hey y'all dat guy here and today I want to explain one of the hottest topics um in the industry right now you know you've probably seen salesforce's ads for for agent force um and that is AI agents um so this is referring to kind of artificial intelligence agents basically small applications that are designed to perform uh you know things from simple rule based chat Bots to really sophisticated autonomous systems capable of self-learning adaption know solving problems uh actually coding things and so what I want to do in this video both for my own edification and
also to help you guys is really just dive into what are AI agents what are the different types of AI agents that you might have out there how do they operate what is a technical definition talk about why they're useful some you know use cases where AI agents are already having a big impact and also ones maybe in the future where I can see them becoming more impactful uh and then finally end off with just a quick summary of you know some of the common tools that you can use to actually build your own I
AI agent uh at home um so without further Ado let's get into it and start talking about what is an AI agent so AI agents you know technical definition is a software entity that perceives its environment through sensors processes information and then takes action to achieve specific goals um and these agents can operate either autonomously or SE semi-autonomously using techniques from machine learning natural language processing and reinforcement learning into a really you know High use case dependent production environment now ai agents can be classified into several types there are simple reflex agents which respond to
specific inputs with predefined actions so you let's say if I were to ask question how do I find the help menu it has a predefined menu That's or action that says hey if I detect the word help menu bring up and ex display The Help menu then you also have modelbased reflex agents these are kind of the next step of that where it's hey I have an internal model of the world that I'm analyzing um and I have a goal to you know make a decision that optimizes for a specific outcome um so this is
you know say hey maybe I'm made an AI agent that's assigned to analyze a database and my goal is to reduce query speed based on the type of queries I see being performed um and so there you would be constantly analyzing you have n agents constantly analyzing the queries um and then making adjustments to the data a schema to allow those queries to run uh more efficiently faster cheaper whatever then you also have goal-based agents um and so these are where you plan actions based on desired outcomes um and this is where you know you
would be given a goal saying Hey I want you to build me a website um and then a goal-based agent would go through all the different steps needed to actually build that website you know go contact and God Daddy buying the domain Etc um you also have utility based agents this is kind of a tighter definition on modelbased agents so saying Hey I want to optimize decisions to Maxim maximize performance measures this is an agent that would be constantly analyzing queries to say hey you should be maybe changing this um to get better results from
your queries or run this query faster um then you also have learning agents um and these continuously improve their performance by learning from past experiences so if you've heard of like kind of neural networks or reinforcement learning this is where you know for a chat bot that's instead of being a simple reflex agent um it actually is going to figure out hey based on these customers what they're asking for I'm going to now find out hey was this an effective response to them if it wasn't an effective response they gave I'm going to adjust my
response for the next time that customer asks a Sim similar question and then finally of hierarchical Agents um and these are kind of you know basically Master almost manager AI agents that break down complex tasks into simple or subt and then organize them in a hierarchal structure which allows the agent to then manage different levels of abstraction solve more complex problems more efficiently by just breaking it down into comp composite units so it's not trying to solve too big a problem all at once um so really useful for kind of more like multi-step processes of
hey I want to build an application can't just build an application you know with with kind of a couple commands it's going to require many different components and a hierarchical agent will help you address and then build those different uh sub components so D we've talked about the different types of AI agents why are they useful um and really they're useful because AI agents are very adaptable across a ton of different domains um and helping enhance efficiency decision- making use interaction and also reducing the human load um but their utility can kind of be categorized
in several key areas uh number one Automation and efficiency um AI agents can automate repetitive and labor intensive tasks which massively reduces the need for human intervention and also minimizes human error um and in Industries like manufacturing and Logistics AI powered robots and scheduling agents are much more reliable um and also can work 247 to optimize production and Supply chains uh and there's also always a system of record if you're using AI AG there's never a risk of hey some Rogue human forgot to check a box um obviously Ai and you know computers can break
down but having everything be automated um by robots is a lot more uh reliable than than humans in a traditional supply chain setting then you also have data analysis and decision-making AI agents can process just insanely massive volumes of data identify patterns and make data driven recommendations faster than humans can um and financial institutions can also use II agents for fraud detection risk assessment uh algorithmic trading and really just do all those things faster than a human would be after they've developed that initial algorithm uh also enhance user experience um you know chatot virtual assistants
like Google assistant Siri Alexa all are able to provide personalized user experiences but also really what we're seeing now is because AI agents are so cheap you now have that kind of agent experience virtual assistant experience brought back to every website to every application um and so now you have ai agents in e-commerce that can do things like recommend products based on user behavior um so really just a dissemination of those kind of AI assistants throughout the entire spectrum of daily life um and then you also have adaptability and learning so unlike traditional software AI
agency can learn from new data um and adapt to changes in their environment uh so things like self-driving cars they can use reinforcement learning to learn how to improve navigation and safety as they're driving um and also things like scalability AI agents can handle multiple interactions simultaneously and you can always add additional AI agents without buy paying for a new employee so makes it really ideal for customer service or cyber security applications where you might have you know a very elastic load maybe you don't have any requests for a while and you have thousands of
requests at once AI agents can help you scale to meet that load appropriately rather than just having a ton of people or a bunch of computers or a bunch of things standing on by just in case for those Peak loads um and so lot of different benefits that all boil down to really better efficiency for your existing resources reducing manual labor for humans um faster iteration faster decision- making um and also able to attack new sources of revenue um and just unlock your Workforce to be more productive so now finally I want to end with
just a discussion of you know where what are the different options you have out there for building and implementing AI agents in the real world um because there's a lot of different Frameworks architectures development approach and it's also a very early industry so there's a lot of new development a lot of Rapid iteration happening right now um and so first I kind of just want to divide into different categories um and then talk about hey what are some options for each of those different categories um so starting off with you know very basic rule-based system
that's where you know chat Bots really early customer support automated response agents that's where you don't even really I wouldn't even classify them as AI agents because they're really just you know hey rule based um and that's where something like rasa something like dialogue flow um that you would use for you know just basically saying hey if you get this message then do that um where you start to see you know actual machine learning based Agents come into play is with things like Predictive Analytics systems for stock market forecast Healthcare Diagnostics and that's where you're
going to be using actual agent developer Frameworks um like or actual you know machine learning Frameworks like tensorflow like pytorch like scikit learn um but then the next step to that is reinforcement learning agents um so this is where I think you kind of really get into actual true AI agents so self-driving car systems um you know things that are learning from Real World simulated environments um this is where you have tools like open AI gym Ray R lib um really any kind of systems that allow you to build and then maintain a model run
for a long period of time also things like ampy High touch offer fit and so here's actually an example of one of those so GPT 4 or Auto GPT this is a tool that allows you to actually use open AI GPT models to create autonomous a agent set XU tasks independently and you can actually see the kind of memory flow here where you're constantly adding new Tas tasks task creation agent is also just learning doing problem solving uh within the context of that so things like research assistance data collection problem solving tasks these are all
really good use cases for these kind of autonomous AI agents decision-making systems that can make those complex decisions and in Dynamic environments build on top of someone else's rails don't try to build everything from scratch don't try to develop your own L llm build what's already out on top of what's already out there um and then just cut it down to whatever your domain is of expertise that you actually want to focus on and kind of in keep keeping with that I also want to talk about you know like just to round off this video
go to your cloud provider um every cloud platform has a lot of AI tools that they're trying to push right now for people to use um so Google Cloud AI has agents for different AI models generative AI you can use uh you can have it you know orchestrated from an agent brain so that hierarchical agent and have multiple sub AI agents that each work on their own individual tasks has a lot of apis functions databases built into it Azure also has a very similar stack kind of built around fabric AWS has a similar stack um
so highly recommend getting into one of your whatever cloud provider you're using or if you can get a really good deal from one using their existing AI Services especially if you're working for a business it'll help you get around a lot of compliance issues if you start using it from a cloud Friday you already have a lot of contracts with um so yeah if you want me to continue talking about this topic and actually show you you know how to build your own AI agents do this in production let me know I just kind of
want to make an intro video for people that are just figuring out what an AI agent is so you can have a better understanding of that um but hope you learned from me today hope you enjoyed this video have a great rest of your day dat a guy out
