This video is a must-watch for anyone just starting out! I don’t know why, but people tend to overcomplicate things with multi-agent frameworks when you can solve problems in much simpler ways. Also, it would be awesome if you made a video about evals, there’s not much content about it online
I think there are only 4 important agent-flow: 1. Routing 2. Prompt Chaining 3. Loops 4. Paralellization The others are combinations of these 4. And each flow uses augmented-LLM-nodes.
Old guy, degreed engineer here (Computer) diving deep with as much time as I can spare on the AI train now after watching and slowly incorporating the surface level tools into my life and business for the past 2 years. This is exactly the type of content I was looking for. I will be joining the community shortly.
I am very new to this and have zero knowledge of coding, but I really liked the way you explained everything and I truly appreciate that. Can you create a full video tutorial on this topic, explaining it in detail?
Dave, congrats on this video. There is too much fluff today about "agentic aI" - and unfortunately, the youtube crowd that has no real idea about it, have found this 'hammer' and looking at everything as a 'nail'. Great video!
this was the first video I've watched and I am incredibly grateful for your efforts here. I am attempting to change career paths from over a decade of IT Channel B2B to AI development. It is still quite daunting, but so glad I've found your channel and will be looking to join Discord + all the other stuff. Great work, thanks once again!
Excellent video. I wrote my thesis on how AI is applied in schools and Khanmigo, magic school, and all the other teacher tool builders necessarily have the workflow AI because these tools can't be agents in the classroom. Text or information on a screen would never calm a child or teach a class without encountering frustrated students.
I just came across your channel, and my first impression is that it’s so well-explained and incredibly welcoming! I’m hitting like and follow right away. Keep up the amazing work! :)
It's amazing how many showcase of AI agents exist in LinkedIn, UA-cam, X etc can be implemented by a single LLM call or a workflow with at most three LLM calls with tool calls. As in the traditional ML space, a single LLM call with a well-engineered prompt should be the baseline solution.
Timestamps 00:03 - Building reliable AI agents is challenging despite widespread enthusiasm. 01:56 - AI agents are often misunderstood amid hype and automation discussions. 04:09 - Understand the distinction between workflows and AI agents for effective system building. 06:28 - Key building blocks for effective AI applications around language models. 08:50 - Combine memory concepts to enhance LLM API calls for better results. 11:17 - Routing and parallelization optimize AI agent workflows. 13:24 - Introducing agent patterns for more efficient AI workflows. 15:34 - LLMs utilize feedback loops for adaptive decision-making in agent systems. 17:41 - Building effective AI agents requires understanding core patterns and starting simply. 19:27 - Build core AI components with a focus on simplicity and clarity. 21:19 - Scaling AI applications requires careful planning and rigorous testing. 23:06 - Implement robust checks to safeguard your brand in AI projects.
I've been mulling over a lot of these ideas over the last 6 months or so, and you just laid them all out perfectly and explained everything in between for me. This is definitely a video I'm going to come back to when I'm not as high XD
For reference, I've been a devops engineer for a decade or so now. I'm familiar with automation pipelines, but I've only recently been getting into LLM's and python at the same time. It's been unlocking so many potential new workflows that I can't keep up with them all.
I know plenty people who need listen to this (but prefer the warm and comfort of the illusionary hype). Thank you so much for share so meritorious view.
Hey Dave, I’m really impressed with what you’ve built with your AI agency-it’s solid, innovative, and clearly impactful. I believe your model has incredible potential to be replicated in Mexico, where the AI market is still in its early stages. I’d love the chance to connect and dive deeper into how your work model operates. There’s nothing quite like this in Mexico right now, and I’m convinced that with the right approach, it could become a massive success. Your expertise could pave the way for something groundbreaking here. Let me know if you’d be open to discussing this further!
Thank you, Thank you and Thank you! Finally someone down to Earth! It's a must watch video. I just came across your channel and I'm so grateful for it! Gracias for sharing great thoughtful information!
Phew... As a noobie to the space, and being a non-developer, with zero coding knowledge, I can say that I'm really glad you continuously give basic, hard examples of these descriptions. Like... instinctively, I know there's a difference between agents and workflows, but... having you detail those is useful. There are just a few spots I don't understand, but overall this video has been incredibly helpful. I even saved it to a Notion page. 😝😝 Thank you, Dave!
Totally agree and thank you for sharing this information. As a beginner in AI Agent space, i noticed that most of these frameworks or low code/no code platforms are not production ready. No one talks about redacting PII (sensitive) data or how to deploy these Agents in production environments. Mostly all UA-cam video showcase Jupyter notebooks. 😊
Thanks for keeping it simple, this was easier than other videos I’ve seen. I read about this AI agent Workbeaver AI that uses a show-and-tell method where it's taught through task demonstration then the automation would run on local PC. Haven’t tried it yet but I joined the waitlist. If it works, no coding or templates would be needed. Thoughts on AI agents like these? Is it even feasible?
Oh lord Thank You!! I've been looking for you Red, I just didn't know it yet. Finally. Someone on here talking about real shoot instead of making flight plans, or getting a stupid hotel, or getting it to 'create' God-awful cartoony videos of a tiger eating doritos or whatever. I swear if i had to hear another one call that nonsense "Amazing" I was gonna punch the screen. Where are they all flying to anyway, the great Doritazan jungle? You rock dude. Please make about Fifty-eleven of these bc it's coming...Quick. We need people like you.
Hey Dave, your SKOOL link is broken. Can you update with a working one? Interested in joining! Really appreciate how you broke down the concepts systematically so we can better understand how to think and design in a simplified and robust way.
Yeah bro, that's why we use agents for the Prompt Chaining and Routing. It's far better to have a personalized for the task AI than to just call the same general one.
I appreciate your down-to-earth approach to the whole thing and i start watching your videos very often but as a beginner i always lose you somewhere at the half of the videos. I'd like to see more real-life examples when explaining the concepts. Thanks!
Great video! I'm starting to do more with automation at my current position and I just watched an "AI Agent" course with Digital Workforce. They keep referring back to sema4 ai as being on the cutting edge of AI Agents. I don't think you mentioned anything about them...so I was curious.
I aspire to leave my full-time job eventually and embrace a freelancing lifestyle that allows me to travel the world. Although I have a foundational understanding of AI and its frameworks, my coding experience is limited. What would be the best way for me to get started? Should I begin by taking a coding course, such as Python?
Great video, thanks for sharing! Just watched Salesforce CEO interview, he mentioned implementing 25k AI agents, targering Agent first oganization. Can someone explain what does it mean to deploy "25k AI agents" ? I'm very sure it is not 25k physic robots, and also not 25k different types of agents. What kind of deployment it is? 25k droplets working separately ? If that is the case, then the deployment seems not so optimal.
While I agree with you there, that's exactly the context a lot of people need that no one talks about on UA-cam. At the end of the video, I do give 5 practical tips, though, based on our experience building AI systems.
Good explanation Thanks. As of the time watching this video. OpenAI has introduced Operator, an AI agent that can complete tasks on its own using a web browser.
Fantastic content! n8n provides a nice way to understand the diference between just making the api call and actually using the AI Agent. Although, as you said: many problems don't really need a agentic approach! Nice!
Won't either of these be too much time-consuming in context of UX ? Like whatever the techniques - every additional LLM call will greatly influence response time and thus - user experience.. Especially as you mentioned the extra last LLM call guardrail - things can quickly escalate to multiple LLM calls - causing a long response waiting time. How would one approach RAG implementation or some other form of improvements without trading off computation time?
UA-cam has a lot of AI agentic demos that aren't useful in real life. Personally, I use AI workflows - AI agents are not reliable in business scenarios.
It seems i was building an agent in 2024 without realizing it. Except, i limited the iterations to 3. The task was simple and yet it would fail. We tried everything. Gave it examples, gave it output from the tools it would call. The system wasn't 100 and we just kept it like that
Whats really odd are the Y Combinator backed startups that offer AIs to completely replace humans. For example, a voice AI agent that makes outbound sales calls without any human oversight. At the end of the day, it's just an LLM, and it will hallucinate.
Not sure why I am struggling with the augmented LLM nodes as much as I am. Does anyone have any guides handy for getting the basic building blocks for retrieval and memory down? 🙏
What frameworks will useful agents run on? In other words, is there a supportive platform with semantic metadata that will enable them to become useful?
basically what you are discussing is efficiency and lack of scaling ability, using 3rd party "Agents" work, but are not efficient, and you cannot go to scale relying on some third party.
I'm working on a Chrome extension to help my disabled friend control their browser by speaking to it, just like having a conversation. This would make browsing much easier for him. I'm not sure how to build it, though-does anyone have any advice? Your help would mean a lot!
I'm just getting started with ai can anyone suggest what topics should i know other than traditional machine learning algorithms supervised and unsupervised learning???
This video is a must-watch for anyone just starting out! I don’t know why, but people tend to overcomplicate things with multi-agent frameworks when you can solve problems in much simpler ways. Also, it would be awesome if you made a video about evals, there’s not much content about it online
Evals is on my video list! There's indeed hardly anything on this topic
I think there are only 4 important agent-flow:
1. Routing
2. Prompt Chaining
3. Loops
4. Paralellization
The others are combinations of these 4.
And each flow uses augmented-LLM-nodes.
Great breakdown! Thanks for this comment 🔥
@MisCopilotos thanks bro
Nowadays, so many "agents" lack agency... Very refreshing video, I can smell experience on real-world projects. Thank you
Finally, found someone who isn’t jumping on the fake AI hype train just for views and shares! Thanks Sir.
No hype here. Appreciate it! 🙏🏻
@@daveebbelaar ditto - liked and subscribed
Old guy, degreed engineer here (Computer) diving deep with as much time as I can spare on the AI train now after watching and slowly incorporating the surface level tools into my life and business for the past 2 years.
This is exactly the type of content I was looking for. I will be joining the community shortly.
Love his mission “Make more money, work on fun projects and create freedom”. Instruction is terrific. Well structured and explained.
Am gonna research on this. Finally someone that's not telling us we can make $2000 a week with A.I. Just the channel i was looking for. Thanks
Just keeping it real ;)
I am very new to this and have zero knowledge of coding, but I really liked the way you explained everything and I truly appreciate that. Can you create a full video tutorial on this topic, explaining it in detail?
Appreciate that! Yes, I have a follow-up video planned where I explain how to implement these patterns in practice.
Dave, congrats on this video. There is too much fluff today about "agentic aI" - and unfortunately, the youtube crowd that has no real idea about it, have found this 'hammer' and looking at everything as a 'nail'. Great video!
this was the first video I've watched and I am incredibly grateful for your efforts here. I am attempting to change career paths from over a decade of IT Channel B2B to AI development. It is still quite daunting, but so glad I've found your channel and will be looking to join Discord + all the other stuff. Great work, thanks once again!
Excellent video. I wrote my thesis on how AI is applied in schools and Khanmigo, magic school, and all the other teacher tool builders necessarily have the workflow AI because these tools can't be agents in the classroom. Text or information on a screen would never calm a child or teach a class without encountering frustrated students.
I just came across your channel, and my first impression is that it’s so well-explained and incredibly welcoming! I’m hitting like and follow right away. Keep up the amazing work! :)
Appreciate that! Thanks!
same
Building an "AI Agent" rn and this video is good send to understand the fundamentals. Loved it. Thanks Dave
It's amazing how many showcase of AI agents exist in LinkedIn, UA-cam, X etc can be implemented by a single LLM call or a workflow with at most three LLM calls with tool calls. As in the traditional ML space, a single LLM call with a well-engineered prompt should be the baseline solution.
Timestamps
00:03 - Building reliable AI agents is challenging despite widespread enthusiasm.
01:56 - AI agents are often misunderstood amid hype and automation discussions.
04:09 - Understand the distinction between workflows and AI agents for effective system building.
06:28 - Key building blocks for effective AI applications around language models.
08:50 - Combine memory concepts to enhance LLM API calls for better results.
11:17 - Routing and parallelization optimize AI agent workflows.
13:24 - Introducing agent patterns for more efficient AI workflows.
15:34 - LLMs utilize feedback loops for adaptive decision-making in agent systems.
17:41 - Building effective AI agents requires understanding core patterns and starting simply.
19:27 - Build core AI components with a focus on simplicity and clarity.
21:19 - Scaling AI applications requires careful planning and rigorous testing.
23:06 - Implement robust checks to safeguard your brand in AI projects.
Thanks ❤
Very nice. Thanks a lot. How did you create these Timestamps? Did you do it with some AI tools?
Thank YOU!!!
I've been mulling over a lot of these ideas over the last 6 months or so, and you just laid them all out perfectly and explained everything in between for me. This is definitely a video I'm going to come back to when I'm not as high XD
For reference, I've been a devops engineer for a decade or so now. I'm familiar with automation pipelines, but I've only recently been getting into LLM's and python at the same time. It's been unlocking so many potential new workflows that I can't keep up with them all.
Really great video. So different than so many that just hype the agent buzzword. Great example of someone who builds practical systems for customers
I know plenty people who need listen to this (but prefer the warm and comfort of the illusionary hype). Thank you so much for share so meritorious view.
Super helpful, learnt so much! Took me an hour to watch that 20mins so I could take notes and understand it all. Thank you very much
Hey Dave, I’m really impressed with what you’ve built with your AI agency-it’s solid, innovative, and clearly impactful. I believe your model has incredible potential to be replicated in Mexico, where the AI market is still in its early stages. I’d love the chance to connect and dive deeper into how your work model operates.
There’s nothing quite like this in Mexico right now, and I’m convinced that with the right approach, it could become a massive success. Your expertise could pave the way for something groundbreaking here. Let me know if you’d be open to discussing this further!
Wow, great video! Awesome explanation. For the first time found someone who really explains the under the hood logics of AI Agents! Thank You...
A video on problems you encountered scaling RAG engines and how you solved them would be useful. Great content Dave.
Thanks!
Thanks, Nick! Appreciate it!
Thanks Dave, This Video teach me how I build more robust solitions. keep it up ! 😀
Great video, Dave! Learned a lot. Thanks for keeping it clear and practical.
man you have helped me so much. I am so thankful. you are the real deal
Thank you, Thank you and Thank you! Finally someone down to Earth! It's a must watch video. I just came across your channel and I'm so grateful for it! Gracias for sharing great thoughtful information!
I would really appreciate if you would give a real world example in your future videos. Subscribed! Thanks much.
Yea I think that would be a great follow-up video, implementing these patterns
Phew...
As a noobie to the space, and being a non-developer, with zero coding knowledge, I can say that I'm really glad you continuously give basic, hard examples of these descriptions.
Like... instinctively, I know there's a difference between agents and workflows, but... having you detail those is useful.
There are just a few spots I don't understand, but overall this video has been incredibly helpful. I even saved it to a Notion page.
😝😝
Thank you, Dave!
Excellent post, very cohesive and straight to the point. Congratulations!! I'm one more subscriber now!
Totally agree and thank you for sharing this information. As a beginner in AI Agent space, i noticed that most of these frameworks or low code/no code platforms are not production ready. No one talks about redacting PII (sensitive) data or how to deploy these Agents in production environments. Mostly all UA-cam video showcase Jupyter notebooks. 😊
Exactly. It's my goal to fill that gap on UA-cam
n8n on your local machine connect to your personal Google drive
I subscribed. I’m thinking of making this my second career. Your explanation is awesome.
Thanks for keeping it simple, this was easier than other videos I’ve seen. I read about this AI agent Workbeaver AI that uses a show-and-tell method where it's taught through task demonstration then the automation would run on local PC. Haven’t tried it yet but I joined the waitlist. If it works, no coding or templates would be needed. Thoughts on AI agents like these? Is it even feasible?
By far the best explanation of Ai I have come across
Oh lord Thank You!! I've been looking for you Red, I just didn't know it yet. Finally. Someone on here talking about real shoot instead of making flight plans, or getting a stupid hotel, or getting it to 'create' God-awful cartoony videos of a tiger eating doritos or whatever. I swear if i had to hear another one call that nonsense "Amazing" I was gonna punch the screen. Where are they all flying to anyway, the great Doritazan jungle? You rock dude. Please make about Fifty-eleven of these bc it's coming...Quick. We need people like you.
Great video! Evals from the start seems like the most important thing. Any resources you recommend?
Hey Dave, your SKOOL link is broken. Can you update with a working one? Interested in joining! Really appreciate how you broke down the concepts systematically so we can better understand how to think and design in a simplified and robust way.
It's working on my side. This is the correct link: www.skool.com/data-alchemy
@@daveebbelaar oof! for some reason it wasn't loading in Safari but it works now in Chrome. Thanks!
Yeah bro, that's why we use agents for the Prompt Chaining and Routing. It's far better to have a personalized for the task AI than to just call the same general one.
This is great. Real-world info rather than click hype. I'd love to see a vid on evals.
This video deserves all the love, amazing!
I appreciate your down-to-earth approach to the whole thing and i start watching your videos very often but as a beginner i always lose you somewhere at the half of the videos. I'd like to see more real-life examples when explaining the concepts. Thanks!
So much hype about AI ..I just love your honesty and hard talk
Finally something informative and not overhyped!
Great video! I'm starting to do more with automation at my current position and I just watched an "AI Agent" course with Digital Workforce. They keep referring back to sema4 ai as being on the cutting edge of AI Agents. I don't think you mentioned anything about them...so I was curious.
Great stuff! I love the diagrams in the explanations. Really helped the message stick.
worth it to watch.
And the workflows you said, I just learnt from langgraph academy.
Fantastic video, thank you for clarifying it all.
Your style is exceptional!
I aspire to leave my full-time job eventually and embrace a freelancing lifestyle that allows me to travel the world. Although I have a foundational understanding of AI and its frameworks, my coding experience is limited. What would be the best way for me to get started? Should I begin by taking a coding course, such as Python?
The only video i found that is really valuable and without the hype stuff
Great video, thanks for sharing!
Just watched Salesforce CEO interview, he mentioned implementing 25k AI agents, targering Agent first oganization. Can someone explain what does it mean to deploy "25k AI agents" ? I'm very sure it is not 25k physic robots, and also not 25k different types of agents. What kind of deployment it is? 25k droplets working separately ? If that is the case, then the deployment seems not so optimal.
Amazing, I like the Routing approach; makes very versatile AI
Great video..very concise and to the point!
Great and clean explanation
This is fantastic. Subscribed. Thanks man!
My understanding is that this is just a review of the Anthropic post and general software engineering rules. Is there any added value there?
While I agree with you there, that's exactly the context a lot of people need that no one talks about on UA-cam.
At the end of the video, I do give 5 practical tips, though, based on our experience building AI systems.
Nice explanation. Thanks
Great eye opener on AI agents
Good explanation Thanks. As of the time watching this video. OpenAI has introduced Operator, an AI agent that can complete tasks on its own using a web browser.
Good job explaining in a simple way 👍
Keep it up! You’re doing an amazing job!
This was ERY helpful, thanks Dave!
Quality video thanks for putting it out!
Excellent information, thank you.
Good post Dave!
Subbed one minute in, excellent content!
Fantastic content! n8n provides a nice way to understand the diference between just making the api call and actually using the AI Agent.
Although, as you said: many problems don't really need a agentic approach! Nice!
Amazing Dave!
Really great video Dave. Much appreciated. 👍🏾
Glad you enjoyed it
Thanks for this video. Good Work..
Simple explanation..AI Agent = a tool that automates some of your tasks..lets not over complicate.
I am just loving this.. ❤
Awesome content without the hype!
Won't either of these be too much time-consuming in context of UX ?
Like whatever the techniques - every additional LLM call will greatly influence response time and thus - user experience..
Especially as you mentioned the extra last LLM call guardrail - things can quickly escalate to multiple LLM calls - causing a long response waiting time.
How would one approach RAG implementation or some other form of improvements without trading off computation time?
Best video I’ve watched so far !
Thanx Dave. It was very useful
thx, can you do a video regarding the gdpr and ai?
Clear explanation bros. Subscribed already.
Excellent content.
UA-cam has a lot of AI agentic demos that aren't useful in real life. Personally, I use AI workflows - AI agents are not reliable in business scenarios.
100%
Interesting.
It seems i was building an agent in 2024 without realizing it. Except, i limited the iterations to 3. The task was simple and yet it would fail. We tried everything. Gave it examples, gave it output from the tools it would call. The system wasn't 100 and we just kept it like that
Whats really odd are the Y Combinator backed startups that offer AIs to completely replace humans. For example, a voice AI agent that makes outbound sales calls without any human oversight.
At the end of the day, it's just an LLM, and it will hallucinate.
Not sure why I am struggling with the augmented LLM nodes as much as I am. Does anyone have any guides handy for getting the basic building blocks for retrieval and memory down?
🙏
Excelent content sir.
Currently the price of integrating LLM into workflows is too expensive to be realistic. Am I wrong?
Why? You can integrate it easily, but what for? How does your business model profit from having an automated bullshitter?
How are you continuosly looking at the camera and talking? Is that AI as well?
Haha no that's just a lot of takes and cutting in between ;)
First 1 min was straight from heart! So true brother💯
What frameworks will useful agents run on? In other words, is there a supportive platform with semantic metadata that will enable them to become useful?
No frameworks but databricks Is a very interesting platform
great presentation
Thank you :)
basically what you are discussing is efficiency and lack of scaling ability, using 3rd party "Agents" work, but are not efficient, and you cannot go to scale relying on some third party.
I'm working on a Chrome extension to help my disabled friend control their browser by speaking to it, just like having a conversation. This would make browsing much easier for him. I'm not sure how to build it, though-does anyone have any advice? Your help would mean a lot!
Use voice recognition, simplify commands to actions like move right move up, programm scrolling speed etc
Video starts at 6:39
Sir, I just want to know how much your program costs. I don't want to waste your time booking a call. Please let me know how much your program costs.
can you please mkae a video on the ai solutions selling for a lot of money
Can you please post the download able file of the document that you are reading in the video .
Reliably incredible content. Thanks, Dave.
Crazy how I thought I was learning about agents and all other influencers have been saying workflows are agentic
Who has the best outbound AI agents?
I'm just getting started with ai can anyone suggest what topics should i know other than traditional machine learning algorithms supervised and unsupervised learning???