Get startedGet started for free

What's in the (Tool) Box?

1. What's in the (Tool) Box?

Welcome back! In the last lesson, we unpacked how prompting frameworks like the ReAct framework enable models to break down problems into small steps.

2. What's in the (Tool) Box?

But thinking alone doesn't make an AI system agentic. In this lesson, we’ll uncover how agents use tools to turn thoughts into actions.

3. Tool Use: An Essential Component

Tool use is essential for making AI systems truly agentic.

4. Tool Use: An Essential Component

Without tools, they can only generate responses and lack the crucial “Agency” that enables them to interact with the external world.

5. The Different Type of Tools

There are many types of tools available within agentic systems. They generally fall into three categories:

6. The Different Type of Tools

The first are extensions.

7. Extensions

Extensions connect agents to the outside world through Application Programming Interfaces, or APIs for short.

8. Extensions

Or emerging protocols like the Model Context Protocol, or MCP for short. MCP is an open standard that lets AI assistants connect to external data sources and tools through a unified interface. That said, we'll focus on APIs in this course, while MCP will be covered in separate courses.

9. Application Programming Interface (API)

So what actually is an API? As mentioned, API stands for Application Programming Interface, and they act as messengers between software applications, taking a request to a system and receiving a response containing data or services.

10. Application Programming Interface (API)

APIs are like waiters in a restaurant; they take our order, or request, communicate it to the kitchen, the system providing the service, and finally, deliver the food, or response from the system, back to our table.

11. Extending Agent Capabilities with APIs

For example, let’s say you have an financial advisor AI agent that can provide you with up-to-date information about stock data.

12. Extending Agent Capabilities with APIs

If you ask it, "What’s the stock price of Nvidia today?”

13. Extending Agent Capabilities with APIs

It can connect to the Yahoo Finance API — query the data for you, and use this real-world data in its response.

14. The Different Type of Tools

The second type of tools are functions.

15. Functions: Custom Abilities

Functions let agents execute a specific set of code. These are like custom tools you give your agent designed for your specific needs.

16. Improving Capabilities with Functions

Imagine you're extending your financial advisor agent's capabilities to also provide analysis on stock data. You might create a function that calculates moving averages on the provided stock data, which provides a signal over how a stock is performing.

17. Improving Capabilities with Functions

In this context, a user can ask, “Analyze Nvidia's stock performance over the last 30 days” The model retrieves the stock data using the Yahoo finance API, executes the moving averages function, and returns the results.

18. The Different Type of Tools

The third type of tools is data stores

19. Data Stores: Retrieving Information

Data stores let agents retrieve information from databases and documents.

20. Data Stores: Retrieving Information

This can be structured data, like databases, excel files, product directories, and more.

21. Data Stores: Retrieving Information

or unstructured data like video, audio, PDFs, emails and text files

22. Improving Capabilities with Data Stores

Let's go back to our financial advisor agent. Say we want to extend its capabilities, and provide it access to industry analysis written by expert human analysts.

23. Improving Capabilities with Data Stores

In this context, a user can ask, “Analyze Nvidia's stock performance over the last 30 days” The model retrieves the stock data using Yahoo finance, executes the moving averages function we defined, reads an analyst report on the state of the semi-conductor industry.

24. Improving Capabilities with Data Stores

And provides a detailed response.

25. The Tool Box

If you’re building agents, both low-code/no-code agent builders and agentic frameworks come with a set of pre-loaded tools you can rely on. If you want to use tools customized to your specific use case, you’ll probably have to create them yourself.

26. Let's Practice!

But we'll cover that in other courses. For now, let's test your skills on the agentic tool box.

Create Your Free Account

or

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.