The Four Key Capabilities

AI researcher Andrew Ng identified four fundamental design patterns that give AI agents their autonomous capabilities. Understanding these helps you recognize what agents can—and can't—do.

1. Reflection

What it is: The ability to review and improve their own work.

Imagine an agent writing a report. Instead of just producing a first draft, it can:

• Read what it wrote
• Identify weaknesses or errors
• Revise and improve the output

Business example: A quality assurance agent that reviews customer service responses, identifies tone issues, and suggests improvements—before the message is sent.

2. Tool Use

What it is: The ability to interact with external systems and software.

Agents aren't limited to just generating text. They can:

• Search the web for current information
• Query databases
• Send emails
• Update spreadsheets
• Control other software through APIs

Business example: A research agent that searches your CRM, pulls relevant sales data, queries industry reports, and compiles a competitive analysis—all from a single request.

3. Planning

What it is: The ability to break complex goals into executable steps.

When given a high-level objective, agents can:

• Decompose it into smaller tasks
• Determine the logical sequence
• Adapt the plan when obstacles arise

Business example: An event planning agent that takes "organize a team retreat for 50 people in Q2" and creates a detailed project plan with venue research, catering coordination, activity planning, and budget tracking.

4. Multi-Agent Collaboration

What it is: Multiple specialized agents working together.

Complex tasks often benefit from division of labor:

• A research agent gathers information
• An analysis agent interprets the data
• A writing agent produces the report
• A review agent checks for accuracy

Business example: A content creation pipeline where one agent researches topics, another writes drafts, a third handles SEO optimization, and a fourth manages publication scheduling.

Why All Four Matter

These capabilities compound each other. An agent that can:

• Plan a task
• Use tools to execute it
• Reflect on the results
• Collaborate with specialized agents

...can accomplish far more than one with only a single capability.

In the next lesson, we'll look at how these capabilities come together in real-world agent architectures.

:::