What is Loop Engineering? Agentic AI Engineering Explained!

Hey everyone, this is Lakshmi Kant Tiwari. Most of you have been asking about the loop engineering through my YouTube channel and the Udemy channel as well. One thing probably you don't know, you have been using this loop engineering since quite some time. So today we will be demystifying that. So this is something which we have seen since so many years that is known as React agent. That means reason and thereafter take the action. React agent work? You have your model here and inside model you give an input and then your model decide whether it need to call a tool or not. Thereafter your model decide whether it has answered your question or not. If it has not answered then it will keep calling the tool and it will keep doing this process until it has finished. Overall this you can say that it gives a sense of the loop engineering but it is not a loop engineering. Why? Let me tell you. So we all know there are many kind of engineering with the generative AI or with the respect of the LLM. Previously we had seen our prompt engineering. And then we had seen our context engineering and apart from that we had also seen their harness engineering. After this we have got our another engineering that is known as loop engineering. So the one thing you will notice here between all these four So these three what you see here, these three are applied here at the input side of your model or at the input side of your agent. And what you see here, this loop engineering that is applied at the output side of your agentic AI. Let's go ahead and break it down into multiple steps so that we can understand our loop engineering. So, what we did here, we had actually expanded our react agent into a five-step process. So, the one thing you will notice here, you have your action. This is where your model is going to sit. There after you have here the object. That means this is the output. All right? So, these three things what you see here currently, all these are making your react agent. But, if you add these additional block in your react agent, then it is going to work as the loop engineering. So, basically, the discipline or the technique which you are using to create this overall agentic behavior, that is known as the loop engineering. One more thing you will notice here, you are going to have here two model. All right? Not just a single model. In another way, you can say that it is like a multi-agentic behavior as well. So, that is why I told you at the start of this lesson that this is something which you have been using since quite some time, but you don't know you have been doing the loop engineering. So, why this term came? Basically, what happened here, earlier people used to use they used to give the different different name to it, but there after people came together and they decided to give it a proper name that is known as the loop engineering. So, what happened here? You have your model here which is working on your task, whatever the task you have given here as the input task, then its output is coming. There after you are going to have here the another model. One thing you need to put here your attention on this particular part, not the model that acted. Basically, you cannot use the same model. You cannot use the same model which was used to generate the output. You have to use here a different model. Why? Because if you use the same model for generating I mean if you use the same model which had used to generate this particular output, that same model is going to give you always a biased answer for its output because the same model cannot evaluate itself. That is why we are going to have here the separate model. So, the separate model is going to decide whether the output of this particular model is correct or not. If that output is not correct, then it is going to give here the prompt. Basically, it is going to give here the prompt and then this model is going to rework again on the particular problem statement. So, overall people have been giving its a name that is known as the self-prompting. So, what happened here? Let's say you have designed here a system. Again, think about this. Self-prompting is not like agent is going to self-prompt itself. That cannot be done because as you see here in this diagram, a single agent cannot work on this part. So, self-prompting I would say that here if you have created here a solution, in that solution you have a two agent agent one and the agent two. First prompt comes from you as the user. There after this agent two can self-prompt to the agent one. Basically, this overall solution can self-prompt itself where a multiple agents are working together and they are improving its output. Apart from that, one more thing here is very important in case of the loop engineering. Earlier what used to happen, our agent actually maintains its memory and observation everything in the estate. But in case of this loop engineering, what you do here, you maintain all memory everything outside of the agentic state so that if you have your multiple agents here, they can share these things here. So, basically you have your agent one which is working, then you have your agent two. All these agents can share this information so that this agent two can create a correction prompt and this agent one can correct itself. All right, everyone. So, this is all about in this lesson. I hope you have liked this video. The presentation which I'm currently using in this lesson that is already given in the video description. You can get that presentation from there. And also, please don't forget to subscribe this channel. In the next video, I'll be bringing the graph engineering so that you can keep yourself up to date. All right, guys. Bye-bye and take care.