I started building my AI agency and documenting it on my UA-cam earlier this year and I just turned 20 . I’ve built some Ai for gym’s and Accounting firms And this is all thanks to you Mark Thank you brother you gave me the confidence to try and start a persona brand. From watching your videos
@Mark_Kashef No the person that really deserves thanks is you 🙏🏽. I know my channel isn't as big yet. But hopefully it'll grow.. And that will all be thanks to you sir 🙏🏽
@@MubangachandaAi hey man, this is not the first time I see you in ai agent related youtube comment sections, seems like me and you are on similar paths. Keep up the grind 👍
Mark - another absolutely outstanding video again. I follow so many different people that have similar channels, but no one quite explains things in such an easy to understand and articulate manner as you do. Anybody reading my comments, I’ve hired Mark for consultation and he’s very helpful and insightful in addition to a great UA-cam content you can Hire him to consult over video or work on projects and many other things and I highly recommend him
I have been heavily consumed by this world of ai agents and automations and for the past couple of months, and I really want to start a youtube channel and AI Agency just like yours, but I feel like I am not even close to amount the amount of knowledge you have in this feild. Do you think that I should start my youtube channel and agency, and just learn as I go? Or do you think I should just keep learning about ai agents and automations for now? Love the video by the way 👍
Thanks so much! If I were starting out from scratch, I'd recommend going deep on a particular pain point that can be solved with AI that affects many business owners -- there are 'very real' problems that aren't as well broadcasted on UA-cam because they're not as flashy and clickbait-y. I'd recommend spending some time deeply learning and building unique solutions to those painpoints, then go to UA-cam to show off a comprehensive, polished, and easy-to-understand way to share to 'the people' on what problem you've managed to solve. If you continue doing this for long enough and you put real thought into your videos (along with good audio and a decent camera), it's almost impossible not to grow with quality content. Hope this helps.
Thanks so much Christian! I agree, while everything is super interesting and cutting edge, there will be serious conversations to be had in a very short amount of time.
Thanks a lot of your videos I am a bit confused about the following: -What is the difference between RAG and Agent? I see a lot of similarity, and I am also thinking that an Agent is just an enhanced RAG system. Is that right? -I also see the Agentic Framework with similarities to Microservices Architecture. Each Microservice performs a specific task, and the main Microservice collects all the responses from Microservices and responds back to the client. Kindly confirm these steps in building an Agent in simple technical words: Step 1: I would dump all the relevant data into a Vector database with embeddings. Step 2: User or API asks a question to the Agent (Microservice) in natural language through REST API. Step 3: Agent (Microservice) interprets natural language and connects with Vector DB and fetches data from the Vector for the given query Step 3.1: Agent (Microservice) calls other APIs like weather API, map API, etc., builds the entire context. It can also call other Agents. Step 4: Agent (Microservice) calls an LLM like ChatGPT and gets the response. Step 5: Agent (Microservice) provides the response back to the user or API. Thanks and Best Regards, Satish
thanks for watching! super appreciate it as well as the detailed question: in short, 1) rag focuses on searching a database for context and using that to help an llm produce answers. 2) an agent includes that capability but also decides which tools or apis to call, can orchestrate multiple steps, and can rely on other agents. it’s a more flexible, action-oriented system 3) your microservices analogy fits -- each microservice does one thing well, and the agent coordinates them, returning a single response (in most cases) 4) the steps you listed are correct for many use cases a. collect data into a vector database, b. let the user ask a question, c. have the agent retrieve info and call necessary apis, d. then generate a response with the llm and send it back. hope that helps!
@@Mark_Kashef Yes, this is super helpful! I have one last question to make my understanding solid. I need clarification on the following point: "An agent includes that capability but also decides which tools or APIs to call, can orchestrate multiple steps, and can rely on other agents." Let's consider a scenario where a user types a question: Hi Agent, please book an economical flight from Philadelphia to Colorado Springs during the long weekend with very good accommodation where I can cook. Bear in mind that I want to see a lot of snow. Based on what I’ve understood, the agent should catch the keywords and derive the related APIs or other agents. This means the agent should have NLP capabilities. Here’s how it might work: a. "Long weekend" - The agent should identify the long weekend (e.g., Columbus Day) and can rely on ChatGPT for context. b. "Bear in mind that I want to see a lot of snow" - The agent should determine the snowiest months in Colorado and make an API call to WeatherAPI using an API key. c. "Economical flight from Philadelphia to Colorado Springs" - The agent should call flight booking APIs like Expedia to find the best flight deals. d. "Good accommodation where I can cook" - The agent should call APIs like Homestay.com to find suitable accommodations. The agent should compile all this data in JSON format, which a ReactJS application can use to display the frontend. The user can then simply click and pay through their credit card. The booking information would be handled by Expedia and Homestay and sends an email to user If I’ve understood this correctly, I now grasp the complete concept of an agent. All I need to do is build a use case and develop the agent! Thanks a lot!
@@satish1012 exactly! you’ve got it. the agent approach is all about using nlp to understand the user’s request, identifying which apis or services are needed, calling them in sequence or parallel, and then compiling the results into a usable format (like json for ex)
I try to be hyper mindful that not all of my audience is code-savvy; so I'm serving that deeper video once we get more stable LLMs with high reasoning capability that: 1) are easier to implement 2) are more stable and worth any coding or configuring that you'd need to implement 3) are ideally usable with both closed and open-source models I'd rather create evergreen content that doesn't become obsolete with one 'new release'.
This is Awesome Mark ! , it clarified many questions I had in my mind regarding AI agents
I started building my AI agency and documenting it on my UA-cam earlier this year and I just turned 20 . I’ve built some Ai for gym’s and Accounting firms
And this is all thanks to you Mark
Thank you brother you gave me the confidence to try and start a persona brand. From watching your videos
Honoured to hear I could instil the confidence to take these steps, thanks so much for sharing 🦾
@Mark_Kashef No the person that really deserves thanks is you 🙏🏽.
I know my channel isn't as big yet. But hopefully it'll grow..
And that will all be thanks to you sir 🙏🏽
@@MubangachandaAi hey man, this is not the first time I see you in ai agent related youtube comment sections, seems like me and you are on similar paths. Keep up the grind 👍
@Clinksys thank you 🙌🏾
love your content. You are a leader in this A.I. Revolution I'm taking notes 📝
Appreciate you Jacob! Much much appreciated 🦾
Nice explanation, Mark, thanks again
Always a pleasure!
Mark - another absolutely outstanding video again. I follow so many different people that have similar channels, but no one quite explains things in such an easy to understand and articulate manner as you do.
Anybody reading my comments, I’ve hired Mark for consultation and he’s very helpful and insightful
in addition to a great UA-cam content you can Hire him to consult over video or work on projects and many other things and I highly recommend him
Thanks so much for the kind words and vote of confidence, Fred!
Humbling to receive yet another piece of positive feedback from you, much appreciated.
I have been heavily consumed by this world of ai agents and automations and for the past couple of months, and I really want to start a youtube channel and AI Agency just like yours, but I feel like I am not even close to amount the amount of knowledge you have in this feild. Do you think that I should start my youtube channel and agency, and just learn as I go? Or do you think I should just keep learning about ai agents and automations for now? Love the video by the way 👍
Thanks so much!
If I were starting out from scratch, I'd recommend going deep on a particular pain point that can be solved with AI that affects many business owners -- there are 'very real' problems that aren't as well broadcasted on UA-cam because they're not as flashy and clickbait-y.
I'd recommend spending some time deeply learning and building unique solutions to those painpoints, then go to UA-cam to show off a comprehensive, polished, and easy-to-understand way to share to 'the people' on what problem you've managed to solve.
If you continue doing this for long enough and you put real thought into your videos (along with good audio and a decent camera), it's almost impossible not to grow with quality content.
Hope this helps.
Thought provoking content. Awesome video Mark, Agent swarms sound cool until you realize it’s just Skynet's way of crowdsourcing the apocalypse.
Thanks so much Christian!
I agree, while everything is super interesting and cutting edge, there will be serious conversations to be had in a very short amount of time.
Thanks a lot of your videos
I am a bit confused about the following:
-What is the difference between RAG and Agent? I see a lot of similarity, and I am also thinking that an Agent is just an enhanced RAG system. Is that right?
-I also see the Agentic Framework with similarities to Microservices Architecture. Each Microservice performs a specific task, and the main Microservice collects all the responses from Microservices and responds back to the client.
Kindly confirm these steps in building an Agent in simple technical words:
Step 1: I would dump all the relevant data into a Vector database with embeddings.
Step 2: User or API asks a question to the Agent (Microservice) in natural language through REST API.
Step 3: Agent (Microservice) interprets natural language and connects with Vector DB and fetches data from the Vector for the given query
Step 3.1: Agent (Microservice) calls other APIs like weather API, map API, etc., builds the entire context. It can also call other Agents.
Step 4: Agent (Microservice) calls an LLM like ChatGPT and gets the response.
Step 5: Agent (Microservice) provides the response back to the user or API.
Thanks and Best Regards, Satish
thanks for watching! super appreciate it as well as the detailed question:
in short,
1) rag focuses on searching a database for context and using that to help an llm produce answers.
2) an agent includes that capability but also decides which tools or apis to call, can orchestrate multiple steps, and can rely on other agents.
it’s a more flexible, action-oriented system
3) your microservices analogy fits -- each microservice does one thing well, and the agent coordinates them, returning a single response (in most cases)
4) the steps you listed are correct for many use cases
a. collect data into a vector database,
b. let the user ask a question,
c. have the agent retrieve info and call necessary apis,
d. then generate a response with the llm and send it back.
hope that helps!
@@Mark_Kashef
Yes, this is super helpful! I have one last question to make my understanding solid.
I need clarification on the following point:
"An agent includes that capability but also decides which tools or APIs to call, can orchestrate multiple steps, and can rely on other agents."
Let's consider a scenario where a user types a question:
Hi Agent, please book an economical flight from Philadelphia to Colorado Springs during the long weekend with very good accommodation where I can cook. Bear in mind that I want to see a lot of snow.
Based on what I’ve understood, the agent should catch the keywords and derive the related APIs or other agents. This means the agent should have NLP capabilities.
Here’s how it might work:
a. "Long weekend" - The agent should identify the long weekend (e.g., Columbus Day) and can rely on ChatGPT for context.
b. "Bear in mind that I want to see a lot of snow" - The agent should determine the snowiest months in Colorado and make an API call to WeatherAPI using an API key.
c. "Economical flight from Philadelphia to Colorado Springs" - The agent should call flight booking APIs like Expedia to find the best flight deals.
d. "Good accommodation where I can cook" - The agent should call APIs like Homestay.com to find suitable accommodations.
The agent should compile all this data in JSON format, which a ReactJS application can use to display the frontend.
The user can then simply click and pay through their credit card. The booking information would be handled by Expedia and Homestay and sends an email to user
If I’ve understood this correctly, I now grasp the complete concept of an agent. All I need to do is build a use case and develop the agent!
Thanks a lot!
@@satish1012 exactly!
you’ve got it.
the agent approach is all about using nlp to understand the user’s request, identifying which apis or services are needed, calling them in sequence or parallel, and then compiling the results into a usable format (like json for ex)
@@Mark_Kashef Great! Now I fully understand the essence of an Agent in simple terms, and I can implement it based on my use case
Having said that this was more of a high level view of agentic workflows, could you take us one technical level deeper?
I try to be hyper mindful that not all of my audience is code-savvy; so I'm serving that deeper video once we get more stable LLMs with high reasoning capability that:
1) are easier to implement
2) are more stable and worth any coding or configuring that you'd need to implement
3) are ideally usable with both closed and open-source models
I'd rather create evergreen content that doesn't become obsolete with one 'new release'.
Where should we start?
stay tuned for another video later this week :)