Former Alibaba P8, after being laid off, earns 170,000 per month with WorkBuddy — I dug up 27 real cases and found that money-makers only used 5 features

Let’s start with a person.
A former Alibaba P8, laid off.
Looked for a job for three months, found nothing.
Then he did something: launched three WorkBuddy instances, each managing a different thing.

Instance A scraped competitor prices, sold to merchants for 300 a month.

Instance B generated ad creatives, 50 per piece.

Instance C handled automated customer service, taking a 5% commission on sales.

Those three instances served over 20 merchants, earning him 170,000 per month.
This is not a joke. It’s the most explosive case among the 27 real cases I found online.

I dug into the remaining 26 cases.
Found one thing: people who actually make money with WorkBuddy only use 5 core features.
Not coding. Not automated deployment. Not some advanced operations.
Just 5 features you can see when you open the software.
But 90% of people have never used a single one.

This article breaks down these 5 features for you — each with real cases, real numbers, real people.
Read it, then decide for yourself if you should act.

▌ Chapter 1: The 5 features money-makers use

Here’s the conclusion.
Among these 27 cases, every money-maker used at least 3 of the following features:

1. Scheduled Automation — it works while you sleep

2. Multi-Agent Parallel — one person commands an AI team

3. Skill Packs — 80,000 plugins, install with one sentence

4. MCP Connector — connect 30+ external services

5. Memory System — gets smarter the more you use it

What do non-money-makers do?
Typing in the chat box: “Write a copy for me,” “Summarize this for me.”
Close it, open it again the next day, start from scratch.

Chat is one-time. Agent is continuous.

That’s the gap.
Let’s break down each one.

▌ Chapter 2: Scheduled Automation — it works while you sleep

This is the most frequently used feature among the 27 cases.

Case 1: Liberal arts student, zero code, AI research investment earns enough for monthly meals

A liberal arts student who can’t code.
He set up 4 scheduled tasks:
8 AM – auto-fetch top 5 tech news, organize into a brief, push to WeChat.
12 PM – pull position-related announcements from Xueqiu (a Chinese investment platform), generate a daily report.
8 PM – aggregate unusual stock movements for the day, flag items needing attention.
Sunday night – generate a weekly tracking report.

Spends only 20 minutes a day reading reports, everything else is automated.

Monthly earnings cover his grocery bill.
His exact words: “I’m not investing — I’m letting AI watch for me while I just make decisions.”

Case 2: 12-person sales team, admin reduced from 45 minutes daily to zero

A company with 12 salespeople needed daily data aggregation for reports.
An admin named Xiao Li used to spend 45 minutes each day manually compiling data.
Then she set up one scheduled task:

5:50 PM – WorkBuddy automatically pulls data from the system, generates a Word brief + Excel table, pushes to Feishu group chat.

15 minutes to set up, permanent automation.
Now Xiao Li leaves work at 5:50 PM sharp, no overtime.

Case 3: Daily news automation, 5 tasks staggered by 3 minutes

A developer set up 5 scheduled tasks for tech, education, environment, world events, and people’s news.
Started with constant failures, now runs steadily, pushing every day at 7 AM sharp.
He summarized three pitfalls:

First, don’t let all 5 tasks run at the same time; stagger them by 3 minutes, or your credits will blow up.

Second, write clear format requirements in the task prompt; otherwise output varies every time.

Third, scheduled tasks won’t execute when the computer is sleeping; either disable sleep or use cloud execution.

He spent a whole month figuring these out himself.
Now every morning when he wakes up, 5 briefs are already waiting in his inbox.

Common thread across these three cases: Set up the task once, it runs forever automatically.
You configure it once, it runs for a lifetime.
That’s the power of scheduled automation — you’re not earning hourly wage, you’re earning compound interest.

▌ Chapter 3: Multi-Agent Parallel — one person commands an AI team

If scheduled automation is “it works even when you’re not there,” multi-agent parallel is “one person does the work of an entire team.”

Case 4: Former Alibaba P8’s 3 WorkBuddy instances, earning 170,000 per month

Back to the opening case.
He didn’t open one WorkBuddy to do everything; he opened three, each managing its own domain:
Instance A: Daily scheduled scraping of competitor prices, generates comparison tables, auto-sends to merchants.
Instance B: Receives product images and requirements from merchants, generates ad creatives (copy + images), delivers.
Instance C: Connects to merchants’ customer service systems, auto-replies to common questions, handles after-sales.

Three instances don’t interfere with each other; each runs its own course. He only monitors quality and collects money.

That’s the essence of a “one-person company” — not that you do everything, but that you command AI to do it for you.

Case 5: 7-person AI editorial team, 10 minutes to produce a 20-page planning document

Someone used WorkBuddy’s multi-expert parallel feature, simultaneously invoking 7 virtual roles:
Marketing operations, product manager, development engineer, finance, copywriter, data analyst, legal.
One sentence: “Create a full plan for a new product launch.”
7 agents start working at the same time, each outputting from their own perspective.

10 minutes later, a 20-page complete planning document appears — including market analysis, product positioning, budget allocation, copy plan, data model, legal risk notes.

All you need to do is review and edit.
This isn’t “AI helps you write”; it’s “AI holds a meeting for you.”

Case 6: Double-Agent unmanned factory — Daji + Hermes

Two developers each run an agent, one named Daji, the other Hermes.
How do they collaborate?

Shared folder as mailbox.

Daji finishes writing some content, saves it to the shared folder.
Hermes detects the new file, automatically reads it, processes it, outputs results, and saves them back.
Naturally decoupled, no interference.
Plus an IMA knowledge base serving as a “single source of truth” — all information follows the knowledge base, both agents read data from the same place.

Result: A fully automated content production pipeline, with humans only responsible for final review.

Common thread across these three cases: Not one AI doing everything, but multiple AIs each doing their own thing, with you as the commander.
One person + multiple agents = one team.
What you save isn’t just salary, it’s management cost.

▌ Chapter 4: Skill Packs — 80,000 plugins, install with one sentence

This is the feature I most want to talk about, and 90% of people have absolutely no idea it exists.

What is a Skill?

In simple terms: Someone has written a certain workflow into a “skill pack.” You install it with one sentence.
Once installed, you don’t need to write a long instruction every time; one sentence triggers the whole process.

People without Skills: Every time they use WorkBuddy, they have to write a long prompt — “Analyze XX for me, note XX, output format is XX…” Then tweak it.

People with Skills: One sentence: “Run a competitor analysis for me.” The Skill automatically loads all steps, formats, and notes.

That’s the gap.

Case 7: Security team triples SRC bounty

A security team used WorkBuddy’s AI Skills to chain a vulnerability discovery pipeline:
Asset mapping → vulnerability scanning → JS information mining → privilege escalation testing.
Used to take a whole day manually; now automated in sequence, averaging double-digit vulnerabilities per month, bounties jumping from four figures to five figures.

Key point: It’s not that AI got smarter; it’s that Skills solidified experienced workers’ processes into repeatable flows.

A new person installs this Skill and directly works to the standard of a veteran.

Case 8: WeChat public account AI daily report, from 90 minutes to 8 minutes

Developer CodeRin used a Skill to solidify the daily report production flow:
Topic selection → scraping → deduplication → summarization → formatting → image matching → publishing.
A daily report used to take 90 minutes; now 8–15 minutes.
In-depth research used to take 3–4 hours; now 30–45 minutes.

He said: “The essence of a Skill is not speed, but turning my experience into a reproducible pipeline.”

Case 9: 7×24 bug-hunting factory

Another security team used enterprise multi-agent parallel, each agent loaded with different Skills, automatically hunting bugs 7×24.
Bounties in five figures per month.
Humans do only one thing: confirm whether a vulnerability is exploitable, submit the report.

How to install a Skill?

In WorkBuddy, find the “Skill Marketplace,” search for the Skill you need, click install.
Or use skill-creator to build your own — write your workflow into a SKILL.md file, then trigger it with one sentence later.

What separates money-makers from non-money-makers? Skills.

Non-money-makers start from scratch every time with a prompt. Money-makers solidify processes into Skills, write once, reuse forever.

▌ Chapter 5: MCP Connector — connect to the outside world

If Skills are “internal process automation,” then MCP is “connecting to the outside world.”

What is MCP?

MCP is WorkBuddy’s bridge to external services. Tencent Docs, Feishu, GitHub, Notion, Tongdaxin (a Chinese stock trading platform), WeCom, WeChat Pay… over 30 external services can be connected.
Once connected, WorkBuddy can directly operate these services — not “generate content for you to post,” but directly post it for you.

Case 10: Tongdaxin MCP quantitative backtesting

Someone used Tongdaxin MCP to connect to a local market data source, pulled 5.8 years of data, and ran an ETF backtesting strategy.
Before, this required writing Python, calling APIs, setting up environments — impossible for non-technical people.
Now, one sentence: “Help me backtest a dollar-cost averaging strategy for the CSI 300 ETF over the past 5 years.”
WorkBuddy automatically pulls data, calculates returns, draws charts, generates a report.

Case 11: WeCom group collaboration — @the AI and it works

A team connected WorkBuddy to WeCom.
In the group, @the AI and it works:
“Help me organize today’s meeting minutes” — auto-reads meeting recordings, compiles a document, sends it back to the group.
“Help me break this requirement into development tasks” — auto-dismantles, generates a requirement document, assigns tasks to people.

The AI isn’t waiting in another window; it’s right in your work group chat.

Case 12: Community group buying — 200 orders sorted in 3 minutes

This is the most down-to-earth case I found.
A community group-buying leader named Amin used to handle over 200 order chains in her WeChat group every day.
Before, manually calculating who wanted what, how many portions, how to sort, how to deliver — it took over an hour just to organize.
Now she throws the chain messages into WorkBuddy, and in 5 seconds they become an order table, automatically breaking out picking lists and delivery lists.
In 3 minutes she finishes what used to take 1 hour.
She doesn’t understand technology, doesn’t know what MCP is — she just knows “throw messages in, table comes out.”

That’s the value of MCP — not making you learn technology, but making technology work for you.

Three cases, three industries, one common thread: AI is not an island; it connects to your real work scenarios.
Connected? It’s a tool. Not connected? It’s just a chat box.

▌ Chapter 6: Memory System — gets smarter the more you use it

This is the most overlooked of the five features, but also the key to widening the gap.

WorkBuddy has a memory system.

What does that mean?
Things you’ve told it — your preferences, your work habits — it remembers.
Next time you chat, it automatically loads those memories; you don’t have to repeat yourself.

Case 13: One sentence fed 4 pieces of info; the system remembered that night

Someone casually said in conversation: “I mainly work in catering, target customers are 25–35 year old white-collar workers, average ticket price 30–50 yuan, focusing on healthy light meals.”
That sentence contains 4 pieces of information: industry, target customer, average ticket price, product positioning.
WorkBuddy automatically extracted them that night and wrote them into a memory file.
After that, every time he asked AI to write copy or do analysis, AI automatically followed those 4 dimensions — no need to repeat “I’m in catering” every time.

Every word you say, it’s learning.

Case 14: Don’t blindly trust Auto for model selection

WorkBuddy supports 15 models; many people always use Auto for automatic selection.
But someone tested and found:
Long document analysis → DeepSeek-V4-Pro is more accurate.
Automation tasks → GLM-5.0-Turbo is faster and more stable.
Creative copy → Claude series is more inspired.

Different tasks, different models — difference of over 30% in results.

He wrote this preference into the memory system. After that, every time a task runs, WorkBuddy automatically selects the best model.

Case 15: Three features combined — exponential explosion

The most powerful cases are those who use three features together:

Scheduled automation + manual model selection + feeding memory.

Every morning at 7 AM, the scheduled task triggers automatically.
WorkBuddy, based on the memory system, knows what you do and what style you prefer.
Based on the task type, it automatically selects the best model.
Generates content and pushes it to you.

What you see when you wake up is not “a draft AI wrote for you,” but “a finished product generated in your style, using the best model, automatically.”

None of what these three people did is technically difficult.
The hard part is — they knew these 5 features existed and combined them.

▌ Chapter 7: I tested them myself for a few days — here’s my honest take

In my previous two articles, I mentioned I only recently started using WorkBuddy deeply.
Over the past few days, I tested all 5 features. Here’s my honest feedback.

Scheduled Automation: It works, but there are pitfalls.

The biggest pitfall: tasks don’t run when the computer goes to sleep. Here in Pattaya, when I close my laptop at night, I wake up to find tasks didn’t run.
Solution: either disable sleep or use cloud execution. But cloud execution costs extra credits.

Multi-Agent Parallel: Powerful, but credits burn faster than expected.

Running 3 agents simultaneously consumes about 3x the credits of a single agent. If you’re on the standard plan, you might run out after two rounds.
Suggestion: not all tasks need parallelism; simple tasks use a single agent, complex ones use multi-agent.

Skill Packs: Easiest to start, fastest results.

Install skill-creator, solidify your article writing process, then trigger the whole workflow with one sentence.
This is the feature I use most right now.

MCP Connector: Configuration has a learning curve, but once set up, it’s a game changer.

I connected Feishu MCP to let WorkBuddy directly push finished articles into Feishu Docs for formatting.
It took me over an hour to find the documentation and configure it, but now it saves 20 minutes of manual copying every day.

Memory System: You need to actively feed it.

It won’t proactively ask “what are your preferences?”; it extracts info naturally from your conversation.
My suggestion: first thing as a new user, chat with WorkBuddy for half an hour, telling it your industry, clients, products, and style. Let it remember.

Summary: Using all 5 features, efficiency noticeably improved. But it’s not perfect — credit consumption, sleep issues, configuration hurdles are real pain points.

No hype, no hate. The tool is good, but you need to spend time troubleshooting.

▌ Chapter 8: What separates money-makers from non-money-makers

After digesting 27 cases, I found a pattern.

Money-makers treat WorkBuddy like an employee.

Non-money-makers treat WorkBuddy like a search engine.
What does treating it like an employee mean?
Give it division of labor — which agent manages what.
Give it processes — solidify workflows with Skills.
Give it permissions — use MCP to connect to your real work scenarios.
Give it memory — tell it who you are, what you do, what you want.
Give it a schedule — use scheduled tasks to make it work 7×24.

What does treating it like a search engine mean?

Open, ask one question, close.
Open again the next day, ask another question, close again.
No Skills, no MCP, no memory, no scheduled tasks.
Start from zero every time.

The gap isn’t in AI; it’s in how you use AI.

None of these 27 people made money because they were technically strong.
The former Alibaba P8 didn’t earn 170K by coding; he earned it through the business model of “3 agents serving 20 merchants.”
Amin didn’t conquer community group buying by being tech-savvy; she did it through the simple action of “throwing chain messages in and getting a table out in 5 seconds.”
The security team didn’t triple bounties because AI got smarter; they did it through the process of “solidifying veteran experience into a Skill.”

What they did was transform AI from a “chat box” into a “workbench.”

This doesn’t require a technical background. What it requires is — action.

▌ Chapter 9: If you only want to do one thing today

Don’t try to adopt all 5 features at once.
Start with one thing: install a Skill.
Open WorkBuddy, go to the Skill Marketplace, search for a Skill related to your work, install it.
Trigger it with one sentence.
Feel the efficiency boost from “process solidification.”
If even this step feels like too much trouble, you won’t need to look at the other 4 features.

AI doesn’t lack features. What it lacks is people willing to spend 10 minutes configuring them.

10 minutes, in exchange for saving 2 hours every day.
You do the math.

Three rules for you who’ve read this far:

Rule 1: Solidify process first, talk about efficiency later.
AI without Skills is just an advanced typewriter — starting from zero every time.

Rule 2: Connect to real scenarios for real value.
AI not connected via MCP is an island; connected, it’s a tool.

Rule 3: Let it remember you.
AI without memory treats you like a stranger every time; feed it, and it gets better the more you use it.

90% of people use the best AI in the country as a chat box.

The remaining 10% have already made it work for them.

Which group will you be in?

Reference sources:
Tencent Cloud Developer Community: “90% of People Use WorkBuddy as a Chat Tool”
53AI: “Skill in Practice: 3 Pitfalls, 1 Case, 4 Tips, 12x Efficiency Boost”
Juejin: “Dual-Agent Collaboration Building an Unmanned Factory”
Toutiao: “WorkBuddy Scheduled Automation 7×24 Unattended”
Zhihu: “2026 AI Agent Enterprise Implementation Explosion Year: 5 Trends”
Tencent Cloud: “Multi-Agent Parallel: 7 Virtual Roles”

Pattaya, a Sunday afternoon. After digging through 27 cases, I decided to start by installing a Skill.