WEBVTT 00:00.040 --> 00:01.240 Okay, so here we go. 00:01.280 --> 00:02.640 I'm going to get rid of the logs. 00:02.640 --> 00:04.280 I'm going to add a first step. 00:04.440 --> 00:06.120 It's going to be on chat message. 00:06.120 --> 00:07.680 We'll keep it simple to start with. 00:07.720 --> 00:09.800 We'll come back later and juice it up. 00:10.080 --> 00:12.080 Let's add in an AI agent. 00:12.920 --> 00:14.760 It's like the stuff is so old hat to you. 00:14.920 --> 00:17.600 We'll we'll also tidy this up at a later point. 00:17.600 --> 00:19.040 We'll add in a chat model. 00:19.040 --> 00:20.080 I'm going to use OpenAI. 00:20.120 --> 00:21.680 You can use whatever model you like. 00:21.720 --> 00:23.040 We'll use deep seek earlier in the week. 00:23.040 --> 00:24.160 You could stick with that. 00:24.560 --> 00:27.360 I am going to pick like a top frontier model. 00:27.360 --> 00:30.800 I'm going to pick GPT 5.2. 00:31.160 --> 00:36.640 Uh, and uh, that's that's definitely the one that, that I'd like to, to pick for here. 00:36.920 --> 00:41.680 Um, but you can, you can feel free to, uh, to use whichever one you want to use this. 00:41.680 --> 00:46.480 This should be very cheap, but obviously a mini would be, would be even even cheaper fractions of 00:46.480 --> 00:46.960 a cent. 00:47.160 --> 00:53.120 Um, and I'm going to add the option reasoning effort low because I want it to be quick and relatively 00:53.120 --> 00:54.560 cheap using 5.2. 00:54.680 --> 00:56.600 So that is a sensible choice. 00:56.800 --> 00:57.840 That is our model. 00:57.840 --> 01:00.440 We're not going to bother with memory because this isn't going to be a conversation. 01:00.440 --> 01:05.370 It's going to be one message in and then the results come out all right. 01:05.370 --> 01:09.490 And now it's time for us to think about what we want to do with Pipedrive. 01:09.650 --> 01:11.490 And so let's go to tools as that. 01:11.490 --> 01:12.650 It would be the most natural. 01:12.650 --> 01:18.410 First choice would be let's equip our agent with tools that will allow it to take a message like, please 01:18.410 --> 01:20.210 load this person into Pipedrive. 01:20.210 --> 01:21.690 And then to do that in Pipedrive. 01:21.690 --> 01:23.890 And indeed there is a Pipedrive tool. 01:24.010 --> 01:26.330 It's right here and we can come into it. 01:26.530 --> 01:31.130 And uh oh, it already knows my credentials, so I will wipe that so we get to set it up. 01:31.370 --> 01:34.770 But, uh, we then would choose, what do you want it to do? 01:34.810 --> 01:38.770 Do you want it to create a lead, an organization or a person? 01:39.010 --> 01:43.570 And if you create a lead, which is what we want to ultimately do, it needs to be associated with an 01:43.570 --> 01:45.570 organization with an ID. 01:45.810 --> 01:47.250 So we would have to do that. 01:47.250 --> 01:52.090 And similarly, if we, if we, uh, we'd want to create an organization and we'd also which, which 01:52.090 --> 01:53.010 needs a name. 01:53.370 --> 01:59.970 And here you'll see an example of something which is a bit tiresome, which is that there are some constraints 02:00.110 --> 02:02.350 around some of the way that these things work. 02:02.390 --> 02:07.870 It appears that unless it's just a current temporary feature of this tool, that you can't actually 02:07.870 --> 02:12.790 get the model to set the name of an organization when it creates it. 02:12.830 --> 02:16.750 However, you can do it when it updates an organization. 02:16.790 --> 02:17.230 There you go. 02:17.270 --> 02:19.230 Let the model define this parameter. 02:19.350 --> 02:23.990 So we'd actually have to have a tool to create the organization and then update its name. 02:24.030 --> 02:27.070 We'd have to create it with a with a temporary name in there. 02:27.070 --> 02:30.990 So there would actually need to be four different tools that we need to equip the LM with. 02:31.030 --> 02:38.110 And we need to teach the LM to take the ID that it gets back from one tool call and put it into another. 02:38.550 --> 02:43.190 And now that's expecting a lot from an LM because it's quite fiddly. 02:43.190 --> 02:46.950 And to be honest, it's something which lends itself really well to code. 02:46.990 --> 02:53.390 We're simply saying, look, always call this first with the ID that gets returned from this, I want 02:53.390 --> 02:54.430 you to call this. 02:54.590 --> 02:57.710 So it's really a set of of workflow steps. 02:57.710 --> 03:02.920 It's a set of steps that should happen one after another with the data from one going into the other. 03:02.920 --> 03:05.360 And that should be a flag for you. 03:05.360 --> 03:07.400 That should be a bit of a red flag for tools. 03:07.400 --> 03:09.720 We don't need autonomy here. 03:09.720 --> 03:13.320 We don't want to give the LLM that much flexibility. 03:13.440 --> 03:20.840 It seems like a better use case for using workflow for using core nodes, not tools. 03:20.840 --> 03:25.520 And that's why your mind should be going to to structured outputs. 03:25.680 --> 03:32.960 This is a great example of where structured outputs and workflow is going to perform better than tools. 03:33.000 --> 03:33.360 Okay. 03:33.400 --> 03:34.920 So I'm going to delete this. 03:35.080 --> 03:35.720 Get rid of it. 03:35.720 --> 03:37.680 From there we're going to stick with this. 03:37.680 --> 03:42.960 And we're now going to build out something which is going to need the agent to produce structured output 03:42.960 --> 03:45.080 in a format that's going to be perfect. 03:45.080 --> 03:46.200 So let's do that. 03:46.200 --> 03:52.280 We're going to to come in here and we're going to say I do require a specific output format. 03:52.480 --> 03:55.720 And once I've done that, you'll see that the output parser has appeared. 03:55.720 --> 03:59.000 And we will choose a structured output parser. 03:59.120 --> 04:00.010 Here we have it. 04:00.050 --> 04:05.570 Okay, so the agent is going to be responsible for taking in some input in text that describes some 04:05.570 --> 04:06.130 lead. 04:06.130 --> 04:13.130 And outputting that same thing turned into something that's structured and organized based on its understanding. 04:13.210 --> 04:18.210 So the JSON that we want is going to have something that is going to describe a lead. 04:18.210 --> 04:25.530 So maybe it's going to have like name the name of the lead, the full name of the lead. 04:28.210 --> 04:28.890 Company. 04:31.210 --> 04:35.130 The company where they work. 04:38.090 --> 04:38.850 Role. 04:41.410 --> 04:43.690 Their job title. 04:45.970 --> 04:46.770 Email. 04:48.610 --> 04:52.850 Their email address. 04:53.490 --> 04:54.330 That's it. 04:54.530 --> 04:59.170 So we're using an LLM to take natural language as an input and output it. 04:59.300 --> 05:01.340 organized and structured in this form. 05:01.340 --> 05:03.260 It's a great application of llms. 05:03.300 --> 05:07.740 They're great at taking structured data and turning it into something that's unstructured. 05:07.860 --> 05:09.580 Uh, and the other way as well. 05:09.780 --> 05:13.060 Uh, either way, they can they can parse and they can parse. 05:13.220 --> 05:15.540 Uh, which is, which is part of their magic. 05:15.700 --> 05:21.140 Uh, so this is going to be a really effective way to use GPT 5.2. 05:21.260 --> 05:21.700 Okay. 05:21.740 --> 05:28.940 Now I'm going to go back to the, uh, agent, and I'm going to set a system message to tell it what 05:28.940 --> 05:29.620 we want it to do. 05:29.660 --> 05:35.180 Basically what I just said, instead of that, I'm going to say this based on the input from the user, 05:35.180 --> 05:38.660 you should reply with a structured version of the lead. 05:38.780 --> 05:42.780 Your reply will be sent to Pipedrive to create a lead record. 05:43.100 --> 05:50.580 It's a bit clear based on the input of the user which describes a sales lead. 05:50.940 --> 05:51.420 There we go. 05:51.460 --> 05:52.900 That seems very clear to me. 05:52.900 --> 05:53.780 So nice. 05:53.780 --> 05:55.020 Simple directions. 05:55.020 --> 06:01.440 It's taking some unstructured data and it's turning it into something that can be put in Pipedrive in 06:01.440 --> 06:02.640 JSON format. 06:03.080 --> 06:03.800 All right. 06:03.840 --> 06:04.760 So far, so good. 06:04.760 --> 06:06.440 Let's come over here now. 06:06.440 --> 06:08.880 So this agent is going to take some chat message. 06:08.880 --> 06:11.800 It's going to turn it into that structured output. 06:11.960 --> 06:15.760 And I'm saying that's kind of equivalent thinking of it as making tool calls. 06:15.760 --> 06:21.640 Because what we can now do is put in in here not a tool but a workflow step. 06:21.760 --> 06:30.960 Pipedrive of course, has a um, a node which has node operations, which includes an operation to create 06:30.960 --> 06:32.160 an organization. 06:32.160 --> 06:33.480 And that's the first thing we're going to do. 06:33.520 --> 06:38.440 We're going to create an organization, uh, and we're going to I'm going to come back to this in a 06:38.440 --> 06:38.720 second. 06:38.720 --> 06:43.560 But when we've done that, we're then going to also use a Pipedrive node. 06:44.240 --> 06:49.280 And here we are going to come in and we're going to create a person record. 06:49.440 --> 06:52.880 So create an organization and then create a person. 06:52.880 --> 06:53.440 Here we go. 06:53.440 --> 06:54.440 Create a person. 06:54.880 --> 06:58.930 And once we've done that we are then let's resize this. 06:58.970 --> 07:01.730 We are then going to do another Pipedrive node. 07:03.370 --> 07:07.730 And finally after doing all of this we're going to create a lead. 07:09.050 --> 07:10.130 So you can see what I mean. 07:10.130 --> 07:13.610 We're just laying out the uh the workflow here. 07:13.650 --> 07:18.170 Here it is three steps based on the JSON that gets outputted from this agent. 07:18.170 --> 07:19.290 Create an organization. 07:19.290 --> 07:21.090 Create a person, create a lead. 07:21.130 --> 07:23.090 Because we just wanted to do it in that order. 07:23.130 --> 07:25.810 There's no point in giving that as a bunch of tools to the model. 07:25.810 --> 07:27.450 We might as well lay it out. 07:27.530 --> 07:29.010 Now we just have to populate these. 07:29.050 --> 07:32.450 Well, the first step to populating it is to set the API key. 07:32.650 --> 07:33.890 And that's pretty simple. 07:33.890 --> 07:35.930 You can see I deleted my the one that was already there. 07:35.970 --> 07:37.610 I press Create New credential. 07:37.650 --> 07:43.170 I click here and I paste in the token that I've got over here from Pipedrive. 07:43.210 --> 07:44.530 Let's go and copy that now. 07:44.570 --> 07:44.890 Okay. 07:44.930 --> 07:46.930 So obviously I go to Pipedrive. 07:46.930 --> 07:48.370 I go to Personal Preferences. 07:48.370 --> 07:49.530 I click on API. 07:49.570 --> 07:52.370 Press the green button to copy your API key. 07:52.530 --> 07:54.050 Come back over here. 07:54.090 --> 07:55.410 Come in here. 07:55.450 --> 07:56.530 Select credential. 07:56.530 --> 07:58.090 Create a new credential. 07:58.180 --> 07:58.900 click here. 07:58.900 --> 08:00.620 Paste in your token. 08:00.620 --> 08:01.540 Press save. 08:01.580 --> 08:02.300 Drum roll. 08:02.340 --> 08:03.140 Green. 08:03.180 --> 08:04.300 Successful. 08:04.300 --> 08:05.460 Congratulations. 08:05.460 --> 08:06.900 We're connected to Pipedrive. 08:06.940 --> 08:09.820 It automatically does that to all of these guys. 08:09.820 --> 08:11.500 We are in good shape. 08:12.020 --> 08:16.740 Next up we need to populate the name of this organization. 08:16.740 --> 08:19.020 And we also need to populate the name of the person. 08:19.020 --> 08:23.580 And we need to populate details of the lead as well and how to associate it. 08:23.580 --> 08:25.660 For now I'm going to leave all of this blank. 08:25.900 --> 08:30.820 I just want to kick this off and see what happens, so I can populate the data so that we can do our 08:30.820 --> 08:32.140 drag and drop stuff. 08:32.140 --> 08:33.020 Let's do that now. 08:33.060 --> 08:37.060 Okay, so when we test this out we're just going to have to turn these guys off. 08:37.100 --> 08:39.580 Otherwise they will complain because they're not populated yet. 08:39.580 --> 08:41.580 So I'm pressing the power button on all three of them. 08:41.580 --> 08:42.700 So they're all turned off. 08:42.740 --> 08:49.420 Then we'll start up the chat and I'll say, please record this lead. 08:50.100 --> 08:57.960 John Smith, CFO at IBM, John Smith at IBM. 08:58.200 --> 08:58.920 Com. 08:59.240 --> 09:02.600 So this is going to be just natural text. 09:02.600 --> 09:07.680 And the other way of doing it would be for this agent to be equipped with various tools and to try and 09:07.680 --> 09:10.000 call all the different tools, but that's not what we're doing. 09:10.000 --> 09:16.360 Instead, we're telling it to use structured outputs to return that unstructured form and pass it on. 09:16.360 --> 09:17.720 And this is what it returned. 09:17.720 --> 09:19.560 And you can see it did just that. 09:19.560 --> 09:23.520 And we can think of this as being analogous to to calling a series of tools. 09:23.520 --> 09:27.760 But that's what we're actually going to do ourselves in this workflow right now. 09:27.880 --> 09:28.320 Okay. 09:28.360 --> 09:30.040 So let me just clean it up with that button. 09:30.040 --> 09:31.760 Let's hide this for a moment. 09:31.920 --> 09:32.800 Let's come here. 09:32.800 --> 09:34.560 Let's reactivate this guy. 09:34.600 --> 09:35.600 Let's double click. 09:35.880 --> 09:36.720 In we come. 09:36.840 --> 09:39.520 So this is creating an organization. 09:39.520 --> 09:41.120 What name do we give it? 09:41.160 --> 09:43.680 In the past we've said let the model decide. 09:43.680 --> 09:45.600 But we can't do that because it's not a tool. 09:45.600 --> 09:46.920 It's just part of the workflow. 09:46.960 --> 09:48.320 We have to select the name. 09:48.360 --> 09:49.840 Well look we have the name right here. 09:49.880 --> 09:51.440 Sorry, the name of the company right here. 09:51.480 --> 09:52.520 It's an organization. 09:52.520 --> 09:53.520 We just drag that in. 09:53.560 --> 09:55.920 It's of course dollar JSON output company. 09:55.920 --> 09:56.560 There it is. 09:56.600 --> 09:58.570 IBM is what will be populated with. 09:58.690 --> 09:59.210 Done. 09:59.490 --> 10:00.050 Okay. 10:00.690 --> 10:01.370 Next. 10:01.730 --> 10:02.570 This one over here. 10:02.610 --> 10:03.530 Turn the power off. 10:03.570 --> 10:04.370 Double click. 10:04.410 --> 10:08.010 It's a little bit more involved because we can't take something from the previous node. 10:08.010 --> 10:09.810 We have to go back one node. 10:09.970 --> 10:14.490 And if you don't remember the syntax for that it's quite easy because over here on the left, if you 10:14.530 --> 10:18.250 scroll down, you'll see that it has the prior nodes. 10:18.250 --> 10:20.370 You can just expand a prior node. 10:20.410 --> 10:22.570 The AI agent is the node we care about. 10:22.610 --> 10:26.130 And here you'll find this is the person record the name John Smith. 10:26.130 --> 10:27.210 That's what we care about. 10:27.250 --> 10:28.410 Drag it and drop it. 10:28.410 --> 10:34.970 And that is the expression dollar AI agent JSON output name. 10:35.130 --> 10:35.930 Fine. 10:36.090 --> 10:38.810 You can have your cunning expression. 10:38.930 --> 10:41.690 Uh, we believe drag and drop work just fine. 10:41.690 --> 10:42.810 This is going to work. 10:42.850 --> 10:43.610 Thank you. 10:43.770 --> 10:46.890 Uh, so that is the name now populated. 10:47.050 --> 10:47.610 Okay. 10:48.010 --> 10:48.690 Escape. 10:48.970 --> 10:49.690 Next. 10:49.730 --> 10:50.130 Okay. 10:50.170 --> 10:51.290 And now the juicy part. 10:51.290 --> 10:53.450 We want to create a lead in our database. 10:53.450 --> 10:55.250 This is what we're really trying to do all along. 10:55.290 --> 10:56.450 We're going to turn the power back on. 10:56.450 --> 10:57.820 We're going to double click. 10:57.820 --> 10:58.460 Here it is. 10:58.500 --> 10:59.340 We need to create a lead. 10:59.340 --> 11:00.620 The lead needs a title. 11:00.660 --> 11:04.580 Now we can populate that with a title that we want for the for the title of the lead. 11:04.620 --> 11:09.660 We can have this be something like, uh, we can have it be the name of the of the person, I guess. 11:09.700 --> 11:13.980 And then, um, and then the, the, uh, the, the lead next to it. 11:13.980 --> 11:20.540 So to do that again, we scroll all the way down, we find the name of the AI agent that is the node. 11:20.540 --> 11:25.180 And we're looking here probably for, for the, the title of the lead should just be the name of the 11:25.180 --> 11:26.060 person I think. 11:26.380 --> 11:30.420 Um, and because normally the lead represents a potential deal. 11:30.420 --> 11:34.540 So this is something that, uh, that could turn into a deal, but I think that's probably how you'd 11:34.540 --> 11:35.260 normally do it. 11:35.300 --> 11:42.340 We want to associate it with an organization and that has an organization ID, and this is where we 11:42.340 --> 11:47.300 might get a bit confused because where are we going to find that organization ID from? 11:47.300 --> 11:49.860 The AI agent didn't tell us the organization ID. 11:50.180 --> 11:54.980 Well, I have news for you that is going to come from this node right here. 11:54.980 --> 12:01.070 When we create an organization, it's going to output the ID and it's just going to be called ID. 12:01.390 --> 12:06.350 And so we can actually take this ID here and drop that in here. 12:06.550 --> 12:09.350 And we're seeing that we're getting create an organization. 12:12.030 --> 12:15.230 That's going to be the organization ID okay. 12:15.550 --> 12:21.350 And you know what we want to lead to be associated not just with an organization but also with a person. 12:21.350 --> 12:25.950 And we can populate a person ID and we want that to come from this node. 12:25.950 --> 12:30.350 Create a person that also has an ID that comes in the response. 12:30.390 --> 12:31.310 These are coming in. 12:31.350 --> 12:32.030 That didn't work. 12:32.150 --> 12:40.190 Uh, expression uh, these are coming in the response back from uh from from this, this previous node. 12:40.470 --> 12:45.150 So I dropped that in there JSON ID because it was just the last node. 12:45.150 --> 12:50.990 And that is now going to, to populate the ID that came back from that prior node. 12:51.430 --> 12:52.430 And we're done. 12:52.550 --> 12:53.750 This should be it. 12:54.150 --> 12:57.130 So I'm just going to recap in case you weren't following me there. 12:57.130 --> 13:03.650 So we're taking all of the data that's flowing in from prior nodes and using that to populate this new 13:03.690 --> 13:04.810 lead record. 13:04.850 --> 13:06.490 And in fact, there are there are three things. 13:06.490 --> 13:08.810 We're setting the title of the lead overall. 13:08.810 --> 13:15.490 We're just setting it to be the the name field that came in from from the agent, the organization ID 13:15.490 --> 13:23.130 and the person ID that we're populating based on the output of creating those records in Pipedrive from 13:23.130 --> 13:30.090 from these prior two nodes, they both output an ID and we're pushing that ID in when we create the 13:30.090 --> 13:30.570 lead. 13:30.570 --> 13:33.770 And that's such a great example of stitching together data. 13:33.770 --> 13:39.450 And it's I hope you I hope this is a eureka moment for you, because you see how much better suited 13:39.450 --> 13:46.690 this is to a workflow in N810 when you're managing the data using a product that's ideally suited for 13:46.690 --> 13:51.810 it, than trying to futz around with tools and telling the agent, hey, you've got to move the ID from 13:51.810 --> 13:54.050 the output of this tool to the input of this tool. 13:54.370 --> 13:55.530 That would be messy. 13:55.570 --> 13:57.010 This is going to work great.