B.C.L. · OPC CO-CREATION PLATFORM

AI + OPC · OPC = ONE-PERSON COMPANY

FROM AN MCN TEAM
TO ONE AI ORGANIZATION

AI + OPC = AI SUPER ORGANIZATION

20

1

付济嘉 / James · Founder, B.C.L. · Alibaba Visit 2026

01 / 10

My name is 付济嘉.

You can also call me James.

I am the founder of B.C.L. AI Company, an OPC co-creation platform that helps founders and small businesses grow and achieve their goals at every stage.

First of all, thank you for having me today.

I am very happy to meet all the incredible founders here.

Thank you also to Alibaba for giving China's OPC community this opportunity to share.

I know many of you have already built profitable companies and have practical experience with AI.

So I come here with respect and humility.

In Chinese, we have an expression called “pao zhuan yin yu.”

It means offering a simple brick first, with the hope of inviting more valuable ideas, like jade, from others.

My brick today is one change in my own work.

I once needed an MCN team of nearly twenty people.

Now, with AI agents, I can complete much of that team's work as one operator.

OPC means a one-person company.

But with a connected AI system, one person no longer means a small organization.

This is what I mean by an AI super organization.

MCN OPERATOR → OPC CO-CREATOR

付济嘉 / James

Content, founder IP and commercial conversion.

B.C.L. helps founders and small businesses grow and achieve their goals at every stage—usually as a project partner, not only a service provider.

MCN track record

600K+followers built from zero across platforms

¥10M-a-yearbusiness built around one creator IP

~20people in the former MCN team

B.C.L. today

Content marketing builds attention and trust.

Founder IP makes expertise visible.

Financing connections support the next growth stage.

No fixed limit on the IPs and brands the system can incubate.

SkincareStem-cell anti-agingIndependent mediaExpert-led businesses

02 / 10

THE OPERATING CHANGE

NEARLY TWENTY PEOPLE.
NOW ONE OPERATOR.

From team structure
to system structure.

Old MCN structure

~20people

SignalsTrends · creators · topics

ContentWriting · review · assets

DeliveryVideo · publish · report

Problem: many handoffs and knowledge trapped in personal files.

AI-native structure

1operator + connected AI

AgentsOwn defined tasks

SkillsReuse operating methods

MCPConnect tools and data

Human owns: business goal · relationships · facts · final decisions.

From scripting to publishing—all done by a full-stack creator.The workflow is the team.

03 / 10

In the old MCN structure, different people followed trends, studied creators, planned topics, wrote scripts, reviewed content, searched for materials, produced videos, published posts, and prepared reports.

The team needed nearly twenty people because every handoff required time and coordination.

Important knowledge often stayed inside one person's files or experience.

When someone left, part of the operating memory left as well.

[Pause]

Today, I do not use one model to imitate twenty employees.

I redesign the workflow itself.

Specialized agents perform defined tasks, reusable skills provide the method, and MCP connections give the system access to tools and data.

From scripting to publishing—all done by a full-stack creator.

I still own the business goal, relationships, fact checks, and final decisions.

The system gives one operator the execution capacity of a much larger content team.

THREE CONNECTED BUSINESS WORKFLOWS

THE AI CONTENT FACTORY

Agents do the work. Skills carry the method. MCP connects live tools and data.

HUMAN CONTROL

J

James

One operator sets the direction and signs off the result.

Business goalWhat must change

Brand positionWho we serve

Final decisionFacts, risk and release

WORKFLOW 01

Discover

Positioning and network

◎

Positionbrand or founder

⌕

Searchmarket and creators

▦

Classifybenchmark accounts

≋

Screenscore and deduplicate

▣

Build assetsintelligence + network

WORKFLOW 02

Create

Daily content production

◷

Scan08:03 and 20:03

◆

Selecttopic and IP fit

¶

Writeevidence-linked draft

7

Challengereview panel

✓

Approverevise + human

WORKFLOW 03

Deliver

Operations and learning

▧

Collectmaterials and cover

⇄

Routemodel + prompt + cost

✓

Checkvideo and assets

↗

Publishtrack every post

↻

Learnperformance to next brief

DIGITAL EMPLOYEE AGENTS

DISCOVER CREWCreator Discovery · Profile · Screening · Benchmark Analysis · Research

CREATE CREWWorkflow Monitor · News & Trend · Topic Fit · Writer · 7-Agent Review

DELIVER CREWMaterials · Cover · Video Router · Asset QA · Publishing · Performance

SHARED OPERATING MEMORY

SkillsPositioning · scoring · writing · QA

MCPCreator data · browsers · ASR · model and platform APIs

Local + FeishuPrivate files and logs · shared assets, versions, reviews and dashboards

04 / 10

This is the operating system behind the one-person team.

The first workflow starts with a brand or founder position.

Creator Discovery, Creator Profile, Creator Screening, Benchmark Script Analysis, and Research and Materials Agents search the market, classify benchmark accounts, and build two assets.

One asset supports positioning with content patterns, hooks, audiences, and commercial models.

The other becomes a creator network for future cooperation and continuous learning.

The reusable Skills include positioning frameworks, creator classification, audience-fit scoring, script-DNA extraction, deduplication, and contactability rules.

MCP connections give those Agents access to Huitun creator data, browsers, Douyin, Xiaohongshu, speech-to-text, Feishu Bitable, and local SQLite.

The second workflow runs at 8:03 in the morning and 8:03 in the evening.

It scans news, selects topics by IP fit, verifies facts and heat, writes scripts, and sends them through seven review Agents before human approval.

The third workflow finds materials, generates covers, routes AI video tasks, reviews assets, publishes content, and tracks performance.

Monthly and quarterly reviews send the learning back into positioning and topic selection.

Private files, raw media, SQLite, and logs stay in the local control layer.

Creator assets, script versions, reviews, and dashboards support collaboration in Feishu.

This is a business workflow map, not a collection of disconnected AI tools.

WORKFLOW 01

SIGNAL IN.
ONE CLEAR BRIEF OUT.

No huge folder. Just the customer, position, creators and topics.

SCATTERED MARKET SIGNALS

◷News

⌕Creators

¶Scripts

▧Assets

FILTERbusiness question + evidence

ONE ACTIONABLE BRIEF

WHOCUSTOMER

WHEREPOSITION

WHATTOPICS

POSITIONING LIBRARYHooks, audiences, angles

CREATOR NETWORKPeople worth learning from

◎

MonitorSchedule and sources

↗

TrendSelected signals

▦

BenchmarkHooks and structure

≋

ScreenFit and deduplicate

▣

ResearchFacts and material

05 / 10

WORKFLOW 02

SEVEN AI EXPERTS
DEBATE EVERY SCRIPT

The goal is simple: publish what the audience actually loves to watch.

SCRIPT DRAFTNOT FINAL

HOOK?RISK?PROOF?

L

LeoQwen

Hook + PaceOpening speed and audience retention.

P

PeterGLM

Business ValueCommercial logic and user value.

C

ChloeDoubao

ClarityNatural language and comprehension.

E

EthanDeepSeek

Facts + LogicEvidence and technical reasoning.

Lu

LucasGemini

Real VoiceFeels human, not like an ad.

F

FrankQwen3

ChallengeOpposing view and trust risks.

V

VictorKimi K3

ExecutionCan this become a useful action?

CLAIMSEVIDENCE ATTACHED

DEBATESEVEN ANGLES

REVISEBETTER SCRIPT

FINALFACT CHECK + HUMAN SIGN-OFF

06 / 10

Next comes content and quality control.

The Writing Agent creates the first script and keeps evidence next to every important claim.

To make this work, I actually created a virtual team of 7 AI experts.

I gave each of them a unique persona, a specific English name, and backed them up with different AI models.

They literally sit in a virtual room and debate with each other.

They audit every piece of my content from different angles.

My only goal? To make sure that whatever we publish is exactly what the audience loves to see.

Leo, using Qwen, checks the hook and pace. Peter, using GLM, checks business value. Chloe, using Doubao, checks natural language and clarity. Ethan, using DeepSeek, checks facts and technical logic. Lucas, using Gemini, checks whether the script feels real, not like an ad.

Frank, using Qwen3, argues against the script and looks for trust risks. Victor, using Kimi K3, checks execution and commercial value.

These are seven review angles. They do not replace my judgment.

I check the latest facts, balance the different views, and approve the final version.

The Publication Review Agent then checks the cover, caption, product facts, brand rules, and final format.

WORKFLOW 03

MAKE IT.
PUBLISH IT. LEARN.

The system improves only after real data comes back.

Make

Route each scene to the right model.

Quality
Accuracy
Time
Cost

Publish

Check the full package before release.

Cover
Caption
Facts
Format

Learn

Turn real response into the next brief.

Watch time
Comments
Leads
Sales

MONITORING AGENT schedules regular reviews and catches execution problems

↻

REAL WORK MAKES THE SYSTEM BETTER Lessons return to the next topic, next script and next campaign

07 / 10

REAL OPERATING EVIDENCE

DOCUMENTS THAT KEEP
THE WORK VISIBLE

IDENTIFIERS REMOVED

Creator-screening reportANONYMIZED · AUTO-SCROLL

Anonymous ID	Category	Follower range	Avg. likes range	Status
Creator A01	Fishing	10M+	100K+	Shortlist
Creator A02	Fishing	5M–10M	25K–50K	Shortlist
Creator A03	Fishing	2M–5M	10K–25K	Shortlist
Creator A04	Fishing	2M–5M	2.5K–5K	Shortlist
Creator A05	Fishing	2M–5M	100K+	Shortlist
Creator A06	Fishing	2M–5M	100K+	Shortlist
Creator A07	Camping / fishing	2M–5M	100K+	Shortlist
Creator A08	Fishing	2M–5M	Under 1K	Review
Creator A09	Fishing	2M–5M	25K–50K	Shortlist
Creator A10	Fishing	2M–5M	2.5K–5K	Shortlist
Creator A11	Fishing	2M–5M	10K–25K	Shortlist
Creator A12	Fishing	2M–5M	5K–10K	Shortlist

Anonymous ID	Category	Follower range	Avg. likes range	Status
Creator A01	Fishing	10M+	100K+	Shortlist
Creator A02	Fishing	5M–10M	25K–50K	Shortlist
Creator A03	Fishing	2M–5M	10K–25K	Shortlist
Creator A04	Fishing	2M–5M	2.5K–5K	Shortlist
Creator A05	Fishing	2M–5M	100K+	Shortlist
Creator A06	Fishing	2M–5M	100K+	Shortlist
Creator A07	Camping / fishing	2M–5M	100K+	Shortlist
Creator A08	Fishing	2M–5M	Under 1K	Review
Creator A09	Fishing	2M–5M	25K–50K	Shortlist
Creator A10	Fishing	2M–5M	2.5K–5K	Shortlist
Creator A11	Fishing	2M–5M	10K–25K	Shortlist
Creator A12	Fishing	2M–5M	5K–10K	Shortlist

Recent script headlinesFULL ENGLISH TITLES · AUTO-SCROLL

RECENT SCRIPT 01

My Entire Content Team Is Made of AI Digital Employees

RECENT SCRIPT 02

Fei-Fei Li’s Atlas: One Photo, One Parallel Universe

RECENT SCRIPT 03

Jensen Huang Says AGI Is Here. Forbes Calls It a Marketing Gimmick.

RECENT SCRIPT 04

Behind the Largest IPO in History, the People Building AI Are Leaving First

RECENT SCRIPT 05

Everyone Says the AI Bubble Is About to Burst. Capital Is Selling Something Else.

RECENT SCRIPT 06

Everyone Is Comparing AI Video Models. ByteDance Quietly Changed the Game.

RECENT SCRIPT 07

A Unicorn Burned Millions on AI in Months, Then Found It Could Not Explain the Bill

RECENT SCRIPT 08

SpaceX Beat Expectations in Its First Financial Report. Its Private-Market Shares Fell 12% as Investors Closest to the AI Boom Sold First.

RECENT SCRIPT 09

Karpathy Gave Opus 5 Two Hours to Write 5,500 Lines of Code. It Built an Unplayable Game, and Everyone Missed the Point.

RECENT SCRIPT 01

My Entire Content Team Is Made of AI Digital Employees

RECENT SCRIPT 02

Fei-Fei Li’s Atlas: One Photo, One Parallel Universe

RECENT SCRIPT 03

Jensen Huang Says AGI Is Here. Forbes Calls It a Marketing Gimmick.

RECENT SCRIPT 04

Behind the Largest IPO in History, the People Building AI Are Leaving First

RECENT SCRIPT 05

Everyone Says the AI Bubble Is About to Burst. Capital Is Selling Something Else.

RECENT SCRIPT 06

Everyone Is Comparing AI Video Models. ByteDance Quietly Changed the Game.

RECENT SCRIPT 07

A Unicorn Burned Millions on AI in Months, Then Found It Could Not Explain the Bill

RECENT SCRIPT 08

SpaceX Beat Expectations in Its First Financial Report. Its Private-Market Shares Fell 12% as Investors Closest to the AI Boom Sold First.

RECENT SCRIPT 09

Karpathy Gave Opus 5 Two Hours to Write 5,500 Lines of Code. It Built an Unplayable Game, and Everyone Missed the Point.

Real creator-screening structure and recent working-script titles. Creator names, account IDs, locations, personal notes and exact metrics removed.

08 / 10

RECORDED OPERATING CASES

CAPACITY UP.
COORDINATION COST DOWN.

One monthly brand / creator-IP case

180scripts produced

Same monthly case

66selected as shoot-ready

Cumulative screening runs

2,897creator profiles processed

Cumulative screening runs

472usable or contactable candidates retained

RMB 50K+previous monthly labor reference

RMB 50K–100Kprevious base service fee

→

1 operatorcore workflow + AI tools

RMB 20K–30Kproject-dependent; can cover 1 brand + 2 founder-IP accounts

17 keyword groups combined and deduplicated in the displayed report.Different case levels are labeled separately · No capacity or pricing guarantee.

09 / 10

These figures come from our operating records, and they describe different recorded cases.

In one monthly brand or creator-IP case, the system produced one hundred and eighty scripts.

We selected sixty-six as ready to shoot.

Across recorded creator-screening runs, the system processed two thousand eight hundred and ninety-seven profiles and retained four hundred and seventy-two usable or contactable candidates.

The displayed report combined seventeen keyword groups and removed duplicates.

The previous monthly labor reference was more than fifty thousand renminbi.

Today, one operator can run the core workflow with AI tools.

Previous base service fees were about fifty thousand to one hundred thousand renminbi.

Our current range is about twenty thousand to thirty thousand, depending on the project.

This range can even cover a content matrix with one brand account and two founder-IP accounts.

These are recorded examples, not capacity or pricing guarantees.

RETURN TO THE OPENING

ONE CAPABLE FOUNDER · A MUCH LARGER OPERATING FORCE

AI+OPC=AI SUPER ORGANIZATION

NO CEILING

Industry knowledgeCustomer understandingCommercial judgmentFast iterationConnected AI execution

“This is my modest brick, and I look forward to learning from your experience.”

付济嘉 / James · B.C.L.

10 / 10

Let me return to the change at the center of this talk.

I once needed nearly twenty people to run an MCN content operation.

Today, one operator can coordinate research, writing, review, production, publishing, and learning through a connected AI system.

This does not make business easy.

AI lowers the cost of execution, but it also increases the speed of competition.

The founder still needs industry knowledge, customer understanding, commercial judgment, and the ambition to improve quickly.

AI multiplies those abilities.

B.C.L. brings this approach into projects as a co-creation partner.

For a large company, the same structure can connect specialists and preserve knowledge across departments.

For an OPC, it can turn one capable founder into a much larger operating force.

AI plus OPC equals an AI super organization with no ceiling.

This is my modest brick, and I look forward to learning from your experience.

Thank you.

FROM AN MCN TEAMTO ONE AI ORGANIZATION

FROM AN MCN TEAM
TO ONE AI ORGANIZATION