Today you need AI agents to translate from user prompts to structured API calls of traditional websites. But that doesn't have to be the future. Each site would evolve their API calls to handle natural languages too. So the agents of the future would be one that converts user prompts to multiple sub prompts to be injected into different agent functions.
True, the frameworks have to consider the security and governance aspects. We have done that in the past but AI Agents is tricky. So will evolve as we roll these out.
I think we still have a ways to go before LLMs are reliable enough to just hand off entire workflows to them. I think they are great at specific NLP tasks, but I haven't seen them handle every situation in a super reliable way - which means I couldn't build an automated solution that way and hand it off to a customer.
I just started learning it a few weeks ago with Examspm being run by Helena Liu. They're extremely good but the required tools and apps are too expensive for a beginner!😊😊
A very good effort! However, LLMs do have memory for specific details about past interactions with a user. Though there are techniques (e.g RAG) to mimic memory, by including past conversations within the context window along with the prompt.
Great video....I was just wondering that LLMs like Gemini, OpenAI are really powerful but how can we use their power??? I mean their responses are so random even at 0.4 temperature. What if i want to use the output of the LLM to drive my workflow? For this the LLM must always output "exact" answer & not just "correct answer". For Eg : To classify the issue "My Laptop is showing blue screen", LLM will classify it as "OS" / "Windows"/ "Blue screen error". All are correct but to drive my workflow who is only looking for "OS" value condition, other suggestions will break my workflow in spite of being correct. So we need a kind of control on output for automating using LLMs. That's where AI Agents will rock. Need your advice on how to learn creating AI Agents using LLMs. Is there any formatting language, framework etc. to build agents in the form of prompts? I have found Guidance as one but its bit difficult to learn. :)
Agreed, there are many videos. Look up videos on how to build AI agents - the current best are Crew AI or Langchain (needs coding) or Flowise (no code). I will also plan to make one. Regarding Halucination issues, I am soon coming out with a video on RAG - pls watch that. There are multiple techniques and LLM providers are working on improving the reasning abilities as well. thanks for the detailed comment.
Sir u mentioned point 3. Memory(long and short), I See in comments someone commented about techniques (e.g RAG). I am facing issue on this concept only that I am using an underline model like GPT, GEMINI etc and want to train them specifically on my own data and devise a algo to act in a certain direction but I am not sure how to save the state where underline model is aware of my data and customisation to give a directional response when ever someone prompts with my AI agent. Can you guide me in a direction to learn this or where is my understanding flawed.
Video was informative. Just few feedback, Video: Jump cuts, zooms do loose eye contact and focus, transitions can be improved. Try integrating visual elements more. For audio the shhh sound is quite a noise, will suggest to use Audition. Overall video was excellent!
AI agents are versatile digital assistants capable of diverse tasks and interactions, whereas SmythOS is designed to optimize specific user experiences within its dedicated technological environment.
I have been articulating these concepts with many peers at my work 9 months ago on exactly what you are mentioning here. I believe this is where AI agents will take AI to a whole new level and start changing industries. Such an amazing time to be part of. AI agent software engineers is where it will be at in 6-9 mths!
10s of TRILLIONS of dollars? Annual global spend on ENTIRE IT this year is expected to be just over $5 trillion. So you expect just AI Agents (latest flavor of rapid introduction of AI buwords) is going to exceed 2 years of global spend in the future ? Just some RANDOM VERY LARGE ROUND NUMBER? Or is there some computation behind this claim ?
Bro, just a constructive feedback. You talk a lot unrelevant stuff. The 11 min video could have been a 5 min crisp and to the point video. Unfortunately, I have to close the video after 5 min only since you didn't even start talking about the agents after 5 min
Thanks for the amazing response🙏🏽
Appreciate your thoughts on AI Agents. What do you think?
Awesome video!
How we can learn ai agents?
Look up videos on how to build agents with Crew AI or Langchain (needs coding) or Flowise (no code). I will also make one soon.
Today you need AI agents to translate from user prompts to structured API calls of traditional websites. But that doesn't have to be the future. Each site would evolve their API calls to handle natural languages too. So the agents of the future would be one that converts user prompts to multiple sub prompts to be injected into different agent functions.
That's an interesting way to do it. Appreciate the add!
Extremely Inspiring. Thank you. Subbed.
Awesome. Thanks for the sub!
We are building a massive agentic system. 1 agent ≠ 1 role/job; 1+n agents ≈ 1 role/job.
Awesome!
This video is a real gem.
Thanks 🙏🏽
Very good explanation. It was easy to understand. Thank you!
Truly appreciate the feedback. Thanks 🙏🏽
ai agents definitely be the most important path in the field of llm
Does seem so. Thanks!
Good content.
Imagine how many layers of Security is needed to allow agent to do financial transactions.
True, the frameworks have to consider the security and governance aspects. We have done that in the past but AI Agents is tricky. So will evolve as we roll these out.
I think we still have a ways to go before LLMs are reliable enough to just hand off entire workflows to them. I think they are great at specific NLP tasks, but I haven't seen them handle every situation in a super reliable way - which means I couldn't build an automated solution that way and hand it off to a customer.
Agreed. It will likely be a while.
That’s the idea of AI Agent frameworks. Will make more videos on how that’s progressing. It’s still a WIP though.
The great challenge is access cloud, sap and crm as a salesforce e.g. After that, agents Will spread a lot
Yes, Enterprise data and apps wil be a challenge.
I have to say AI agent can replace all the programmers in SAP, Oracle , Microsoft , Google and other big tech companies
Yes, big changes coming. I think augment by replacing some of the work.
Thanks - learned a lot.
Glad it was helpful!
please tell me how to become an ai agent builder from scratch especially for non cs background people Thank you
Appreciate the comment. Will plan to make such a video.
Meanwhile look up building AI Agents with flowise or nocode. There are videos out there.
I just started learning it a few weeks ago with Examspm being run by Helena Liu. They're extremely good but the required tools and apps are too expensive for a beginner!😊😊
A very good effort! However, LLMs do have memory for specific details about past interactions with a user. Though there are techniques (e.g RAG) to mimic memory, by including past conversations within the context window along with the prompt.
Absolutely they do. I should have mentioned that explicitly. Thanks for the add! Coming up with a RAG video next.
Thank you, I finally get it now.
Thanks for the feedback. Truly appreciate it!
Good video mate
Thanks much mate
Such an informative video , thanks .
Appreciate the comments. Sorry they were all held for moderation.
Great video....I was just wondering that LLMs like Gemini, OpenAI are really powerful but how can we use their power???
I mean their responses are so random even at 0.4 temperature.
What if i want to use the output of the LLM to drive my workflow? For this the LLM must always output "exact" answer & not just "correct answer".
For Eg : To classify the issue "My Laptop is showing blue screen", LLM will classify it as "OS" / "Windows"/ "Blue screen error".
All are correct but to drive my workflow who is only looking for "OS" value condition, other suggestions will break my workflow in spite of being correct.
So we need a kind of control on output for automating using LLMs. That's where AI Agents will rock.
Need your advice on how to learn creating AI Agents using LLMs. Is there any formatting language, framework etc. to build agents in the form of prompts?
I have found Guidance as one but its bit difficult to learn. :)
Agreed, there are many videos. Look up videos on how to build AI agents - the current best are Crew AI or Langchain (needs coding) or Flowise (no code). I will also plan to make one.
Regarding Halucination issues, I am soon coming out with a video on RAG - pls watch that. There are multiple techniques and LLM providers are working on improving the reasning abilities as well.
thanks for the detailed comment.
Sir u mentioned point 3. Memory(long and short), I See in comments someone commented about techniques (e.g RAG). I am facing issue on this concept only that I am using an underline model like GPT, GEMINI etc and want to train them specifically on my own data and devise a algo to act in a certain direction but I am not sure how to save the state where underline model is aware of my data and customisation to give a directional response when ever someone prompts with my AI agent. Can you guide me in a direction to learn this or where is my understanding flawed.
Check out the next video on RAG coming up shortly.
@@BotNirvana sure!!
Video was informative. Just few feedback, Video: Jump cuts, zooms do loose eye contact and focus, transitions can be improved. Try integrating visual elements more. For audio the shhh sound is quite a noise, will suggest to use Audition. Overall video was excellent!
Many thanks for the feedback. Will take care next time.
Thanks for sharing this
Welcome and appreicate the comment!
Great video with a lot of good insights / updates, thank you for sharing Nandan 👍
Thanks Krish. Great to see you here. Appreciate the comment and feedback.
AI agents are versatile digital assistants capable of diverse tasks and interactions, whereas SmythOS is designed to optimize specific user experiences within its dedicated technological environment.
Agents sure are versatile!
Thanks for sharing the info....
My pleasure
Hi, I work on AI agents, and have a good knowledge of Langchain and Langgraph. Would like to have a chat with you about my startup.
Awesome Ujwal. Hit me up on Linkedin www.linkedin.com/in/nandanmullakara/
How do I become an A.i agent?
You make AI agents. You remains human 😃
I have been articulating these concepts with many peers at my work 9 months ago on exactly what you are mentioning here. I believe this is where AI agents will take AI to a whole new level and start changing industries. Such an amazing time to be part of. AI agent software engineers is where it will be at in 6-9 mths!
Glad you see it similarly. Looking forward to the AI Agent revolution!
very good video.. thank you so much.
Glad you liked it!
Thanks...
Welcome!
📌 How do you think AI agents will transform the tech landscape? Share your insights below 👇🏼
Thanks
Welcome!
Great video , lot of insight shared , i am sure things are going to be disruptive in next 5 years
Looking forward to it too. Thanks!
Good content....Naatil evadaya....
Thanks @stanleyvictorprakash. From Thrissur.
10s of TRILLIONS of dollars?
Annual global spend on ENTIRE IT this year is expected to be just over $5 trillion.
So you expect just AI Agents (latest flavor of rapid introduction of AI buwords) is going to exceed 2 years of global spend in the future ?
Just some RANDOM VERY LARGE ROUND NUMBER? Or is there some computation behind this claim ?
Pls check towards end of video for the source (Seqouia Capital).
Military application will be autonomous drone swarms.
Scary!
Ai is best
Atleast for now!
I wouldn't like to see an AI agent go ahead and book tickets for me before it shows me the itinerary and the ticket prices
Absolutely! Humans First.
I agree 100%
Glad oyu do!
Glad you do!
Cool
Thanks
Bro, just a constructive feedback. You talk a lot unrelevant stuff. The 11 min video could have been a 5 min crisp and to the point video. Unfortunately, I have to close the video after 5 min only since you didn't even start talking about the agents after 5 min
Many thanks for the feedback. I do start on AI agents earlier than mentioned but point taken.
Its not new, people are building them for 2 years now...
It's becoming more feasible now. I am talking beyond AI chatbots which I think is what you are referring as Agents.
@@BotNirvana majority of support chats are running on llm agents already. Youve been under a rock, my dude.
That's called new.