WEBVTT 00:00.160 --> 00:03.680 So I'm in the newest LinkedIn documentation on agents. 00:03.680 --> 00:09.720 And we have here a button which is going to allow us to copy this documentation as text. 00:09.720 --> 00:16.320 And this is actually something super innovative and something very useful that the team did, because 00:16.320 --> 00:20.920 it's going to allow us to use all these documentation as context. 00:21.040 --> 00:27.200 We'll be sending to our AI powered code editor like Cursor or Cloud Code, which is going to help us 00:27.200 --> 00:29.720 figure out how to do things with the new docs. 00:30.000 --> 00:36.120 This is a very interesting topic, which is called lm dot txt, which I do actually cover later in this 00:36.120 --> 00:36.600 course. 00:36.600 --> 00:38.360 You can check out those videos. 00:38.480 --> 00:41.880 They are under the MCP section right. 00:41.880 --> 00:49.960 So now I want to go and migrate my code to be using link chain v1 to be using the new interface. 00:50.240 --> 00:52.240 So this is going to be actually very easy. 00:52.480 --> 00:55.920 And the way I'm going to do it is actually with cursor. 00:56.120 --> 01:03.560 Now I'm doing it not because I'm lazy, it's because you can really see the difference with the generated 01:03.560 --> 01:08.150 diff cursor is going to create for us so we can really see everything side by side. 01:08.150 --> 01:09.350 What is being removed? 01:09.350 --> 01:10.750 What is being added? 01:10.790 --> 01:13.630 This is actually easier to understand in my opinion. 01:13.910 --> 01:22.470 So let me just fire up cursor agent and let me prompt it something like migrate the code in Main.py 01:22.630 --> 01:24.910 to the newest create agent interface. 01:24.910 --> 01:26.270 And here is the docs. 01:26.270 --> 01:29.510 And here I'm simply pasting everything from the docs. 01:30.070 --> 01:30.590 All right. 01:30.590 --> 01:36.230 So I'm going to fast forward everything because this query may take a minute or two. 01:37.390 --> 01:40.950 And let me go and wait until we get here. 01:41.270 --> 01:41.870 The diff. 01:43.150 --> 01:43.630 All right. 01:43.630 --> 01:46.430 So this query took us around three minutes. 01:46.630 --> 01:47.870 And let's review the diff. 01:47.870 --> 01:52.550 And it's actually going to match what we saw in the documentation in the earlier video. 01:53.070 --> 01:58.350 So we do not need the prompt hub because we do not need the react prompt. 01:58.390 --> 02:02.630 We do not need the agent executor because this is all going to be abstracted. 02:02.630 --> 02:06.190 And this is all going to be running on a graph under the hood. 02:06.670 --> 02:12.340 We do not need the create react agent we imported instead the create agent function. 02:12.620 --> 02:18.820 We don't need prompt template because we're not writing a custom react prompt, and we do not need a 02:18.820 --> 02:25.260 runnable lambda because we're going to be using a built in argument, which is going to help us accept 02:25.260 --> 02:26.580 structured output. 02:26.740 --> 02:32.500 And you can see we do not need this special react prompt with format distractions, simply because we 02:32.540 --> 02:37.740 are not going to be using a react prompt, because the reasoning is going to happen through the function 02:37.740 --> 02:39.980 calling capabilities of LMS. 02:40.260 --> 02:40.940 Alrighty. 02:40.940 --> 02:47.380 So after we finished with the imports, let's go and discuss the implementation and we can see we still 02:47.380 --> 02:51.300 need the agent response schema and we still need the list of tools. 02:51.340 --> 02:58.740 These haven't changed, but we did remove a lot of code that is around the structured output and is 02:58.740 --> 03:01.060 around the react prompt, which we do not need. 03:01.300 --> 03:06.700 And we simply have here a model variable which is going to be an instance of chat. 03:06.700 --> 03:08.500 OpenAI with GPT four. 03:08.540 --> 03:11.020 Maybe we can edit it later for GPT five. 03:11.700 --> 03:12.300 All right. 03:12.300 --> 03:19.120 And let's check out the implementation itself, so we do not use any more the create react agent function. 03:19.280 --> 03:26.480 Instead, we are using the create agent which is going to receive an input the model that tools like 03:26.480 --> 03:27.360 we did before. 03:27.640 --> 03:33.440 And this time we're also going to provide it with a response format argument, which is going to be, 03:33.640 --> 03:38.600 um, the schema of the agent response, which we covered in earlier videos. 03:38.600 --> 03:44.440 So that's going to make sure that the answer that the agent is going to return us is going to be a pydantic 03:44.440 --> 03:46.640 object of agent response type. 03:47.280 --> 03:54.200 And we can see we do not need the agent executor so that while loop, we do not need it anymore, we 03:54.200 --> 03:56.600 do not need the extract output lambda. 03:56.600 --> 04:01.800 We do not need the chain to extract the output link chain is going to handle everything for us. 04:01.800 --> 04:02.200 Now. 04:02.200 --> 04:08.040 It's super important to note that I will be showing in this course how this is implemented. 04:08.040 --> 04:08.880 Within Lang graph. 04:08.920 --> 04:14.320 We will actually build everything from zero and you'll know exactly what's the magic over here. 04:14.680 --> 04:17.120 So right now we're just reviewing the interface. 04:17.560 --> 04:23.710 So let's show how to use it and how to invoke it, because there is one small difference. 04:23.910 --> 04:28.910 So now the input we're going to be sending is simply a list of messages. 04:28.910 --> 04:31.350 So it's going to have the role of the user. 04:31.350 --> 04:35.870 And it's going to have the same content as we had in our input key. 04:36.310 --> 04:38.670 And the reason why we do it. 04:38.670 --> 04:43.590 So why we send messages instead of sending in the key of input. 04:43.790 --> 04:46.630 This is something I'm going to leave open right now. 04:46.710 --> 04:49.550 But I promise you, we do answer in this course. 04:49.550 --> 04:54.870 Why this subtle change which actually encompasses within it a lot of depth in it. 04:54.870 --> 04:57.030 So I don't want to give any spoilers. 04:57.030 --> 04:58.430 We will be seeing it later. 04:58.710 --> 05:05.670 And lastly, in order to access the answer of the agent, which is going to be the structured response 05:05.670 --> 05:13.670 as we want it to be, we need to access the key of structured response within the result here after 05:13.670 --> 05:15.070 we invoke the agent here. 05:15.310 --> 05:19.550 Um, so this is simply how long chain is going to structure the answer. 05:19.590 --> 05:20.030 Okay. 05:20.230 --> 05:21.630 So this is it. 05:21.630 --> 05:23.030 Those are all the changes. 05:23.030 --> 05:27.860 And you can see it's actually much simpler now to use the create agent. 05:27.860 --> 05:31.420 And the interface got way more simpler. 05:31.660 --> 05:38.500 And there is actually a reason why I did not start immediately with this implementation, and simply 05:38.500 --> 05:44.700 because I want to show you the evolution for you to really appreciate and for you to understand the 05:44.700 --> 05:51.860 motivation and the architecture decision that the link chain team and the link chain framework made 05:52.100 --> 05:54.460 in order for it to be where it is today. 05:54.700 --> 06:01.380 And in order to see why this create agent function, why it has much more benefits over earlier versions 06:01.380 --> 06:05.060 of the create react agent or create tool calling agents. 06:05.180 --> 06:12.060 And it's going to give us a much more in-depth understanding of the ecosystem and agents in general. 06:12.100 --> 06:17.100 Okay, so I just remind you that up until now we only saw interface. 06:17.100 --> 06:20.820 So we haven't discussed what link chain react is doing under the hood. 06:20.860 --> 06:25.300 And even what is this function is doing under the hood and how it's working. 06:25.300 --> 06:29.460 And this is going to be happening right in the next couple of sections.