6.2 Setup & Your First Agent: From Zero to Working AI Assistant¶

What You'll Accomplish¶

By the end of this section, you'll have AgentKit installed and a working agent that can help with a real business task.

The Assembly Line Analogy¶

Think of building your first agent like setting up a new workstation in a factory: - First: Install the equipment (AgentKit setup) - Second: Train a new worker (configure your agent) - Third: Give them their first assignment (test functionality) - Fourth: Watch them work independently

Prerequisites Check¶

Before we start, ensure you have: - ✅ OpenAI account with API access - ✅ Computer with internet connection - ✅ Basic file navigation skills (from Module 0) - ✅ Email access for verification steps

Step 1: AgentKit Account Setup¶

Creating Your AgentKit Account¶

Visit the AgentKit Portal
Go to https://agentkit.openai.com
Click "Get Started"
Link Your OpenAI Account
Choose "Sign in with OpenAI"
Authorize the connection
Verify your email address
Choose Your Plan
Start with "Developer Free Tier"
Provides 100 agent actions per month
Perfect for learning and testing

Understanding Your Dashboard¶

When you first log in, you'll see:

┌─ AgentKit Dashboard ─────────────────────┐
│ ┌─ Quick Start ─┐  ┌─ Templates ──────┐ │
│ │ • New Agent   │  │ • Email Helper    │ │
│ │ • From Template│  │ • Data Processor │ │
│ │ • Import      │  │ • Task Manager   │ │
│ └───────────────┘  └──────────────────┘ │
│ ┌─ Recent Activity ─────────────────────┐ │
│ │ No agents created yet                 │ │
│ └───────────────────────────────────────┘ │
└─────────────────────────────────────────┘

Step 2: Understanding Agent Components¶

The Three Essential Parts¶

Every agent needs these components:

1. Identity (Who)¶

Name: What you call your agent
Role: What job they perform
Personality: How they communicate

2. Instructions (What)¶

Primary task: Main responsibility
Secondary tasks: Supporting activities
Constraints: What they cannot do

3. Tools (How)¶

Input sources: Where they get information
Output destinations: Where they send results
Processing capabilities: What they can do with data

Step 3: Building Your First Agent¶

Choosing Your First Project¶

Start with something simple but valuable: - ✅ Email summarizer (good first choice) - ✅ Task organizer - ✅ Meeting note formatter - ❌ Complex data analysis (save for later) - ❌ Customer-facing interactions (too advanced)

Example: Building an Email Prioritizer Agent¶

Let's create an agent that helps you prioritize your daily emails.

Step 1: Create New Agent¶

Click "New Agent" on your dashboard
Choose "Start from Scratch" (not template yet)
Fill in basic information:

Agent Name: Email Priority Assistant
Description: Helps prioritize daily email inbox
Category: Productivity

Step 2: Define Agent Identity¶

Role: Professional Email Analyst
Personality Traits:
  - Concise and clear
  - Focuses on urgency and importance
  - Professional tone
  - Actionable recommendations

Step 3: Write Agent Instructions¶

## Your Primary Job
Analyze emails in my inbox and categorize them by priority:
- HIGH: Requires immediate action (same day)
- MEDIUM: Important but can wait (2-3 days)
- LOW: Informational or non-urgent (when time permits)

## For Each Email, Provide:
1. Priority level (HIGH/MEDIUM/LOW)
2. One-sentence reason for priority
3. Suggested action (Reply, Schedule, File, etc.)
4. Estimated time needed

## Guidelines:
- Emails from my boss = always HIGH priority
- Customer complaints = HIGH priority
- Internal meetings = MEDIUM priority
- Newsletters/promotions = LOW priority
- Consider deadlines mentioned in content

## Output Format:
Create a simple priority list I can review in 2 minutes each morning.

Step 4: Configure Tools and Permissions¶

For this first agent, we'll use: - Input: Email reading capability - Output: Text summary report - Permissions: Read-only access to inbox

Tools Enabled:
  - Email Reader (Gmail/Outlook)
  - Text Generator
  - Priority Classifier

Security Settings:
  - Read-only access
  - No email sending capability
  - No attachment downloading

Step 5: Set Automation Schedule¶

Trigger: Every weekday at 8:00 AM
Action: Scan inbox from last 24 hours
Output: Send priority report to me
Duration: Stop after 30 days (review period)

Step 4: Testing Your Agent¶

The Test Run Process¶

Start Small: Give your agent 5-10 test emails
Review Results: Check if priorities make sense
Adjust Instructions: Refine based on what you see
Scale Up: Let it process your full inbox

What to Look For¶

Good Signs ✅¶

Priorities match your expectations
Reasons make logical sense
Format is easy to read
Completes tasks within time limit

Warning Signs ⚠️¶

Inconsistent priority assignments
Takes too long to process
Output format is confusing
Missing important context

Making Adjustments¶

Common first-agent improvements:

Before: "Categorize my emails"
After: "Categorize emails using these specific criteria..."

Before: "High priority emails"
After: "HIGH: Needs response within 4 hours"

Before: "Send me a report"
After: "Send a 5-item bulleted list with action items"

Step 5: Monitoring and Iteration¶

Week 1 Checklist¶

Agent runs successfully each day
Priority assignments are 80%+ accurate
You save at least 15 minutes daily
No security or access issues
Output format works for your workflow

Improvement Questions¶

Accuracy: Are priorities correct 8 out of 10 times?
Usefulness: Do you actually use the output?
Efficiency: Does it save more time than it takes to review?
Reliability: Does it run consistently without errors?

Common First-Agent Mistakes¶

Mistake 1: Too Complex Initially¶

Problem: Trying to automate your entire workflow at once Solution: Start with one specific task

Mistake 2: Vague Instructions¶

Problem: "Help me be more productive" Solution: "Categorize my emails by these specific criteria..."

Mistake 3: No Success Metrics¶

Problem: Not knowing if the agent is actually helping Solution: Track time saved, tasks completed, accuracy rate

Mistake 4: Set-and-Forget¶

Problem: Never reviewing or improving the agent Solution: Weekly 10-minute review sessions

Step 6: Preparing for Scale¶

What You've Learned¶

How to translate business needs into agent instructions
The importance of specific, measurable tasks
Testing and iteration cycles
Basic security and permission management

Next Steps¶

Create 2-3 similar simple agents
Learn to use pre-built templates (Section 6.3)
Explore more complex automation scenarios

Troubleshooting Your First Setup¶

Can't Connect to Email¶

Issue: Agent can't access your inbox Solutions: 1. Check OAuth permissions in your email provider 2. Verify AgentKit has necessary scopes 3. Try disconnecting and reconnecting

Agent Doesn't Run on Schedule¶

Issue: No automated reports arriving Solutions: 1. Check your timezone settings 2. Verify trigger conditions are met 3. Look for error messages in agent logs

Poor Quality Results¶

Issue: Agent priorities don't match expectations Solutions: 1. Add more specific examples in instructions 2. Include edge case scenarios 3. Refine your criteria definitions

Security Best Practices for First Agents¶

Start with Read-Only¶

Never give write permissions initially
Test thoroughly before enabling actions
Use sandbox environments when possible

Review Access Regularly¶

Check what data your agent can see
Revoke unused permissions
Monitor for unusual activity

Keep Instructions Private¶

Don't include sensitive information in agent descriptions
Use variables for confidential data
Regular review shared agents for exposure

Your First Agent Success Framework¶

Day 1: Setup and Basic Test¶

Install and configure agent
Run manual test with 3-5 items
Verify basic functionality

Week 1: Daily Monitoring¶

Check agent output each morning
Note accuracy and usefulness
Make small adjustments as needed

Week 2: Optimization¶

Analyze patterns in results
Refine instructions based on learning
Consider adding new capabilities

Week 3: Expansion Planning¶

Identify next automation opportunity
Plan second agent project
Document lessons learned

Ready to explore pre-built templates? In Section 6.3, we'll discover how to use AgentKit's ready-made solutions to fast-track your automation journey.