Discover how a Toss PO built a custom in-house ERP and AI agents using Claude Code. Learn the step-by-step automation strategy that transformed a 9-person co...
How to Build an AI-Native ERP System: CEO Yoon's Complete Guide
Core Insights
- Custom ERP beats expensive SaaS: Building your own ERP system costs less than 50 won per month compared to thousands in subscription fees
- AI-powered automation scales fast: Using Claude Code, new features take under 30 minutes to create, enabling rapid iteration
- Centralized data is the foundation: All company data—meetings, emails, projects, contacts—feeds into one system that AI agents can access and analyze
- Slack integration creates seamless workflows: Adding an AI agent to Slack allows employees to manage tasks, schedules, and projects without leaving the chat app
- From 2 features to 20+ modules: Start minimal with basic project and customer management, then add features on-demand as pain points emerge
The Birth of Yoon Secretary: From Pain Points to Innovation
CEO Yongseung Yoon, founder of Yoonjadong and former Product Owner at Toss, spent four years building and perfecting his company's internal ERP system. The journey didn't start with grand ambitions. Instead, it began with frustration.
When Yoon's automation agency grew from solo operation to nine employees, managing 35+ concurrent projects became chaotic. The team tried Notion, but encountered critical limitations: poor mobile experience, slowdowns with large datasets, and difficulty managing complex relational data. More importantly, no off-the-shelf solution perfectly matched their specific workflow requirements.
"Since I was at a large company, I always noticed that chairmen and CEOs have secretaries," Yoon explained during a recent podcast. "I thought, 'If I had someone like that—an AI secretary—I could work so much more conveniently.' So I decided to create one." He named it Yoon Secretary (윤 비서), combining his name with the title to instantly communicate its purpose.
The turning point came when Yoon realized that modern AI performance had finally made custom development feasible. "Even two or three years ago, anyone could imagine wanting something like this, but the resources required were too extensive," he reflected. "Now, with recent advancements in AI, these things can be built immediately if we just decide to."
This shift in thinking transformed Yoon from a subscriber to a builder. Instead of renewing expensive ERP subscriptions, he spent evenings coding with Claude Code and gradually replaced costly SaaS tools with custom-built alternatives.
The Architecture: How Yoon Secretary Connects Everything
The genius of Yoon Secretary lies in its integration philosophy. Rather than forcing all data into a single monolithic database, Yoon kept specialized tools doing what they do best—Google Drive for files, Gmail for email, Google Calendar for scheduling—and created connectors that pull everything into one searchable interface.
Project and Client Management: The Core Foundation
The system manages over 120 active projects with details on clients, deliverables, timelines, and assigned team members. When a new project launches, all relevant information—contracts, meeting recordings, communications, and deliverables—flows into a unified project view.
The genius element: every team member can see project history instantly. There's no email chain to search, no forgotten Slack conversation. All context is preserved and searchable, dramatically reducing onboarding time when team members join mid-project.
Meeting Intelligence: From 700+ Recordings to Actionable Data
Yoon's team invests in Plaud AI recorders—devices that capture every client meeting automatically. Over 150 days, the team accumulated 700+ recordings averaging 3.5 hours per day.
Rather than let these recordings sit in Plaud's cloud, Yoon automated the entire pipeline:
- Plaud → Zapier → Yoon Secretary: Every recording automatically transfers to the internal system
- Claude Code integration: Ask Claude any question about meeting content—it instantly retrieves relevant recordings and synthesizes insights
- Quote generation: When a client inquires about pricing, the system searches past meeting notes to generate context-aware quotes
This single integration eliminates hours of manual transcription and search. When a sales question arises, the AI has already read all relevant conversations and can provide accurate, contextual advice.
Business Card Automation: From Stack to Address Book
Most companies store business cards in a drawer or desk organizer. Yoon solved this with optical character recognition (OCR) and API integration:
- Scan the card: Take a photo of any business card
- Automatic entry: OCR extracts contact information (name, phone, email, company, address)
- Dual sync: Information appears in both the internal address book AND automatically syncs to the team member's phone contacts via the Google Contacts API
- Zero additional cost: Using Google's free APIs costs roughly 50 won per month
The result: employees never ask "Do you have this person's number?" Instead, they open their phone and the contact is already there.
Email Management: Taming the Inbox With AI Agents
One employee complained about constant context-switching: reviewing a project required reopening Gmail to find related email threads. This simple pain point spawned a comprehensive email management system.
The system now handles:
- Integrated email view: Search and manage emails from multiple employees' accounts in one place
- Automatic classification: Claude Code reads incoming emails and categorizes them as: reply immediately, flag for review, or archive
- Smart replies: For routine inquiries, Claude composes responses in the CEO's tone and voice, asking for approval before sending
- Security controls: Each team member can choose which emails appear in the shared system (company-only communications, not personal messages)
Yoon no longer reviews every incoming email. Claude filters, prioritizes, and handles routine messages automatically. "I used to be very stressed by the volume of emails," Yoon shared. "Now, I don't have to look at emails that don't need my attention."
For complex decisions, Claude asks: "Should I reply 'yes' or 'no'?" The CEO responds briefly—sometimes just a single word—and Claude sends the full, professional response.
Sales and Financial Management: From Manual Invoicing to Automated Tax Integration
Managing small business finances traditionally requires juggling multiple tools. Yoon's system consolidates sales, invoicing, and expense tracking into one interface.
Automated Invoice Generation
When a sale closes, the system can automatically issue tax invoices (통장 영수증) directly to Korea's Hometax system:
- Sales data is entered into the project system
- Click "Issue Sales Tax Invoice"
- Customer information auto-populates from the contact database
- Fill in details and click "Request Issuance"
- Tax invoice instantly appears in Hometax (no manual submission required)
This eliminated hours of administrative work. Employees can now self-issue invoices without waiting for the CEO's approval.
Real-Time Deposit Tracking
Most business owners rely on checking their bank app frequently. Yoon automated this:
- Bank app sends push notifications when deposits arrive
- The system captures these notifications via API
- Deposit details appear automatically in the financial dashboard
No manual entry. No spreadsheets. All deposits logged automatically.
Future: Expense Automation
The next phase involves:
- Corporate card transaction linking
- Receipt OCR and categorization
- Automatic tax filing preparation
- Real-time P&L visibility
When complete, Yoon won't need to scramble before tax season—all documentation will be pre-organized.
Slack Integration: Making AI Accessible Without Opening Another App
The insight: forcing employees to check Yoon Secretary in a web browser creates friction. Instead, Yoon added a Slack bot that understands natural language commands.
Real-World Example: Creating Complex Schedules
The CEO needed to schedule irregular meeting patterns: Monday/Wednesday one week, then Tuesday/Thursday the next week, mixing in multiple sessions daily. Rather than manually clicking calendar fields, Yoon simply messaged the bot in Slack:
"From tomorrow until next Friday, only Monday, Wednesday, Friday, from 3 PM to 5 PM, create a test schedule. The location is Seoul Station."
The bot parsed this natural language instruction, created all schedule entries, and confirmed completion in seconds.
Task Management Through Chat
Similarly, Yoon uses Slack for task creation:
- "Yoon Secretary, create a follow-up task for the ABC project"
- "What do I need to do today?"
- "Summarize today's progress"
The bot retrieves information from the central database and responds instantly. Notifications go to relevant team members automatically.
Channel-Specific Permissions
Different teams need different data access. Yoon configured:
- Marketing channel: sales and lead data (no salary information)
- Finance channel: all expense data (no project details)
- Development channel: project and deployment data (no sales forecasts)
Confidential information stays protected while maximizing team autonomy.
Claude Code: The Strategic AI Partner
Beyond routine task automation, Yoon uses Claude Code as a strategic advisor. This involves uploading company data—meeting transcripts, email exchanges, sales figures, project portfolios—and asking for analysis and advice.
Strategic Analysis Use Cases
Sales Strategy: After losing a deal, Yoon uploads the proposal and meeting transcript, asking: "Why didn't this close? What should I change for the next meeting?"
Schedule Optimization: "Analyze my schedule for last week and this week. How much work did I actually accomplish? Should I reduce next week's commitments?"
Leadership Development: "What are my habitual speech patterns? Any words I overuse? What tone issues came up in this meeting?"
Conversation Analysis: "Review my sales call with Company X. What objections did they raise? What did I miss?"
This transforms Claude Code from a productivity tool into a personalized consultant. The key: all company data is centralized, so Claude has complete context.
The Economics: Why Custom Beats SaaS
Most small-to-medium businesses subscribe to multiple SaaS tools:
- ERP system: $500-1,500/month
- Project management: $200-400/month
- CRM: $300-600/month
- Document management: $100-300/month
- Email management: $0-200/month
Total annual cost: $15,000-35,000+
Yoon Secretary's actual monthly cost: 50 won (approximately $0.04).
This dramatic difference stems from:
- Free/affordable APIs: Google Drive, Gmail, Google Calendar, Hometax all offer free APIs
- Claude Code usage: Roughly 50 won per month for API calls
- No infrastructure costs: Uses Google Workspace for file storage (already subscribed)
- Time investment, not money: Yoon invested 2-3 months building it; ongoing maintenance is minimal
The ROI equation is simple: 2-3 months of CEO time versus 3+ years of SaaS subscriptions.
Moreover, Yoon never has to request features or wait for vendor roadmaps. When a pain point emerges, he builds a solution in 30 minutes.
The Expansion Story: From 2 Features to 20+
Yoon's development strategy contradicts traditional software engineering: start minimal, iterate obsessively.
Month 1: MVP (Minimum Viable Product)
- Project Management
- Customer Management
- Launch and distribute to 9 employees
Month 2-3: Quick Wins (30 min each)
- Task Management (with Kanban board)
- Schedule Management
- Email Management
- Business Card Management
Month 4-6: Integration Features
- Slack bot
- Hometax integration
- Google Drive linking
- Gmail threading
Month 7-8: Intelligence Layer
- Claude Code integration
- Email classification and auto-reply
- Daily progress reports (GitHub-based)
- Sales analysis and forecasting
Ongoing: One-off Solutions
- Certificate of employment generation (created in 30 seconds)
- Invoice issuance (fully automated)
- Schedule synchronization with Google Calendar
- Contact sync with employee phones
The genius: each feature solves a specific pain point. Yoon doesn't build features just because they could be useful. He only automates things that are genuinely inconvenient.
Automation Philosophy: The "Only If Inconvenient" Principle
Yoon's core belief: "I only automate things that are inconvenient. If it's not inconvenient, I don't automate it."
This discipline prevents scope creep and ensures every feature has a clear ROI. For example:
- Business cards: Genuinely inconvenient to manage manually → automate
- Daily schedules: Easy to check Google Calendar → don't automate
- Email replies: Tedious, time-consuming, error-prone → automate aggressively
This philosophy extends to his team management. Instead of daily status reports, Yoon has Claude analyze GitHub commits and project updates automatically at 6 PM every day. The result: teammates no longer waste time writing reports; the system extracts the information from their work.
Building a Truly AI-Native Company
What distinguishes Yoon's operation from typical businesses isn't just technology—it's a fundamental shift in how work gets done.
The Data-Centralization Principle
Traditional companies scatter data across tools. Yoon consolidated everything—meetings, emails, contacts, projects, financial data—into a single queryable system.
This enables his secret weapon: AI agents that actually understand context.
When an employee asks Claude Code for sales strategy advice, the AI doesn't operate on hunches. It analyzes actual conversation logs, project outcomes, and customer feedback. The quality of advice improves exponentially with data availability.
The Automation Mindset
In most companies, when someone identifies inefficiency, they create a workaround or document a process. Yoon's team asks: "Can we automate this?"
Examples:
- Follow-up reminders: Traditionally, people write reminders in their to-do lists and often forget. Now, when Claude Code sends a proposal, it automatically creates a "check on response" task one week later, with notifications if the deadline passes.
- Daily digests: Instead of relying on the CEO to remember checking project status, Yoon Secretary sends a daily 6 PM summary of all work completed.
- Routine tasks: Anything that repeats gets a cron job or scheduled Claude analysis.
The Experiential Shift
The result: employees spend more time on actual work, less time on process management. The system remembers, tracks, and alerts them. They focus on what only humans can do: creative thinking, relationship building, and strategic decision-making.
Future Direction: SaaS vs. Custom Solutions
As Yoon's reputation for automation expertise grew, other companies began requesting Yoon Secretary. This raised a strategic question: should he productize it as SaaS, or offer custom implementations?
SaaS Advantages:
- Scalable
- Recurring revenue
- Reduced support burden per customer
SaaS Disadvantages:
- Generic features satisfy no one completely
- Each company has unique needs
- Support and customization costs explode
Custom Implementation Advantages:
- Perfect fit for each client
- Higher customer satisfaction
- Stronger competitive moat
Custom Implementation Disadvantages:
- Limited scalability
- Resource-intensive
- Requires skilled developers on staff
Yoon's current strategy: pilot programs with existing contacts, then decide.
If custom implementations prove more effective than SaaS, Yoon would offer his automation expertise as a service—building bespoke ERPs for other companies, similar to his automation agency. This leverages his core competency (automation engineering) rather than forcing him into a software vendor model.
Practical Lessons for Your Company
1. Start With Your Biggest Pain Point
Don't build an entire system at once. Identify one thing that genuinely wastes time or causes errors, then automate it. Yoon started with project management. Build from there.
2. Centralize Data First, Automate Second
You can't meaningfully use AI agents if data is scattered. Invest time in creating a single system where emails, meetings, projects, and contacts feed into one place.
3. Use Free APIs Aggressively
Most major platforms offer free APIs: Google, Stripe, OpenAI, Anthropic. Combining these can replace $500/month SaaS subscriptions.
4. Make AI Accessible Where Work Happens
Don't expect employees to open a portal for routine tasks. Add bots to Slack, email, or wherever communication already happens.
5. Iterate Based on Complaints
Your team will tell you what's inconvenient. Listen and build. A 30-minute feature that saves everyone 5 hours per month compounds into massive ROI.
6. Measure the Real Cost of SaaS
Count not just subscription fees but also:
- Time spent learning each tool
- Data sync costs and friction
- Switching costs if you change providers
- Feature requests that never materialize
Custom development often beats SaaS purely on total cost of ownership.
7. Treat Claude Code Like a Consultant
Don't just ask it to execute tasks. Feed it your company data and ask strategic questions. It's most valuable as an advisor, not just an executor.
Conclusion
CEO Yongseung Yoon's Yoon Secretary represents a new model for business operations. Rather than adapting workflows to fit expensive enterprise software, Yoon built software that perfectly matches how his company actually works.
The transformation didn't require a massive engineering team or years of development. Two to three months of CEO time, 50 won per month in actual costs, and a clear automation philosophy were enough to create something that large enterprises pay six figures annually to achieve.
The broader lesson: in 2026, the competitive advantage isn't having the right SaaS subscription—it's having the discipline and ingenuity to build exactly what you need, when you need it. The tools and AI capabilities now exist. The only question is whether you'll use them.
If your company is still manually entering data, writing status reports, and juggling disconnected tools, now is the time to consolidate, automate, and build your own AI-native system. Yoon Secretary demonstrates it's not only possible—it's economically rational.
The future belongs to companies that become their own software vendors. Start building.
원문출처: 토스 PO 출신 대표가 사내 ERP와 AI 에이전트를 직접 만들고 AI 네이티브 컴퍼니로 거듭난 방법 (윤용승 대표, 윤자동)
powered by osmu.app