The Complete Guide To Building AI Agents (10 Lesson Recap)

0:05Welcome back to the course This is the final video on how to build an AI agent from scratch We first started with ChatGPT

0:18We talked about how it's an app around a language model but it's not a language model itself because it has access to tools

0:31Tools allow language models to do things You can execute code call APIs execute CLI commands get information in some way inject it into the context

0:46to make better decisions later But ChatGPT isn't really an AI agent even though it has tools because it has no orchestration loop The

0:59orchestration loop allows the LLM to think and act in its environment It's literally called REACT or Reasoning Action Framework

1:11After you have your basic REACT loop you need to have ways to figure out which tools is the LLM allowed to execute by itself and which tools require

1:24explicit user permission Some tools are cheap and easy and quick and you don't need any approval and some tools are expensive or dangerous and you want a human in the loop just in case

1:38so you don't make any mistakes We talked about memory There's more advanced memory like retrieval augmented generation There's other advanced memory architectures like storing

1:51memory data in a database and then querying for certain fields at runtime and just injecting that into the context There are a number of different ways to do memory It just depends on your actual use case

2:05We talked about evals How do you know your agent is doing what you want it to do You can use a very simple objective matter eval or you can use something

2:17like a weighted score for something that's more subjective and you don't even need to use LLMs for evals at all You might have objective criteria like how much time it took how much was the cost how many

2:30iterations it did or any other things like backtest performance can be used in your evaluation pipeline But I cautioned you that you have to make sure that your evaluation pipeline is

2:43grading how you want it to grade because if you optimize the wrong thing your agent might learn to do something that you don't want it to do If you mention all of these things you

2:56understand all these things then you will pass You will excel in AI native system design interviews Thank you guys for watching

3:09I hope this was helpful but don't just watch this video and that's it Now you are primed with information but you're not going to retain it unless you do something yourself So go out there

3:23open your terminal open link chain open cloud agents or whatever and build your own AI agent to do whatever you want To check your emails to respond to Slack to do anything I

3:37don't care what it is build your own AI agent so you commit the knowledge to your memory and don't just build AI agents Use them Download Cursor download ClockCode or use NexusTrace AI

3:51agent for stock trading All of these things will help you cement into your memory how agents work so you know that it's not a black box It's something that you can build right now