From prompt to action: Exploring tool use
1. From prompt to action: Exploring tool use
Our welcome agent is polite and is great with language, but it might not have all the facts to answer customer questions on its own.2. Smart, but not all-knowing
In real-world applications, we often want AI agents to perform specific tasks3. Smart, but not all-knowing
or fetch information they don’t inherently “know.” This is where tools come into play.4. Smart, but not all-knowing
By equipping our agent with tools, we extend its capabilities beyond just the language model’s built-in knowledge.5. Connecting to external resources
Some ways we could enhance our agents are, for instance, giving them access to external resources, like a database of company FAQs,6. Connecting to external resources
a web search engine,7. Connecting to external resources
or even a calculator. Adding a tool to an agent is a bit like giving a human a new app or a reference book: it empowers them to handle requests they couldn’t before.8. Picking the right tools
There are many kinds of tools an agent could use, depending on the use case: A customer support agent might use a knowledge base lookup tool to fetch answers to specific product questions, instead of relying on the model’s memory, which might be outdated or inaccurate.9. Picking the right tools
A financial assistant agent might use a calculator or spreadsheet tool to crunch numbers when asked about budgets or projections.10. Picking the right tools
A news agent could have a web search tool to get the latest news or information from the internet.11. Picking the right tools
A coding assistant agent might use a code execution tool to run snippets of code or a documentation lookup tool to find API details.12. Reliable answers with tools
Combining tools with agents is especially powerful for tasks that require factual correctness or multi-step processes. Our own project is a perfect example: a support agent that can not only greet users, but also answer their questions using real company data.13. Reliable answers with tools
By giving the agent a tool (like a function to look up FAQs), we ensure it provides reliable answers rather than guessing.14. Reliable answers with tools
The result will be an agent that can answer actual customer questions by consulting an FAQ resource, rather than just replying with general notions that the LLM learned.15. Let's practice!
Let’s head over to the next video to find out how to do this in practice!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.