- 7
- April
From EP.1, you now understand what AI Agents are, what they can do, and how global organizations are using them. In EP.2, we will teach you how to set up your own AI Agent team — from defining roles, choosing tools, designing prompts, to building workflows that actually work. No coding required, no programming background needed. Just follow along.
Series: Can AI Agents Be Team Members? (4 Parts)
- EP.1: Can AI Agents Really Be Team Members? — The AI as Coworker Concept
- EP.2: How to Set Up an AI Agent Team — Step-by-Step Guide (You are here)
- EP.3: Real-World Examples — What Percentage of Work Can AI Agents Handle?
- EP.4: Limitations and Risks — What You Need to Know Before Using AI Agents
Step 1: Define Roles — Assign Responsibilities to Each AI Agent
Just like hiring real employees, the first thing you need to do is define what each AI Agent will be responsible for. Don't create an Agent that does everything — because an AI that does everything usually isn't great at anything. Just as you need to clearly divide work between AI and humans, you need clear role definitions.
| Role | Responsibilities | Data access | Constraints |
|---|---|---|---|
| Content Writer | Write articles, social media posts, emails | Brand guidelines, product info, past articles | No fabricated pricing data |
| Data Analyst | Analyze sales data, generate reports | Sales data, financial data | No access to personal data |
| Customer Support | Answer questions, prioritize tickets | FAQ, product manuals, customer history | Complex issues escalate to humans |
| Code Reviewer | Review code, suggest improvements | Codebase, coding standards | No automatic deployment |
| Research Assistant | Find information, summarize research | Websites, research databases | Must always cite sources |
| Admin Assistant | Schedule meetings, summarize meetings | Calendar, email, notes | No sending emails without approval |
Each role must have four clearly defined elements:
- Name — Give it a memorable name, e.g., "ContentBot" or "DataSensei"
- Responsibilities — Clearly specify what it must do and what it must not do
- Data access — Define what data this Agent can see
- Constraints — What it is not allowed to do, to prevent risks
Step 2: Choose Your Tools
Several platforms are available today for creating AI Agents, each with different strengths:
| Platform | Key strength | Best for | Coding required? |
|---|---|---|---|
| Claude (Anthropic) | Long document analysis, detailed writing, deep thinking | Research, Content, Analysis | No (use Projects) |
| ChatGPT (OpenAI) | Easy Custom GPTs creation, many plugins | General tasks, Customer Support | No (use GPT Builder) |
| Gemini (Google) | Deep Google Workspace integration | Admin, Email, Calendar | No |
| Microsoft Copilot | Works within Microsoft 365 | Organizations using Microsoft | No |
| Custom Agent (API) | Full control, deep customization | Specialized tasks, system integration | Yes (Python, Node.js) |
Start simple: You don't need to write code — ChatGPT Custom GPTs or Claude Projects can create functional AI Agents. Start with simple tasks and expand when you see results.
Step 3: Prompt Design — The Heart of an AI Agent
A prompt is the "instruction" that determines how your AI Agent will work — a good prompt produces excellent results, a bad prompt produces unusable output. The recommended formula is ROLE + CONTEXT + TASK + FORMAT + CONSTRAINTS:
| Component | Description | Example |
|---|---|---|
| ROLE | Who you are, what expertise you have | "You are a senior data analyst" |
| CONTEXT | Background and data needed | "Q1 sales data is in this table..." |
| TASK | What needs to be done, clearly stated | "Analyze sales trends and summarize in 5 points" |
| FORMAT | Output format requirements | "Use tables + bullet points in English" |
| CONSTRAINTS | Limitations, what must not be done | "No fabricated numbers, cite data only" |
Example System Prompt: Content Writer Agent
ROLE: You are a Content Writer for Grand Linux Solution Co., Ltd., an ERP provider
CONTEXT: Write articles about ERP, technology, and organizational management for the grandlinux.com website
TASK: Write articles on assigned topics with 3+ tables, callout boxes, and internal links to other website articles
FORMAT: HTML format, professional English
CONSTRAINTS: No fabricated pricing. Do not claim Saeree ERP has features it doesn't have
Example System Prompt: AI Accountant Agent
Example: System Prompt for an Accounting AI Agent
ROLE: You are a senior accountant specializing in TFRS and IFRS standards
CONTEXT: Help analyze accounting data from ERP systems, including balance sheets, P&L statements, and accounting tasks AI can assist with
TASK: Audit transactions, identify anomalies, summarize analysis results, recommend corrective actions
FORMAT: Tables + explanations with data-backed figures
CONSTRAINTS: No fabricated numbers. Request additional data if insufficient. Always note that tax advice requires professional consultation
Step 4: Build Workflows — Connect Agents Together
The true power of an AI Agent Team lies in connecting multiple agents together into workflows — just like passing work between team members in real life:
Example Workflow: Content Creation
- Research Agent gathers information and summarizes key points
- Writer Agent creates the article from the Research Agent's summary
- Editor Agent checks grammar, adjusts tone, and verifies facts
- Human Review approves before publishing
- Publisher publishes the article (automatic or manual)
Example Workflow: Customer Support
- Triage Agent receives the query and categorizes it (billing, technical, general)
- Support Agent answers basic questions from FAQ and manuals
- Escalation sends complex issues to human agents with a problem summary
- Follow-up Agent checks resolution status and sends satisfaction surveys
Step 5: Test and Iterate
Don't start with critical tasks — begin with small tasks where mistakes won't cause significant damage:
- Start with simple tasks — Such as summarizing documents, drafting emails, searching for information
- Measure results — Compare with the previous approach. Is it better?
- Refine prompts — Identify where AI performs poorly and fix the prompt
- Gradually expand — Increase complexity incrementally
| KPI to measure | How to measure | Target |
|---|---|---|
| Time spent | Compare work time before and after AI | At least 30% reduction |
| Output quality | Human quality rating (1-5 scale) | Average no less than 3.5 |
| Cost | Subscription + API cost per month | Lower than hiring for equivalent work |
| Revision rate | % of work requiring revisions after AI output | No more than 20% |
| User satisfaction | Survey from team members using AI Agents | Average no less than 4/5 |
AI Agent + ERP = A Complete Team
When AI Agents work together with an ERP system, the results multiply:
- Analyze ERP data — AI Agent pulls sales, inventory, and AR data from ERP to analyze and summarize for management
- Automatic report generation — Instead of manual monthly reports, AI Agent extracts ERP data and generates reports with insights
- Anomaly alerts — Stock below reorder point, sales below normal, or overdue receivables
- Help users find information — Instead of clicking through multiple menus, just ask AI "What were last month's sales?"
This is the direction Saeree ERP is developing with its AI Assistant, to provide clients with maximum benefit from AI technology in the near future.
Next EP: Real-World Examples — What Percentage of Work Can AI Agents Handle? will showcase real case studies showing what AI Agents do well, what still requires humans, and the actual results.
Setting up a good AI Agent Team doesn't start with technology — it starts with understanding your team's pain points and using AI to solve them. Start small, measure results, iterate, then expand.
— Sureeraya Limpaibul, Grand Linux Solution
