Intelligent Agent

Shadai's intelligent agent enables autonomous workflows with custom tools through a powerful plan → execute → synthesize pattern.

What is an Agent?

An agent is an AI system that can:

Plan: Decide which tools to use

Execute: Run tools autonomously

Synthesize: Combine results into coherent answers

The agent automatically:

Decides to use search_database

Infers arguments from prompt

Executes the tool

Synthesizes results

Agent Workflow

Traditional Approach (Manual)

Agent Approach (Autonomous)

Agent automatically:

Determines tool sequence

Infers all arguments

Handles tool execution

Combines outputs

Creating Custom Tools

Method 1: Using @tool Decorator

Key Points:

Use @tool decorator

Add docstring (agent uses this!)

Type hints required

Return string

Method 2: Manual Tool Definition

Use @tool decorator - it's simpler and automatically generates schemas!

Tool Design Best Practices

✅ Good Tool

Why Good:

Clear, descriptive name

Detailed docstring

Type hints

Optional parameters

Returns formatted string

❌ Bad Tool

Why Bad:

No decorator

No docstring

No type hints

Vague name

Returns non-string

Agent Capabilities

1. Tool Selection

Agent automatically chooses appropriate tools:

2. Argument Inference

Agent infers arguments from prompt:

3. Multi-Step Execution

Agent chains tools automatically:

4. Result Synthesis

Agent combines tool outputs into coherent response:

Real-World Examples

Example 1: Customer Support

Example 2: Market Research

Example 3: Data Pipeline

Advanced Patterns

Pattern 1: Conditional Execution

Pattern 2: Error Handling in Tools

Pattern 3: Async Tools

Limitations

1. Tool Complexity

Works Well:

Challenging:

Solution: Break into smaller tools

2. Ambiguous Prompts

Clear:

"Search users with revenue > $50k, limit 10, email results to boss@company.com"

Ambiguous:

"Do something with users"

Solution: Be specific in prompts

3. Tool Count

Optimal: 3-10 tools

Too Many: 50+ tools (agent gets confused)

Solution: Group related functionality, provide clear descriptions

Debugging Agents

Check Tool Execution

Validate Tool Descriptions

Make sure docstrings are clear:

Best Practices

✅ Do This

❌ Don't Do This

Next Steps

Custom Tools Deep Dive

Tool Orchestration

Custom Agent Examples

Key Takeaway: Agents automate complex workflows. Design clear tools, write specific prompts, let the agent orchestrate!

What is an Agent?#

Agent Workflow#

Traditional Approach (Manual)#

Agent Approach (Autonomous)#

Creating Custom Tools#

Method 1: Using @tool Decorator#

Method 2: Manual Tool Definition#

Tool Design Best Practices#

✅ Good Tool#

❌ Bad Tool#

Agent Capabilities#

1. Tool Selection#

2. Argument Inference#

3. Multi-Step Execution#

4. Result Synthesis#

Real-World Examples#

Example 1: Customer Support#

Example 2: Market Research#

Example 3: Data Pipeline#

Advanced Patterns#

Pattern 1: Conditional Execution#

Pattern 2: Error Handling in Tools#

Pattern 3: Async Tools#

Limitations#

1. Tool Complexity#

2. Ambiguous Prompts#

3. Tool Count#

Debugging Agents#

Check Tool Execution#

Validate Tool Descriptions#

Best Practices#

✅ Do This#

❌ Don't Do This#

Next Steps#