WEBVTT 00:00.080 --> 00:01.000 They're eating here. 00:01.000 --> 00:06.640 And in this video we're going to talk about the definition and what are agents. 00:06.680 --> 00:07.600 AI agents. 00:07.880 --> 00:12.480 Now if you're going to ask two different people what are AI agents. 00:12.480 --> 00:14.880 You're going to get ten different answers. 00:14.960 --> 00:19.200 So within those ten different answers, they all have some common grounds. 00:19.200 --> 00:22.080 And this is what I'm going to cover in this video. 00:22.080 --> 00:25.160 And I'm going to give you my $0.02 and how I view agents. 00:25.400 --> 00:27.560 It's pretty aligned with the LinkedIn ecosystem. 00:27.920 --> 00:37.600 An agent is a software system that uses Llms as a reasoning engine to decide what actions to take, 00:37.840 --> 00:40.480 and then execute those actions. 00:40.920 --> 00:49.200 So unlike simple chains in renewables, where the sequence of actions are hard coded agents dynamically 00:49.200 --> 00:56.880 determined which tools or steps need to be taken to solve a specific task, or to answer specific questions. 00:57.440 --> 01:03.450 The important thing to note here is that the LLM in an agent is deciding what to do next. 01:04.570 --> 01:05.090 Where? 01:05.370 --> 01:06.570 In chains. 01:06.610 --> 01:11.610 We might use an LLM, maybe to summarize something or to generate some text. 01:11.770 --> 01:16.490 However, we as developers, we define the entire control flow. 01:16.650 --> 01:19.250 We defined what is going to happen next. 01:19.450 --> 01:25.530 And the LLM is maybe used in one of those steps, but it's not deciding what to do next. 01:25.530 --> 01:29.730 So this is the main difference between an agent and a regular chain. 01:30.210 --> 01:39.170 Now the core manifestation of agents in the real world is the example of taking an LLM and equipping 01:39.170 --> 01:47.050 it with tools so we can give it, maybe a tool to make an API call, to do a search, to read from a 01:47.050 --> 01:50.490 database, or to simply write and execute some code. 01:50.650 --> 01:57.930 And this means that we can basically give our llms superpowers to do anything, because we can connect 01:57.930 --> 01:59.970 them to do anything with agents. 01:59.970 --> 02:05.780 And every time I think about this idea, it's simply mind blowing because we know how powerful large 02:05.780 --> 02:06.820 language models. 02:06.820 --> 02:14.140 If we give them tools and ability to do stuff, then we're going to see lots of innovation, a lot of 02:14.140 --> 02:20.340 complex stuff we couldn't do before, and we can automate a lot of stuff that we couldn't because now 02:20.340 --> 02:24.780 we have this LLM, which is a powerful tool which is going to mimic a human right. 02:24.820 --> 02:26.900 So let's talk in a very high level. 02:26.940 --> 02:29.020 What is a react agent. 02:29.300 --> 02:36.540 And it's a specific type of an agent architecture that follows the react paradigm. 02:36.660 --> 02:40.620 Now the name react comes from reasoning in acting. 02:40.820 --> 02:47.940 And this was a paper, um, that was published and was really the stepping stone of implementing agents. 02:48.500 --> 02:55.300 And in this approach, we combine the reasoning power of llms to think through the problem with chain 02:55.340 --> 02:59.140 of thought prompting, and I cover both of them in the theoretical section. 02:59.140 --> 03:01.340 So if you want, I elaborate on them. 03:01.900 --> 03:07.540 And after the LLM reasons about the problem, it Who decides what we need to do? 03:07.580 --> 03:08.140 Right. 03:08.180 --> 03:09.860 So he decides about the acting. 03:09.900 --> 03:13.340 And then we simply go and do what the Lord tells us to do. 03:13.340 --> 03:15.220 So we take those actions. 03:15.380 --> 03:20.700 Now those actions are going to be manifested through tools. 03:20.980 --> 03:21.940 What are tools? 03:21.980 --> 03:27.820 The gist of it is that there are the abilities that we give the LMS, and they can be making an API 03:27.820 --> 03:35.020 call, calling a database, or simply running some Python code that we wrote beforehand because it's 03:35.020 --> 03:38.260 functions that we pre-write before running everything. 03:38.300 --> 03:45.180 Then we have the freedom to to really write anything we want, and we can equip our LM with any capability 03:45.180 --> 03:46.100 that we want to. 03:46.620 --> 03:52.100 Now this is going to happen in an iterative loop until the task is complete. 03:52.620 --> 03:59.820 And link chain and link graph provides us with pre-built react agents that we can create and customize 03:59.820 --> 04:00.420 easily. 04:00.740 --> 04:03.180 And these agents can invoke tools. 04:03.180 --> 04:08.630 They can handle complex workflows and maintain state over long running tasks. 04:09.790 --> 04:14.190 Okay, now I know I was really waving my hands in this video. 04:14.230 --> 04:16.110 I didn't show you anything concrete. 04:16.110 --> 04:18.550 I didn't show you any code or any demo. 04:18.590 --> 04:21.670 This was really an intro to the intro. 04:21.710 --> 04:27.110 Okay, so in the next video I will show you a demo of a search agent examples. 04:27.110 --> 04:32.750 And in this section we're going to be implementing this search agent with Link chain. 04:32.910 --> 04:39.830 Now we're going to be implementing via the react architecture which is the most important architecture 04:39.830 --> 04:40.510 in agents. 04:40.510 --> 04:42.350 And it's the basis for everything. 04:42.350 --> 04:45.790 And in this section we are simply going to learn the interface. 04:45.790 --> 04:49.110 So how do we equip our LM with tools. 04:49.110 --> 04:51.110 This is the entire goal of this section. 04:51.110 --> 04:52.630 Just how to do it with link chain. 04:52.870 --> 04:57.550 And in the section afterwards we're going to see really how it's built under the hood. 04:57.550 --> 05:01.430 And we're really going to understand every bits and bytes of this. 05:01.630 --> 05:09.550 Now this is super interesting and this is the most important topic in LM development applications in 05:09.550 --> 05:10.150 my opinion.