10 platforms without writing anything manually. The full breakdown is in my Content Engine article, it's the perfect companion to this one
Your agency and your audience feed each other. Most people build only one. Build both
## CONCLUSION (what to do)
1. Set up the OS today: folder + Claude Code + skills + GitHub backup. 20 minutes, do it right now
2. Build your first template website tomorrow for a fake business, deploy it free
3. This weekend: create your freelance profile AND launch one small cold campaign
4. Spend 4 HOURS A DAY on outreach. Marketing and selling are the real skills of this business, not AI. This is the part everyone underestimates and the reason most people fail
5. Take your first clients free or cheap. You're buying testimonials, trust and reps, not income
6. Get to $2-3k/mo that covers your life. The success criteria of business is staying in business. 9 out of 10 die here, not at scaling
7. Then raise prices. Fire broke and annoying clients. Fewer clients, higher pay
8. Find the bottleneck, fix it, find the next one. Leads, then calls, then delivery, then leads again. You'll be on this step forever. That's the job
And the honest footnotes:
You don't have to quit your job. Build 5-9am when your brain is fresh, work 9-5, two hours in the evening, weekends
Growth is exponential AND volatile: months of flat, then it clicks, then a dip, then a bigger click
One person is how you START, not a cap. Claude carries 10-15 clients mostly on its own
And don't over-automate. Sometimes hiring one human for calls beats three weeks of building a mediocre automation. Even coming from me
A business used to be an office full of people
Now it's a folder, a subscription, and someone who refuses to quit
If you loved this article and found it valuable, please put the LIKE + RT ❤️
If this article collects 2,000+ Likes, I drop the full skill pack from it (proposal writer, campaign builder, website builder, new-client-system) in a follow-up
Now, it's time to go build it buddies 😎
## 相关链接
- [Ronin](https://x.com/DeRonin_)
- [@DeRonin_](https://x.com/DeRonin_)
- [187K](https://x.com/DeRonin_/status/2076690611399176506/analytics)
- [$3-10k](https://x.com/search?q=%243-10k&src=cashtag_click)
- [$2-15k](https://x.com/search?q=%242-15k&src=cashtag_click)
- [$1-5k](https://x.com/search?q=%241-5k&src=cashtag_click)
- [$10-20k](https://x.com/search?q=%2410-20k&src=cashtag_click)
- [$2-5k](https://x.com/search?q=%242-5k&src=cashtag_click)
- [$2-15k](https://x.com/search?q=%242-15k&src=cashtag_click)
- [Jun 29](https://x.com/Mho_23/status/2071293386724982947)
- [108K](https://x.com/Mho_23/status/2071293386724982947/analytics)
- [$10k+](https://x.com/search?q=%2410k%2B&src=cashtag_click)
- [$2-3k](https://x.com/search?q=%242-3k&src=cashtag_click)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [11:29 PM · Jul 13, 2026](https://x.com/DeRonin_/status/2076690611399176506)
- [187.6K Views](https://x.com/DeRonin_/status/2076690611399176506/analytics)
- [View quotes](https://x.com/DeRonin_/status/2076690611399176506/quotes)
---
*导出时间: 2026/7/14 09:40:33*
---
## 中文翻译
# 用 Claude 开启一人公司（完整教程）
**作者**： Ronin
**日期**: 2026-06-28T18:03:20.000Z
**来源**： [https://x.com/DeRonin_/status/2076690611399176506](https://x.com/DeRonin_/status/2076690611399176506)
---
我运营着一家月经常性收入（MRR）4 万美元的机构，零员工
没有开发者。没有销售团队。没有运营经理。没有抽成 30% 的机构合作伙伴
只有 Claude Code、一个像真正公司那样组织的技能文件夹，以及一条在我睡觉时也能自动填满的客户管道
这不是又一篇“AI 会让你暴富”的文章。我会按顺序向你展示每一套系统、每一个提示词、每一个价格和每一个错误，从零开始
如果你下周就需要钱，请关掉这个页面。如果你准备好在未来 3-6 个月里打造一些真正的东西，这就是你唯一需要的文章
读完并实践本文后，你将拥有：
1. 一套完整的业务 AI 操作系统（Claude Code + 备份在云端的技能文件夹）
2. 两台并行运转的客户获取机器（自由职业平台 + 自动化冷邮件）
3. 一个不强推销售也能成交的销售流程（包括让 Claude 听你销售通话的技巧）
4. 三个可以完全独立交付的服务层级：网站、自动化、完整 AI 系统
5. 一个让你能交付从未做过的事物的构建框架
6. 一个让客户持续付费多年的定价模型（5 倍法则）
[开始构建吧] ↓↓↓
## 为什么选择一人公司？（以及为什么 2026 年是作弊码之年）
简单算一下 AI 在全球的真实普及程度：
大约 84% 的地球人从未用过 AI。连一次提示词都没输入过
约 16% 的人接触过免费聊天机器人后就止步于此
大约 0.3% 的人每月花 20 美元订阅一个真正强大的模型
而知道 Claude Code 这类工具存在的人，只是统计误差级别的极少数
而你现在正在读一篇关于用 Claude 技能运营公司的文章
你没有迟到。你天生就是早鸟
这就是为什么这与赚钱息息相关：
数十亿人和数千万家企业永远不会学习这些工具
他们专注于自己的手艺：屋顶施工、牙科、法律、电商
对他们来说，你读完本文后能构建的东西看起来就像魔法
而魔法是可以收费的
还有一个大多数人都错过的认知解锁：
3 年前，只有美国、英国、加拿大和澳大利亚是 AI 服务的现实市场
现在欧洲、拉美和亚洲正在同时觉醒
如果你会瑞典语、葡萄牙语、希腊语、波兰语或任何非英语语言：那个市场基本上是空的
同样的服务，近乎零竞争，回复率高得多
## 诚实的部分（因为我承诺过不搞大师那套）
每个新生意都遵循同样的曲线：
第 1 天：你热血沸腾。“我能自动化一切，客户遍地都是，我要发财了”
第 3 周：现实降临。第一个客户很难拿。技能需要时间积累。没人回复你的消息
那就是曲线的谷底，95% 的人恰好在那里放弃
他们称之为骗局，然后转头追逐下一个闪亮的新东西
而那撑过平淡月份的 5% 赢得了一切
我的第一个客户花了好几个月的每日拓客，只付了几百美元
如果这句话刚杀死了你的动力，这门生意不适合你
如果没有，请记住贯穿本文余下部分的一条法则：
先变强，再变富
人们付钱给你，是为你投入到他们所不具备的技能上的时间。没有技能 = 没有账单
## 商业模式（3 个层级，从第 1 级开始）
在接触任何工具之前，先看看你实际在售卖什么的完整地图
## 第 1 级：AI 构建的网站
大约 30% 的美国小企业根本没有网站。大多数指南都让你去追他们
别。一家 15 年没有网站还活得好好的企业不想要网站。那是最难卖的东西，不是最容易的
真正的买家是另外那 70%：网站很糟糕的企业
他们已经为网站付过一次钱，所以是被验证过的买家。痛点显而易见：加载慢、手机上显示混乱、在 Google 上搜不到、长得像 2011 年的产物
再加上两个几乎没人瞄准的细分群体：
- 全新企业（新注册的 LLC 是公开数据，他们现在就需要网站，而且还没有供应商）
- 投放广告但落地页很糟糕的企业（最简单的推销话术：“你在为跳出率买单”）
最容易交付，最快赚到第一笔钱，而且每一次重建都自带证明素材：他们现在的网站就是“改造前”照片
关于如何轻松构建一个网站的示例（无需任何专门的设计技能）：
典型客单价：首单约 500 美元，有了案例后 3 千至 1 万美元
## 第 2 级：自动化与智能代理
开票、潜客跟进、客户入驻、评价收集、客服支持
这就是经常性收入开始的地方：每月持续运行的东西就按月收费
这些真正在流失金钱的利基市场：
- 家庭服务（屋顶工、水管工、暖通空调）：一单价值 2 千至 1.5 万美元，而潜客来电时老板正在房顶上干活。未接来电自动回短信 + 跟进在这里就是一台印钞机
- 牙医和诊所：每位爽约患者损失 200-500 美元。提醒、召回和唤醒 dormant 客户的功能自己就能卖出去
- 房地产：5 分钟内回复的潜客转化率远高于一小时后才回复的。响应速度自动化就是产品本身
- 律所：案件受理混乱 + 客户不交材料。光是追文件就值回服务费
- 太阳能和承包商：旧报价单的坟场。一场唤醒营销活动 = 零广告投入换来五位数收入
- 电商、健身房、美容沙龙：客服收件箱、流失客户挽回、弃购提醒
而这还有个作弊码：
每个利基市场都是同样的 4 个自动化。即时跟进、评价收集、提醒、沉睡潜客唤醒
构建一次，封装成技能，永远在每个利基市场部署
典型客单价：1 千至 5 千美元搭建费 + 每月 200-500 美元
## 第 3 级：完整 AI 系统
5-25 个自动化打包在一个客户登录的仪表盘后面
客户看到的是执行记录、节省的小时数、省下的钱。这是一个产品，不是一单零活
典型客单价：1 万至 2 万美元的项目，或每月 2 千至 5 千美元的服务费
## 这个模式的规则
每个人都从等级 1 开始，没有例外。网站可以同时积累技能、信任和客户见证
等级会复利增长：今天花 $500 买网站的客户，8 个月后会买每月 $2k 的系统
而且你卖的不是 AI。没有人会为“一个智能体”付钱
错误：“我们提供 AI 智能体解决方案”
正确：“你再也不会错过任何一个潜在客户，你的发票会自动发送”
[ 实战手册开始 ] ↓↓↓
## 你的 AI 操作系统（20 分钟搭建）
你的整个公司都将存在于一个文件夹中
Claude Code 是操作系统，技能是员工，文件夹是办公室
搭建步骤，一步一步来：
1. 安装 VS Code
你不需要写代码，Claude 会写。VS Code 只是个车库
2. 从终端安装 Claude Code
使用订阅（Pro 或 Max）登录，而不是 API 密钥。开始用 Pro 就够了
3. 创建你的公司文件夹，用你的代理公司名字命名
这个文件夹就是你的生意。你以后构建的一切都放在这里
以下是你要搭建的确切结构：
```
/your-agency
├── CLAUDE.md （你是谁、公司做什么、如何行事）
├── skills/
│ ├── engineering/ （superpowers、context7、mcp-builder、webapp-testing）
│ ├── design/ （frontend-design、taste、brand-guidelines）
│ ├── sales/ （proposal-writer、campaign-builder、call-prep）
│ ├── delivery/ （build-premium-website、mini-automation、new-client-system）
│ └── operations/ （invoicing、onboarding、reporting）
├── clients/
│ └── [每个客户一个文件夹：背景资料、通话记录、交付成果]
├── templates/
│ └── [网站模板、合同模板、提案框架]
└── content/
└── [证明类帖子、Loom 视频、用于营销的截图]
```
现在就去创建这个结构。真的，暂停阅读，马上做
今天文件夹是空的没关系。读完整篇文章它们就会填满
4. 安装你的第一批技能
技能是可移植的指令集，能把 Claude 变成专家。我之前分享过我的完整技能组织架构（42+ 个技能，按部门组织），挑适合的拿去用
安装流程，把这段复制粘贴到 Claude Code：
```
我下载了一个 Claude 技能文件夹到我的 Downloads 文件夹
查阅关于在项目中安装 Claude 技能的官方文档，
然后把它们安装到这个文件夹
安装完成后，列出你现在可以使用的每一个技能
```
安装后重启会话，技能会在启动时加载
5. 备份
创建一个以你的代理公司命名的私有 GitHub 仓库，然后告诉 Claude：
```
把这个文件夹推送到 [repo url]，不要推送任何机密信息
```
如果你的笔记本电脑明天坏了，你的公司不应该随之消亡
核心原则（记住这一条）
一人企业的 IP 不是代码
而是你的 SOP、你的流程和你的技能
每当你重复构建同一个东西两次，就把它变成一个技能
到第 6 个月，你的技能文件夹会比你的客户名单更值钱
操作系统已经运行。现在我们需要每个新手都会跳过的东西：客户
[ 客户机器 #1 ] ↓↓↓
## 个人品牌 + 铅磁（客户主动找上门）
大多数指南让新手去自由职业平台
我不会这么做。一个零评价的新账号要和 6 年的老账号竞争，而且每投一次简历都要付费。这是获取第一单 offer 的残酷之地
这里有更聪明的机器：公开积累证据，让客户私信找你
证据优先原则
没有人会雇佣一个没有业绩记录的陌生人
但这里有个捷径：你不需要客户也能建立业绩记录
你需要的是作品。用 Claude，你可以在一个周末内做出大多数自由职业者要花一年才能积累的作品集
第 1 步：在没人要求之前先构建
选择一个细分领域（比如屋顶承包商或牙医）
然后为该领域的企业构建 3-5 个真实演示：
- 为一个真实的本地企业做一个高端网站（是的，不用先问）
- 一个未接来电自动回拨短信，配上 60 秒演示视频
- 一个评论收集流程，展示前后对比数据
假客户，真作品。这些工作和付费工作完全一样，只差一张发票
第 2 步：发布每一个作品（内容引擎）
每个作品变成 3-5 条帖子：
- 60 秒屏幕录像：“用 Claude 在 10 分钟内为一家本地屋顶公司建好了这个网站”
- 他们现有网站与你重建版本的前后对比图
- 数据帖：“这个自动化能回复老板错过的电话。屋顶承包商每错过一个电话损失 $2k-15k”
- 面向开发者受众的“我是怎么做到的”拆解
每天在 X 和 LinkedIn 上发帖。一个平台管曝光，一个管买家
你发布的不是内容。你发布的是证据
第 3 步：铅磁（把浏览变成私信）
铅磁是实用到让人愿意交出联系方式来获取的东西。你的铅磁是免费的，因为是 Claude 做的：
- 细分领域模板包：“[某领域] 的 5 个网站模板，免费，私信我 TEMPLATE”
- 免费审计：“评论 AUDIT，我会录制一个 2 分钟的视频，讲讲你的网站正在让你损失什么”
- 计算工具：“这个工具能显示一个加载慢的网站让你损失多少收入。链接在下面”
关键机制：让他们必须评论或私信才能获取
每条评论 = 算法会把你的帖子推得更猛
每条私信 = 一场你不需要冷启动的温暖对话
用这种机制销售产品的示例：
> **Miko@Mho_23**: [原文链接](https://x.com/Mho_23/status/2071293386724982947)
>
> 我刚刚破解了 AI 动画广告的密码
> 每一条成本只要 12 美分，制作时间不到 20 秒
> Gemini Omni + Claude Code 太疯狂了。
> 我搭建了一个系统，可以把任何产品变成任何风格的端到端 AI 动画广告……黏土动画、皮克斯风格都行
>
> 
他的推文让 765 个人对这个系统感兴趣
比如 765 人中有 300 人会阅读他私信发送的文档
在这份文档里，他实际上留下了很多销售引导，指向“预约通话”，或者如果你想不费力气就得到这样一个系统，那就是“在此购买”
他的系统售价 $247（假设最低转化率 1%）。那就是 3 个买家 × $247 = $741，而这仅仅来自一条推文，而且它还能规模化……
总的来说，通过这些转化闭环，他已经吸引了683名成员购买这个247美元的系统（总计：168,701美元）
Whop 上的证据：
第 4 步：聪明的销售（演示优先成交法）
这一步让整个闭环完成，也是整篇文章理念在一次行动中的体现：
1. 在你的细分领域找一个网站很差的商家
2. 用 Claude 重建他们的主页（10 分钟，你有这个技能）
3. 部署到一个免费的子域名上
4. 给商家发一段简短的 Loom 视频：“我重建了你们的主页，这是你们原来的和这个的对比。如果你们想要，它就是你们的，我今天就把你们的域名接上”
5. 公开发布前后对比（如果对方没回复，就把名字打码）
你不是在求职。你是把成品交到他们手上，唯一剩下的步骤就是付钱
被拒绝的比率无关紧要：每一个“不”仍然产生了
一份作品集案例和 3 条帖子
第 5 步：前 3 个客户 = 见证价
你的前 3 个客户不用付全价，他们用证明来支付：
“我给你做【便宜/免费】。作为交换：一条推荐语、一个有真实数据的案例研究，如果你满意的话再给我 2 个转介绍”
在有了 3 个带数据的案例研究之后，个人品牌机器
就开始自动运转：发布证据，收取私信，在电话中成交
而且不像自由职业平台的主页，这个资产是属于你的。每条帖子都在产生复利，没人按申请次数向你收费，你积累的受众成为你之后出售的一切产品的分发渠道
个人品牌是你的日常工作。现在这台机器在你睡觉时也在运转
【客户机器 #2】↓↓↓
## 冷邮件（后台引擎）
个人品牌 = 每日主动工作，每天一到两个小时
冷邮件 = 被动管道，设置一次，就在后台持续给你送线索
从第一天起就要两条腿走路。这对组合是新手能用的最高杠杆组合
工具栈：
```
Instantly 发送 + 收件箱管理 每个域名约 $10-15
Apollo 线索数据库 + 筛选 浏览免费
lead scraper 导出 Apollo 列表 每条线索约 $0.005
MillionVerifier 清洗列表 每批很便宜
Claude + MCP 撰写和管理营销活动 你的订阅
```
送达率速成课
跳过这一步，你的域名就废了：
- 绝对不要用你的主域名发冷邮件。一次糟糕的营销活动就会毁掉它
- 在发送工具内购买预加热域名（每个域名约 $10-15，每个含 5 个收件箱）
- 每个收件箱每天安全发送约 20 封邮件，所以一个域名 = 每天约 100 封
- 自己加热域名需要约 30 天，预加热域名可以跳过等待
- 设置域名转发到你的真实网站，这样好奇的潜在客户能落到一个真实的页面上
- 等健康度分数达到 100% 再开始发送
建立名单
公式：
【细分领域】x【国家】x【1-50 人规模】x【老板/创始人职位】x【仅限已验证邮箱】
- 在 Apollo 中筛选：地点、行业关键词、职位（CEO、创始人、老板）、已验证邮箱
- 员工人数是最被低估的筛选条件：1-50 名员工意味着老板会亲自看收件箱
- 通过抓取工具导出：1,500 条线索约 $25，而不是 Apollo 的定价
- 通过验证工具清洗名单：预期约 60-65% 为“良好”，只发给良好的
- 发给失效邮箱 = 退信 = 域名被拉黑 = 从头再来
记住语言套利：瑞典语、葡萄牙语、希腊语、波兰语的收件箱是空荡荡的，而美国收件箱是战场。同样的报价，翻译一下，回复率翻好几倍（先查一下当地冷邮件法律，认真的）
MCP 时刻
现在是让这篇文章成为 Claude 文章的部分
把你的发送工具的 MCP 连接到 Claude（Claude 设置 → 连接器 → 添加自定义连接器 → 粘贴 API URL）
突然之间，Claude 可以自己创建营销活动、撰写序列、拉取回复并管理跟进，约 38 个工具级别的控制权
然后把你营销活动的结构编码成一个技能。这个技能在写作前会先采访你：
```
1. 你在卖什么，卖给谁？（一段话）
2. 这个细分领域每天感受到的头号痛点是什么？
3. 邮件提供的具体实物资产是什么？
（一个演示网站、一个收入损失计算器、一次审计，
这将成为邮件主题行）
4. 这个营销活动应该叫什么名字？
```
回答完毕，Claude 就会起草包含跟进邮件的完整序列，并通过 MCP 在工具中创建它。你审阅、保存、启动
以下是有效的首封邮件框架：
```
主题：[为他们的业务准备的实物资产]
我本来想打电话，但觉得还是先写邮件
我认为【公司】正在失去【工作/患者/订单】，因为
【具体可观察到的问题：网站没有排名、没有评价、
跟进缓慢】
我给你做了【那个资产】，这样你就能明白我的意思：[链接]
值得打个15分钟的电话吗？
```
这些数字
让你知道“有效”是什么样子：
一次好的营销活动：约 10,000 次发送，5% 以上的回复率。行业平均水平约为 0.5%
跟进邮件产生大部分回复。永远不要发一封就完事
经验法则：每天 100 次发送 = 每两天一个感兴趣的线索
每两天一个感兴趣的线索 = 一个月内日程排满
还有驱动这一切的报价规则：永远不要发“我们提供 AI 解决方案”这样的邮件。要发一个具体的实物。“我为你们公司做了一个演示网站，在这里”会彻底扭转整个局面，通话还没开始，他们就已经见过产品了
线索在回复了。现在是每个人都搞砸的部分：通话
【销售系统】↓↓↓
## 如何成交客户（医生式方法）
新手在三个地方丢单：通话前、通话中、通话后
以下是全部三个环节的修复方法
通话前
规则 1：绝对不要在通话前报价格
在私信中报价 = 只按价格被评判，他们还看不到价值
在演示之后报价 = 与他们刚刚看到的价值进行对比
展示 1 万美元的价值，然后说出 2 千美元，他们的大脑会自动替你算账
如果他们在文字中逼你报价：“取决于项目范围，这正是这 15 分钟通话要谈的”
规则 2：在每个渠道都进行确认 + 提醒
人们会忘记自己预约过，感到尴尬，然后消失。用有节奏的跟进杀死这种情况：
立即确认： "已确认，[星期] [时间]。日历邀请已发送"
每3天： "别忘了[星期]的约定！正在为您准备一些东西"
前一天： "明天见的小提醒。我根据您的实际业务做了一个演示，
很期待给您展示"
提前1小时： "[时间]见"
提前5分钟： "刚结束另一个通话，5分钟后见"
```
提到演示是故意的。爽约一个有人专门为你做了东西的会议，会让人感到愧疚。愧疚感是一种提高出席率的策略
规则3：永远不要在私信里推销
文字信息里信息有限，完整价值要在电话里呈现
通话过程中
医生不会一开口就说“买这个处方药”
他们会问哪里疼，先诊断，然后才开药
你说话占20%，他们占80%，而你的20%主要是提问
为什么有效：当客户自己说出痛点时，他们是在说服自己。你无法通过争辩推翻别人自己得出的结论
问题阶梯：
```
1. "现在新客户通常是怎么找到您的？"
2. "下班后打来的潜在客户电话会怎么处理？"
3. "当客户把您的5条评价和竞争对手的150条放在一起比较时，
您觉得他们会怎么做？"
4. "一个新[工作/病人/客户]对您来说值多少钱？"
5. "您已经尝试过哪些办法来解决这个问题？"
```
第3个问题是关键模式：让他们亲口说出损失
第4个问题让他们在听到你的报价之前就先为解决方案定价
通话框架：
1. 痛点提问（上面的阶梯）
2. “我这边可能正好有解决方案，我们先看看是否真的适合”
3. 展示演示，屏幕共享，控制在10分钟以内
4. 报价。一个数字，干脆利落
5. 然后闭嘴。报价后的沉默正是成交发生的地方。不要去填补它
6. 处理异议（通常是伪装成问题的疑虑：“如果没用怎么办？”），回答，成交
还有道德底线：如果他们不需要，就直说。一家有500条评价的商家不需要你的评价系统。一句诚实的“您不需要这个”能带来的转介绍，比任何成交技巧都多
成交机制
- 总是在当前通话中预约下一次通话。日历 = 承诺。“趁我们现在还在通话，我马上把邀请发给您”
- 总是在通话中拿到一个承诺
- 1美元Stripe技巧：创建一个1美元/年的循环付款链接。“我们先把卡挂上，只有1美元，我会退给您”。之后你收取真实金额时零阻力
还有转化利器，概念验证成交法：
预付款 + 满意保证
“您付款启动，我把整个东西做出来，您看到成品，如果不满意：全额退款，不问任何问题”
消除100%的风险，传递十足的信心
如果你的活儿够好，基本上永远不会退款，而你的成交率会飙升
通话之后
- 在第一个项目上超额交付。在这个阶段，你的口碑就是业务本身
- 与每个客户设置双周一次的例行回顾通话。代理商平均每月流失20-30%的客户，回顾通话能把你降到约5%。30%流失率意味着客户只留3个月，5%意味着留几年
- 在这些通话中进行向上销售，绝不做冷推销：网站客户 → 评价自动化 → AI电话客服 → 广告
- 转介绍，两条规则：要具体（“您认识水管工或电工吗？”胜过“您认识什么人吗？”，大脑无法搜索“任何人”）并给出被介绍收入20%的佣金
我 promised 的 Claude 技巧
给 Claude 接入一个 AI 会议记录 MCP（Fathom 或类似工具）。它会录制并转录每一次通话
通话结束后，输入一个提示词：
```
查看我在[时间]与[客户]的通话记录
提取：他们的业务背景、提到的每一个痛点、
我们约定要构建的内容、时间线和预算信号
将完整的项目规格写入 clients/[客户名]/
并提出构建计划
```
Claude 读取会议内容，撰写规格文档，并且可以在你吃午饭之前就开始构建
客户说的话和实际构建的东西之间零信息丢失
客户成交，资金到位。现在开始交付，而这就是 Claude 把你变成一个团队的地方
[ 交付级别 1 ] ↓↓↓
## 网站（你的第一个产品）
约1000万美国小企业没有网站。他们永远不会去学 Claude Code，他们忙着铺屋顶和补牙。这个缺口就是你的工作
市场的收费水平：
```
模板自由职业者 $1-3k
精品工作室 $3-10k
正规代理商 $10-30k
中档代理商 $30-80k
顶级代理商 $80-250k+
你现实的首单价格 约$500
你第一年的目标 $3-10k
```
首单500美元的网站不是贱卖。这是一笔交易买到一份客户见证、一个作品集案例和一条未来向上销售的路径
构建过程
使用一个高级网站技能（自己做一个或从技能仓库抓一个）
/build-premium-website，然后回答访谈：行业、品牌调性、颜色、服务
不到10分钟，一个生产级的 React + Tailwind 网站就在本地跑起来了
然后通过截图迭代：粘贴看起来不对的地方，用大白话描述，Claude 修复
什么让它可销售（而不只是好看）
- 首屏醒目的 CTA，带一键拨号号码
- 社会证明板块：评价、前后对比照片
- 联系表单接入他们的 CRM 或邮箱
- 加载快，动画用代码实现
- 移动端已检查（检查元素 → 切换设备视图）
大多数设计师做的网站好看但不能转化。你做的是能转化的网站。这就是你的卖点
发布上线，免费
1. 把代码推送到 GitHub 仓库
2. Vercel：新建项目，粘贴仓库，部署
3. 约30秒后就在免费子域名上上线
4. 稍后通过任意域名注册商绑定自定义域名
这个阶段的托管成本：$0
作品集策略
本周为5个不同行业构建5个模板网站。虚构的企业也行
一个作品集网站展示全部5个
潜在客户从“挑选”代替“想象”，而挑选比想象更快促成成交
这些模板也是你的 Upwork 作品集和你的第一批成果帖
然后，在你做完第二个真实网站之后：把这个流程编码成一个技能
交付时间从几天压缩到一个下午
你刚刚创造了你的第一个员工
网站打开大门。自动化让钱每个月持续进来
[ 交付级别 2 ] ↓↓↓
## 自动化（持续收入从这里开始）
常见的默认建议是用拖拽式构建器。别用它们
改用基于代码的开源自动化运行器，原因如下：
- Claude 写代码比点 UI 更在行
- 代码可版本管理、可测试，并保存在你的 GitHub 中
- 开源 = 可以托管在任何地方，甚至客户自己的服务器上
- “在你的服务器上运行”这句话能让企业客户多付钱
第三方授权使用 OAuth 代理（Composio 风格），这样 Claude 永远不会接触原始的 Google 凭据。你运行一个脚本，点两个链接，连接就生效了
一个完整案例：发票自动化
填表 → 生成 PDF 发票 → 通过 Gmail 自动发送 → 归档到 Drive
提示词模式：
```
I want to automate my invoicing process
Flow: I fill a simple web form, it builds the invoice,
exports a PDF, emails it to the client via Gmail,
and archives a copy to Google Drive
Use two skills:
1. the composio skill for all third-party authentication
2. the trigger.dev skill to build and host the task
Auto-generate invoice numbers by date. Keep all secrets
in a .env file, nothing hardcoded
Plan mode first: ask me your clarifying questions before building
```
记住这个模式，它适用于所有场景：目标 + 要用的技能 + 约束条件 + 计划模式
Claude 会提出澄清问题（字段、触发方式、邮件风格），你回答、批准计划，它就开始构建
然后配置好：
```
runner API key (from the automation platform)
auth broker key (from the OAuth service)
Anthropic API key (for AI-generated email text)
Drive folder ID (from the folder URL)
business details (logo URL, bank details, currency, net terms)
```
调试循环
第一次测试总会有东西出错。循环如下：
“查看日志，我的测试成功了吗？”
Claude 读取运行日志，找到问题，修复它，你重新运行
反复直到全部通过。你从不需要自己打开堆栈跟踪
这个循环占了现在“懂技术”的 80%
把任务部署到生产环境，把表单部署到 Vercel，搞定
一个安全警告：一个暴露的端点连着你的 API 密钥 = 半夜三点有陌生人在烧你的额度。任何公开的东西都必须加鉴权（完整模式在下一节）
本文最重要的思想
构建一次。技能化。永久交付
任何一次成功的构建之后：
```
Create a skill from what we built this session
Make it GENERIC: any mini automation with a form front end,
a background task back end, and OAuth-brokered integrations,
not invoicing-specific
The skill must interview the user about their use case first,
and must research before building
Make it fully standalone: no references to this project's paths
Verify it's standalone before finishing
```
然后双重检查：“这个技能是独立的吗？它有没有在任何地方引用发票应用？”
你的技能文件夹 = 你的产品目录 = 这门生意真正的知识产权
每个技能都是一个永远免费工作的员工
单个自动化能卖几百美元。打包成系统能卖几千美元
[ 交付级别 3 ] ↓↓↓
## 完整 AI 系统（1 万美元以上档位）
完整 AI 系统 = 5-25 个自动化，全部整合在一个客户登录的控制台中
大多数代理公司交付的是客户从来不打开的“意大利面条式”流程
你交付的是一个产品：本月执行次数、节省的小时数、已回复的工单
正是这个控制台让月度服务费感觉像软件，而不是一张账单
技术栈：
```
front end Next.js + Tailwind + shadcn components
auth magic links via email, no passwords,
restricted to the client's email domain
back end code-based automation runner (open source)
database MongoDB (built-in vector search for RAG)
hosting Vercel (front) + runner cloud (back),
or the client's own servers for enterprise
```
有效的构建顺序
1. 先搭骨架，零设计
仪表盘页面、自动化侧边栏、设置、登录
规则：把坏的改好比一开始就做好难得多，但结构先于美观
2. 用前端技能做设计美化
“给这个应用来一次改头换面，[风格]调性，支持浅色和深色模式”，然后通过截图反复迭代
3. 安装第一个自动化
你已经做好的发票构建，在控制台内换上品牌
4. 入职自动化，几分钟搞定
用客户信息生成的合同 PDF + 日历链接邮件 + 入职调查表
一个触发器，三封邮件，新客户在你睡觉时完成入职
5. 王牌功能：AI 支持工单收件箱
它做什么：
- 每 10 分钟监控一个共享支持收件箱
- 基于客户知识库做 RAG（他们粘贴文档、PDF、网址）
- 自动回答约 60% 的重复性工单
- 每个回答都附带置信度评分 + 引用来源
- 低置信度工单升级到人工队列，并且预先准备好一份草拟回复
- 分析页面：已回复工单、升级率、节省小时数
让客户买单的经济学
```
a support rep costs $2-4k/mo
a 10-person team cut to 5 saves $15k/mo
you charge $2-5k/mo
client still saves $10k+/mo
nobody churns from that math
```
而且同一个系统只需换一个新知识库就能卖给下一个客户。第二次销售 = 几乎纯利润
上线 + 技能化
私有 GitHub 仓库、Vercel 部署、在托管控制台设置所有环境变量，并把鉴权 URL 从 localhost 更新为正式域名（每个人都恰好会忘一次）
然后是最终技能化：把整个脚手架编码成一个"new-client-system"技能，它会询问客户名称、邮箱域名和品牌颜色
每个未来客户在第一天就能拿到一个运行中的控制台，只花几分钟
这一刻，它不再是自由职业，而是一家公司
如果你在想“我永远做不出这个”，下一节会删掉这个借口
[ 元技能 ] ↓↓↓
## 零经验如何构建任何东西（四步框架）
1. 构建计划（设计规格）
你要构建的东西 + 技术栈 + 设计与用户体验
技术栈决定了成本、速度、可扩展性以及日后扩展的难度
与 Claude 反复来回讨论复杂项目，动辄就是好几个小时
这份文档决定了项目的成败
2. 实施计划
怎么做：将构建计划拆分为带复选框的步骤
一个复杂项目大约有 40 个步骤：文件结构、数据库、认证、功能，按顺序排列
强制顺序能强制做出好的决策
3. 构建
一次只喂一个步骤："从第 1 步开始"
在步骤之间进行确认，不要一次性把整个计划全部丢出去
在大型项目上使用子智能体执行以提速
4. 测试与完善
你一定会遇到 bug。把每个错误原封不动地粘贴给 Claude
最后再进行设计改造的环节
黄金比例法则
80% 的时间投入到第 1 步和第 2 步
20% 投入到第 3 步
大多数人的做法正好相反，然后纳闷为什么 AI"做错了东西"
给我六个小时砍树，我会用四个小时磨斧头
零经验作弊码：双模型验证
一个 AI 提出技术栈和计划
另一个会话进行审计："这是一个拟议的架构，这真的是正确的选择吗？规模化后会出现什么问题？"
顾问 + 审计员。他们能互相发现盲点
这就是零基础背景下做出高级工程师级决策的方法
当你不知道某个澄清性问题的答案时，"你会推荐什么？"永远是被允许的回答
你启动任何项目的提示词：
```
I want to build [thing]
First: a build plan. Ask me questions about how it should
work, then write the design spec as a .md file
Recommend the tech stack and justify it: cost to run,
speed, scalability
Don't write any code until I approve the spec
```
框架才是真正的技能。技术栈只是细节
你可以构建任何东西。现在，让我们确保你能按相应的价值收费
[ 定价 ] ↓↓↓
## 如何为 AI 服务定价（5 倍法则）
一切事物的黄金法则：
让每个客户获得其付费金额 5 倍的回报
每月为他们节省 1 万美元，就收每月 2000 美元
为什么有效：
- 客户每个月都会重新算这笔账，然后重新决定留下来。要让这笔账无可辩驳
- 低流失率会累积成转介绍
- 足够多的转介绍 = 负流失：机构无需主动开发就能增长
计算价值
路径一，节省的人力成本：
```
support team: 10 reps at $3k/mo
your system cuts it to 5
value created: $15k/mo
your price at 5x: $3k/mo
client keeps: $12k/mo, forever
```
路径二，新增收入：
只按新增收入的毛利定价，绝不按现有收入
系统每月产生 10 万美元新增收入，约 4 万美元毛利 → 你的定价约每月 8000 美元
三种定价结构
1. 结果导向（毛利百分比，约 20%）
天花板最高，随客户规模无限扩展
要求可追踪性：真正的 CRM、干净的归因
一个结果导向的客户可以产生六位数的 LTV，而月费制只能带来 1.2 万美元
法则：平庸就用月费制，优秀就用结果导向
当价值无法追踪（如一个客服小部件）时，退回到按量计价（每通对话约 1 美元）
2. 预付 + 周期性费用（默认选项）
示例：2000 美元预付 + 每月 500 美元服务费
预付款创造承诺。不付一分钱的客户会玩消失
周期性费用以支持、监控、修复、改进为理由
务必搭配满意度保证
十个这样的客户 = 在没有任何新销售的情况下，每月 5000 美元的保底收入
3. 用量分级（大客户保险）
像软件一样：100 次 / 1000 次 / 5000 次 / 以上自定义
保护你免受那些月付 500 美元却实际打了 1 万通电话的客户的影响
简洁至上
困惑的人不会购买
一个价格，而不是一长串细目
99% 的情况下把软件成本打包进你的价格（600 美元一口价胜过"500 美元 + 100 美元工具费"）
例外：想自己拥有并托管一切的企业客户，他们自己支付工具费用，并支付高得多的服务费
而这条法则高于所有定价法则：
糟糕："一个对你的知识库做 RAG 并带置信度评分升级机制的 AI 智能体"
出色："你的客服收件箱能自动回复，每月为你节省 1 万美元"
没人买 AI。甚至没人买自动化。人们买的是更多的钱、更多的时间，以及更少的员工去做枯燥的工作
再加一个引擎，整个循环就闭合了
[ 增长飞轮 ] ↓↓↓
## 展示你的作品（复利循环）
每个客户项目都是你已经付费的内容资产
你为提案录制的 loom 视频：是内容
仪表盘的截图：是内容
客户网站的前后对比：是内容
客服收件箱本月达到 63% 自动回复率：是内容
这个循环：
为客户构建 → 发布证明 → 证明带来 inbound 线索 → inbound = 无需人脉连接、无需发送成本、无需任何冷启动 → 更好的客户 → 更好的项目 → 更好的证明 → 循环往复
发布什么："用 Claude 十分钟做出这个"的屏幕录像、前后对比图、真实数据，以及你的 skills 文件夹的组织结构图（很 meta，但有效，问我怎么知道的）
我在 10 个平台上运行这套完全相同的发布系统，无需手动撰写任何内容。完整拆解在我的 Content Engine 文章中，它是本文的完美配套
你的机构和你的受众相互滋养。大多数人只建其一。两者都要建
## 结语（该做什么）
1. 今天就搭建你的操作系统：文件夹 + Claude Code + skills + GitHub 备份。20 分钟，现在就做
2. 明天为一家虚构企业构建你的第一个模板网站，免费部署上线
3. 这个周末：创建你的自由职业者主页，并启动一个小型冷启动推广活动
4. 每天投入 4 小时做客户开发。营销和销售才是这门生意的真正技能，而不是 AI。这是所有人都低估的部分，也是大多数人失败的原因
5. 免费或低价接下你的前几个客户。你买到的是客户见证、信任和练习机会，而不是收入
6. 达到覆盖生活开支的每月 2000-3000 美元。商业的成功标准是活下来。十家公司有九家死在这一步，而不是死在扩张阶段
7. 然后提高价格。踢掉挑剔烦人的客户。更少的客户，更高的报酬
8. 找到瓶颈，解决它，再找下一个。先是获客线索，然后是电话，然后是交付，然后又是获客线索。你会永远停留在这第二步上。这就是这份工作的本质
还有一些诚实的脚注：
你不必辞职。早上5点到9点，头脑清醒时搞事业，9点到5点上班，晚上再干两个小时，周末继续
增长是指数型的，同时也很波动：几个月平淡无奇，然后突然通了，然后又下滑，然后迎来更大的突破
一人公司是你起步的方式，而不是上限。Claude 基本上可以独立支撑10-15个客户
另外，别过度自动化。有时候雇一个人来打电话，胜过花三周时间搭一个平庸的自动化流程。这话即使是从我嘴里说出来的
过去，一家公司意味着一办公室的人
现在，它是一个文件夹、一份订阅，以及一个拒绝放弃的人
如果你喜欢这篇文章并觉得有价值，请点赞 + 转发 ❤️
如果这篇文章收集到2000+个赞，我会在后续推文中放出来自其中的完整技能包（提案撰写器、营销活动搭建器、网站搭建器、新客户系统）
现在，是时候去把它做出来了，伙计们 😎
## 相关链接
- [Ronin](https://x.com/DeRonin_)
- [@DeRonin_](https://x.com/DeRonin_)
- [187K](https://x.com/DeRonin_/status/2076690611399176506/analytics)
- [$3-10k](https://x.com/search?q=%243-10k&src=cashtag_click)
- [$2-15k](https://x.com/search?q=%242-15k&src=cashtag_click)
- [$1-5k](https://x.com/search?q=%241-5k&src=cashtag_click)
- [$10-20k](https://x.com/search?q=%2410-20k&src=cashtag_click)
- [$2-5k](https://x.com/search?q=%242-5k&src=cashtag_click)
- [$2-15k](https://x.com/search?q=%242-15k&src=cashtag_click)
- [6月29日](https://x.com/Mho_23/status/2071293386724982947)
- [108K](https://x.com/Mho_23/status/2071293386724982947/analytics)
- [$10k+](https://x.com/search?q=%2410k%2B&src=cashtag_click)
- [$2-3k](https://x.com/search?q=%242-3k&src=cashtag_click)
- [升级到 Premium](https://x.com/i/premium_sign_up)
- [2026年7月13日 晚上11:29](https://x.com/DeRonin_/status/2076690611399176506)
- [18.76万次浏览](https://x.com/DeRonin_/status/2076690611399176506/analytics)
- [查看引用](https://x.com/DeRonin_/status/2076690611399176506/quotes)
---
*导出时间: 2026/7/14 09:40:33*
