# ๐Ÿข Lobster University: One-Person Company in Practice (One Person, One Team) > **Use case**: You're a solo developer, freelancer, or small-team founder who wants AI to build a complete virtual company team โ€” product, design, engineering, marketing, sales, ops, HR, and customer support all covered. **Just describe your business goal in Telegram, and the Lobster breaks it down, assigns tasks to the right AI specialists, and delivers actionable results.** [agency-agents](https://github.com/msitarzewski/agency-agents) is an open-source collection of specialized AI agent personalities, featuring **144 professional AI agents** across **12 functional divisions**. Each agent has its own identity, core workflows, technical deliverables, and success metrics โ€” not generic prompt templates, but battle-tested professional roles. Combined with OpenClaw's Telegram channel, you can send business goals directly from your phone and let the Lobster automatically dispatch the right agents to collaborate โ€” from requirements analysis to code implementation, from marketing strategy to deployment. Truly **run a company as one person**. --- ## 1. What You'll Get (Real-World Value) Once set up, you'll have a **24/7 AI company team** at your command: ### Scenario 1: From Idea to MVP - **Problem**: You have a product idea, but handling requirements, design, development, and deployment alone is overwhelming - **Solution**: Describe your goal in Telegram โ€” agency-agents automatically breaks it into product requirements โ†’ UI design โ†’ frontend/backend implementation โ†’ deployment, with each step handled by the corresponding AI specialist ### Scenario 2: Quickly Build a Marketing System - **Problem**: Product is built, but you can't write copy, don't understand SEO, and have no marketing strategy - **Solution**: The Marketing division's 31 agents handle content creation, social media management, SEO optimization, and email marketing ### Scenario 3: Solo Client Project Delivery - **Problem**: You've taken on a client project that needs product design, frontend/backend development, testing, and documentation - **Solution**: Dispatch the complete delivery pipeline on demand: Product โ†’ Design โ†’ Engineering โ†’ Testing โ†’ Support ### Scenario 4: Business Decision Support - **Problem**: Unsure which direction to pursue; need market research, competitive analysis, and user personas - **Solution**: Product division's Trend Researcher + Marketing division's market analysts collaborate to output structured decision references --- ## 2. Why agency-agents? ### Core Architecture: 12 Divisions ร— 144 AI Specialists ```mermaid flowchart TB subgraph COMPANY["๐Ÿข Your AI Company"] direction TB subgraph BIZ["Business Line"] PRODUCT["๐Ÿงญ Product
5 agents"] SALES["๐ŸŽฏ Sales
8 agents"] MARKETING["๐Ÿ“ฃ Marketing
31 agents"] PAID["๐Ÿ’ฐ Paid Media
7 agents"] end subgraph TECH["Tech Line"] ENGINEERING["๐Ÿ”ง Engineering
24 agents"] DESIGN["๐ŸŽจ Design
8 agents"] TESTING["๐Ÿงช Testing
8 agents"] GAME["๐ŸŽฎ Game Dev
20+ agents"] end subgraph OPS["Ops Line"] PM["๐Ÿ“‹ Project Mgmt
6 agents"] SUPPORT["๐ŸŽง Support
6 agents"] ACADEMIC["๐Ÿ“š Academic
5 agents"] SPECIAL["โšก Specialized
30+ agents"] end end USER["๐Ÿ‘ค You"] --> |"Send goal via Telegram"| COMPANY ``` ### Every Agent Is a "Specialist," Not a "Template" The traditional approach is writing generic prompts and having AI role-play. agency-agents is different โ€” each agent includes: | Field | Description | |-------|-------------| | **Identity & Personality** | Unique persona that influences communication style and decision-making | | **Core Workflows** | Battle-tested standardized workflows | | **Technical Deliverables** | Explicit deliverables checklist (not just concepts) | | **Success Metrics** | Measurable success criteria | | **Memory & Learning** | Pattern recognition for continuous improvement | ### Division Overview | Division | Agents | Capabilities | |----------|--------|-------------| | **Engineering** | 24 | Frontend/backend dev, DevOps, database optimization, security audits, API design | | **Marketing** | 31 | Content creation, SEO, social media, email marketing, platform ad strategy | | **Design** | 8 | UI/UX design, brand visuals, accessibility audits | | **Sales** | 8 | Outbound strategy, customer discovery, deal strategy | | **Product** | 5 | Product management, trend research, behavioral design | | **Testing** | 8 | QA, performance benchmarking, accessibility testing | | **Project Mgmt** | 6 | Project management, process optimization, resource scheduling | | **Support** | 6 | Customer service, analytics, finance, compliance | | **Paid Media** | 7 | PPC, search ads, programmatic buying | | **Game Dev** | 20+ | Cross-engine development, platform adaptation | | **Academic** | 5 | History, psychology, anthropology (world-building) | | **Specialized** | 30+ | Blockchain auditing, Salesforce architecture, and more | ### How It Differs from Plain Prompts | Feature | Plain Prompts | agency-agents | |---------|--------------|---------------| | **Expertise** | Depends on your prompt skill | Pre-built battle-tested workflows | | **Collaboration** | Single role | Multi-agent collaboration with auto-routing | | **Deliverables** | Freeform output | Standardized deliverables + success metrics | | **Reusability** | Rewrite each time | Register as skills, invoke on demand | | **Scalability** | Manual coordination | Unified registry + routing engine | > **Key insight**: The value of agency-agents isn't just "better prompts" โ€” it's a **registerable, composable, routable company function system**. Through OpenClaw's skill registration mechanism, these agents become standardized capabilities that can be dispatched at any time. --- ## 3. Configuration Guide: Building Your AI Company ### 3.1 Prerequisites | Requirement | Description | |-------------|-------------| | OpenClaw installed and running | Base environment ready | | Telegram account | For interacting with OpenClaw | | LLM API Key | Supports OpenAI / Claude / DeepSeek / local models | | Tool profile set to coding/full | Requires command execution permission, see [Chapter 7](/en/adopt/chapter7/) | ### 3.2 Configure the Telegram Channel Same as other Lobster University tutorials โ€” set up the Telegram bot first. If you've already configured it in [Automated Research in Practice](/en/university/vibe-research/), skip to 3.3. **Step 1: Create a Telegram Bot** Open Telegram, search for `BotFather`, select the official account with the blue verification badge, click **Start**, then send `/newbot`: BotFather will ask two questions: 1. **Bot display name** โ€” can be anything, e.g., "Lobster" 2. **Bot username** โ€” must end with `bot`, e.g., `HelloClawClaw_bot` Full conversation example: ```text You: /newbot BotFather: Alright, a new bot. How are we going to call it? Please choose a name for your bot. You: Lobster BotFather: Good. Now let's choose a username for your bot. It must end in `bot`. You: HelloClawClaw_bot BotFather: Done! Congratulations on your new bot. Use this token to access the HTTP API: 8658429978:AAHNbNq3sNN4o7sDnz90ON6itCfiqqWLMrc ``` > **Important**: Keep this Bot Token safe โ€” you'll need it when configuring OpenClaw. **Step 2: Connect Telegram in OpenClaw** On your OpenClaw host, run the onboard command: ```bash openclaw onboard ``` Skip and continue until you reach the **Select channel** page, choose **Telegram (Bot API)**, and paste your Bot Token. **Step 3: Get Your Telegram User ID** In Telegram, find the bot you just created and send `/start`. The bot will reply with your User ID: ```text OpenClaw: access not configured. Your Telegram user id: 8561283145 Pairing code: 6KKG7C7K Ask the bot owner to approve with: openclaw pairing approve telegram 6KKG7C7K ``` Note the User ID and enter it in the `allowFrom` field. After configuration, select **restart** to restart OpenClaw. ### 3.3 Register agency-agents as OpenClaw Skills With the Telegram channel ready, transform the 144 AI specialists into standardized OpenClaw skills. Send this prompt in Telegram: ```text Read repo: https://github.com/msitarzewski/agency-agents Goal: Restructure it into a set of "company function skills" registerable to OpenClaw, for on-demand invocation and auto-routing. Requirements: 1. Abstract existing agents into standardized skills (e.g., product / marketing / sales / ops / hr / support) 2. Each skill must include: - name / description (clearly searchable) - capabilities (what problems it solves) - inputs / outputs schema - tools / dependencies - prompt template (directly executable) - routing tags (for auto-matching) 3. Design a unified skill registry supporting search and composition 4. Define multi-skill collaboration protocol (task decomposition, state passing, feedback loops) 5. Output as "deployable structure" (directory + config), not explanations 6. Provide at least one end-to-end workflow example (auto-select and invoke multiple skills) Constraint: Prioritize engineering usability and extensibility. Avoid staying at conceptual design only. ``` ![Sending agency-agents skill restructuring prompt to Lobster](./images/opc-skills.jpg) The Lobster reads the entire repository and outputs a complete skill registration structure: ```text agency-skills/ โ”œโ”€โ”€ SKILL.md # Main entry โ”œโ”€โ”€ registry/ โ”‚ โ”œโ”€โ”€ registry.json # Skill definitions + routing tags โ”‚ โ””โ”€โ”€ COLLABORATION.md # Multi-skill collaboration protocol โ”œโ”€โ”€ product/ # ๐Ÿงญ Product Management โ”œโ”€โ”€ marketing/ # ๐Ÿ“ฃ Marketing โ”œโ”€โ”€ sales/ # ๐ŸŽฏ Sales โ”œโ”€โ”€ ops/ # โš™๏ธ Operations โ”œโ”€โ”€ hr/ # ๐Ÿ‘ฅ Human Resources โ”œโ”€โ”€ support/ # ๐ŸŽง Customer Support โ”œโ”€โ”€ engineering/ # ๐Ÿ”ง Engineering โ”œโ”€โ”€ design/ # ๐ŸŽจ Design โ”œโ”€โ”€ workflows/ # End-to-end workflows โ””โ”€โ”€ scripts/router.py # Routing CLI ``` Each skill includes standardized fields (name, description, capabilities, inputs/outputs schema, prompt template, routing tags) and supports auto-routing: ```text $ python scripts/router.py "write PRD for feature X" โ†’ ๐Ÿงญ PRODUCT (matched: feature, PRD) $ python scripts/router.py "build outbound campaign" โ†’ ๐Ÿ“ฃ MARKETING + ๐ŸŽฏ SALES ``` > **Core value**: This step transforms 144 scattered agents into searchable, composable, standardized capabilities. For any subsequent business goal, the Lobster can automatically match and dispatch the right skills. --- ## 4. First Run: From Goal to MVP Demo With skills registered, let's run an end-to-end business scenario โ€” **from a business goal to a working demo website**. ### 4.1 Send Your Business Goal Describe what you want to build in Telegram. Here's an example using a harness engineering training website: ```text Goal: Design and implement a harness engineering training website Demo (MVP) Requirement: Quickly build a runnable, demonstrable minimal system, not a complete product. Tasks: 1. Product structure (minimal closed loop) Design 3 modules: - Learn: Core courses (harness engineering concepts) - Practice: Simple agent harness experiments (e.g., prompt + eval loop) - Evaluate: Evaluate and provide feedback on agent output 2. Course content (minimal set) Provide 3 lessons: - What is harness engineering (definition + architecture) - Agent feedback loop (input โ†’ execution โ†’ evaluation โ†’ iteration) - Runtime scaffolding (tools / memory / routing) Each lesson includes a brief explanation + a runnable experiment 3. Technical implementation (must be runnable) - Frontend: Simple page (Next.js or static HTML) - Backend: Lightweight API (Node / Python) - Agent: Simple LLM call wrapper - Eval: Rule-based or simple scoring function 4. Demo scenario (must work end-to-end) Implement full flow: User input โ†’ agent execution โ†’ eval scoring โ†’ optimization suggestions Constraint: Prioritize simplicity and runnability. All modules must connect to form a closed loop. ``` ![Sending MVP build goal to Lobster](./images/opc-mvp.jpg) ### 4.2 Lobster Auto-Dispatches Skills After sending, the Lobster activates the product-launch workflow, dispatching in sequence: 1. **Product** โ†’ Analyze requirements, define product structure and modules 2. **Design** โ†’ Design page layouts and interaction flows 3. **Engineering** โ†’ Implement frontend and backend code 4. **Support** โ†’ Write usage documentation Within minutes, you'll receive the build completion message: ```text Done! Harness Engineering Training Demo is ready. harness-training-demo/ โ”œโ”€โ”€ README.md โ”œโ”€โ”€ backend/ โ”‚ โ”œโ”€โ”€ app.py # Flask API (5 endpoints) โ”‚ โ”œโ”€โ”€ agent.py # LLM Agent wrapper โ”‚ โ”œโ”€โ”€ eval.py # Rule-based scoring system โ”‚ โ””โ”€โ”€ requirements.txt โ””โ”€โ”€ frontend/ โ””โ”€โ”€ index.html # Single-page app (open directly) ``` | Module | Function | |--------|----------| | ๐Ÿ“š Learn | 3 courses, each with a runnable experiment | | ๐Ÿ”ง Practice | Agent single execution / feedback loop | | ๐Ÿ“Š Evaluate | PRD + marketing copy โ€” two complete demos | ### 4.3 Run and Deploy the Demo Code is generated, but you want to see it in the browser. Just tell the Lobster: ```text Please run it and find a way to let me browse the demo ``` ![Lobster exposes Demo to the internet via Cloudflare Tunnel](./images/opc-port.jpg) The Lobster will: 1. Start the backend service (Flask API on port 8080) 2. Configure frontend static file serving 3. Expose the Demo to the internet via **Cloudflare Tunnel** ```text Done! Demo is running and exposed to the internet: ๐ŸŒ Access link: https://domains-elect-atlantic-completion.trycloudflare.com How to use: 1. Open the link 2. Click any module (Learn / Practice / Evaluate) 3. Experience the full harness engineering flow Backend status: โ€ข Flask API: http://localhost:8080 โœ… โ€ข 3 courses loaded โ€ข Agent + Evaluator ready ``` > **Cloudflare Tunnel**: The Lobster uses the free `cloudflared` service to create a temporary tunnel โ€” no domain or port forwarding configuration needed. The link is valid as long as the tunnel process is running. ### 4.4 View the Demo Open your browser and visit the link to see a complete training website demo: ![Harness Engineering Training Website Demo Page](./images/opc-demo.png) The full end-to-end loop is working: - **Learn** โ†’ Browse course content, understand core concepts - **Practice** โ†’ Input a task (e.g., "Write a product requirements document"), let the Agent execute - **Evaluate** โ†’ Agent output is scored by the evaluation system, returns optimization suggestions > **Note**: This demo is just a minimal runnable prototype. If you get a 404 on first visit, the frontend static files may not be configured yet โ€” just tell the Lobster and it'll fix it. In practice, the value of an MVP is rapid idea validation, not perfection. --- ## 5. Advanced Scenarios: From MVP to Operations ### Scenario 1: Full Product Launch Have multiple divisions collaborate on the entire journey from concept to launch: ```text Goal: Complete launch preparation for [your product name] Need: 1. Product: Finalize PRD and user stories 2. Design: Design landing page and core interactions 3. Engineering: Implement core features and deploy 4. Marketing: Write launch copy, prepare social media content 5. Support: Write user guide and FAQ ``` The Lobster activates the `product-launch` workflow: `product โ†’ design โ†’ engineering โ†’ marketing โ†’ support`, with each stage's output feeding into the next. ### Scenario 2: Quickly Build a Marketing Plan ```text Goal: Design a complete content marketing plan for my SaaS product Need: - Target user personas and pain point analysis - Content calendar (Blog / Newsletter / Social media) - SEO keyword strategy - 3 sample blog articles - Landing page copy ``` The Marketing division's Content Creator, SEO Strategist, and Social Media Manager collaborate to produce a complete plan. ### Scenario 3: Client Project Delivery ```text Goal: Build a data dashboard for client [XXX] Tech stack: React + FastAPI + PostgreSQL Need: 1. Requirements analysis and data model design 2. Backend API implementation 3. Frontend pages and chart components 4. Testing and deployment scripts 5. Delivery documentation ``` Engineering's Frontend Developer, Backend Architect, and Database Specialist work together, with Project Management tracking progress. ### Scenario 4: Competitive Research & Strategy ```text Goal: Analyze product strategies of [Competitor A], [Competitor B], [Competitor C] Need: - Feature matrix comparison - Pricing strategy analysis - User review summary - Differentiation opportunity identification - Action recommendations ``` Product's Trend Researcher + Marketing's market analysis agents jointly produce a structured report. ### Scenario 5: Iterative Improvement Not satisfied with the first delivery? Just add requirements: ```text The training website demo needs improvements: 1. Learn module: add syntax highlighting for code examples 2. Practice module: support multi-turn conversations 3. Evaluate module: add radar chart for evaluation dimensions 4. Overall: add responsive layout for mobile access ``` The Lobster iterates on existing code โ€” no need to start from scratch. --- ## 6. Troubleshooting ### Issue 1: Lobster Doesn't Know How to Use Registered Skills **Common causes**: - Prompt doesn't reference skill names explicitly โ€” add "use the product skill" or "invoke the engineering team" in your follow-up messages - registry.json routing tags aren't precise enough โ€” go back to Telegram and have the Lobster optimize the routing tags **Solution**: ```text List all registered skills and auto-select the right ones for: [your goal description] ``` ### Issue 2: Demo Build Fails or Code Errors **Common causes**: - Missing Python/Node.js dependencies โ€” have the Lobster run `pip install` or `npm install` first - API Key not configured โ€” ensure the LLM API Key is properly set - Port already in use โ€” specify a different port **Diagnostic steps**: ```bash openclaw logs --limit 50 # Check OpenClaw logs ``` ### Issue 3: Cloudflare Tunnel Not Accessible **Common causes**: - Backend service not running โ€” check if the Flask/Node process is alive - Frontend static files not configured โ€” backend needs to serve static files too - Free tunnel is unstable โ€” wait a few seconds and retry, or have the Lobster recreate the tunnel ### Issue 4: Telegram Bot Not Responding **Diagnostic steps**: 1. Verify Bot Token is correct: check in BotFather 2. Verify `allowFrom` includes your User ID 3. Verify OpenClaw has been restarted: `openclaw restart` 4. Check OpenClaw health: `openclaw doctor` --- ## 7. Security & Compliance Reminders ### Reminder 1: Telegram Bot Token Security - **Never leak your Bot Token**: Anyone with the token can control your bot - **Never commit tokens to Git**: Use environment variables or `.env` files - **Rotate tokens regularly**: If you suspect a leak, use `/revoke` in BotFather to regenerate - **Restrict `allowFrom`**: Only allow your own User ID to interact with the bot ### Reminder 2: API Key & Demo Security - Never hardcode LLM API Keys in demo code โ€” use environment variables - Don't repeatedly send API Keys in plaintext via Telegram chat - Demos exposed via Cloudflare Tunnel are **publicly accessible** โ€” don't include sensitive data - Free tunnel links are temporary, but anyone can access them while active ### Reminder 3: Deliverable Review - Code and plans generated by agency-agents are AI output โ€” **always review before production use** - Marketing copy, legal compliance content needs professional review - Demo code is suitable as **prototypes and POCs**; production deployment requires security audits --- ## 8. Summary: One Person's Company, 144 AI Specialists' Team The core value of agency-agents is **turning complete company functions into AI capabilities** โ€” just describe your business goal, and 12 divisions with 144 AI specialists handle the rest: - **Full-function coverage**: Product โ†’ Design โ†’ Engineering โ†’ Testing โ†’ Marketing โ†’ Sales โ†’ Ops โ†’ Support - **Auto-routing dispatch**: Just describe your goal; the Lobster automatically matches the best agent combination - **Standardized delivery**: Every agent has explicit deliverables and success metrics โ€” not vague suggestions - **Mobile control**: Send goals and receive results anytime via Telegram - **Rapid iteration**: Not satisfied? Add requirements and iterate on existing work **Remember**: A one-person company doesn't mean "one person does everything" โ€” it means **one person directs an AI team to do everything**. agency-agents transforms you from executor to decision-maker โ€” you decide the "what" and "why," the AI team handles the "how." ## References ### agency-agents - [agency-agents (AI Specialist Team Collection)](https://github.com/msitarzewski/agency-agents) - [msitarzewski (Project Author)](https://github.com/msitarzewski) ### Telegram - [Telegram BotFather (Create Telegram Bot)](https://t.me/BotFather) - [Telegram Bot API Official Docs](https://core.telegram.org/bots/api) ### Related Tutorials - [Automated Research in Practice (Describe It, Get a Paper)](/en/university/vibe-research/) - [Chapter 7: Tools & Scheduled Tasks](/en/adopt/chapter7/) - [Chapter 4: Chat Platform Integration](/en/adopt/chapter4/)