WEBVTT 00:00.080 --> 00:00.440 Okay. 00:00.480 --> 00:04.000 That was the first of our two agents and the bigger of the two for sure. 00:04.160 --> 00:06.160 And very satisfying, I have to say. 00:06.560 --> 00:08.360 And now for the second one. 00:08.400 --> 00:11.200 The second one, the first one was the outbound sales. 00:11.200 --> 00:14.200 The second one is the account executive. 00:14.360 --> 00:17.000 The person that takes inbound calls. 00:17.240 --> 00:23.320 The account executive agent is going to have just two subagents, one of them that is able to record 00:23.320 --> 00:27.520 that a deal is in progress and the other is able to book a demo. 00:27.680 --> 00:33.760 These are the two activities that will be carried out by Subagents under the watchful eye of the account 00:33.760 --> 00:34.480 executive. 00:34.520 --> 00:35.920 Let's go and build them. 00:35.920 --> 00:36.240 Okay. 00:36.280 --> 00:38.280 Here we are in our amplify project. 00:38.280 --> 00:43.720 I'm now going to go and create a new folder and I'm going to call it Account Executive. 00:44.760 --> 00:45.440 There we go. 00:45.560 --> 00:49.760 So we now have our business development and our account executive folder. 00:49.760 --> 00:51.560 Nothing in there just yet. 00:51.680 --> 00:55.440 Let's go in and create our first workflow. 00:55.560 --> 01:02.160 And our first workflow which is going to be called the deal according Subagent. 01:03.840 --> 01:04.720 Okay. 01:05.120 --> 01:07.400 What is this deal recording subject going to do? 01:07.400 --> 01:08.440 Well, let's begin with it. 01:08.480 --> 01:12.800 Having a trigger of a chat message, uh, which we will change later. 01:12.800 --> 01:17.320 And it's going to, as usual, go straight in to an AI agent. 01:17.480 --> 01:18.440 There it is. 01:18.720 --> 01:19.160 Okay. 01:19.200 --> 01:20.480 What are we going to do with this agent? 01:20.480 --> 01:24.720 Well, I'm going to start by having it use the usual OpenAI chat model. 01:24.720 --> 01:26.240 You may pick your own. 01:26.560 --> 01:32.840 I'm again going to go with my favorite GPT 5.2, which will give me nice and strong model. 01:32.840 --> 01:33.840 There it is. 01:33.880 --> 01:39.880 And but I am going to tell it just to have the low reasoning effort, uh, just so that it's nice and 01:39.880 --> 01:42.160 quick and doesn't take too much. 01:42.480 --> 01:43.440 There we have it. 01:43.600 --> 01:48.240 Uh, we, uh, we we won't need memory because we'll just be giving it one instruction at a time. 01:48.520 --> 01:52.240 And now it's time for us to look at the tools. 01:52.240 --> 01:53.680 So what's this agent going to do? 01:53.840 --> 01:59.600 It's going to be informed that some customer is interested in having a demo, which means that it's 01:59.600 --> 02:07.720 time to record that this customer is potentially in a deal flow in our CRM in Pipedrive, you say that 02:07.720 --> 02:08.480 you've got a deal. 02:08.480 --> 02:12.120 And in something like Pipedrive, not necessarily because you've yet won the business, but because 02:12.120 --> 02:16.840 there's somewhere in the deal pipeline that hopefully will close and become a deal in the future. 02:17.160 --> 02:21.440 So that's what we want to do, which means that really, this AI agent needs to have the ability to 02:21.480 --> 02:26.560 read from Pipedrive, to find out the people who've got in Pipedrive to identify a person and then to 02:26.560 --> 02:28.600 be able to create deals. 02:28.880 --> 02:29.920 Now, here's the thing. 02:29.920 --> 02:33.360 I'm actually going to use tools and structured outputs. 02:33.360 --> 02:38.320 In this case, I'm going to use a tool to read all of the people, but I'm going to use structured outputs 02:38.320 --> 02:43.440 to write the new information so that we can be crisp and and bulletproof about it. 02:43.440 --> 02:45.960 And that seems like a good balance to me. 02:46.000 --> 02:48.480 Now, of course, you might say, how did I come up with that? 02:48.480 --> 02:51.920 And the answer is you should experiment and you should see what works best. 02:51.920 --> 02:54.520 I know you know this saying, shut up and get on with it. 02:54.760 --> 02:55.400 All right. 02:55.440 --> 03:00.160 So we're going to add a tool which is going to be a Pipedrive tool, which is going to be the tool to 03:00.200 --> 03:01.560 get people. 03:01.560 --> 03:06.000 We're going to want to have a resource will be a person. 03:06.240 --> 03:09.200 And I'm going to make this to be get many. 03:09.600 --> 03:14.970 And as I say, if you were building this to be something that's ready for prime time, you'd want to 03:14.970 --> 03:19.850 put in filters here and have the model be able to filter on the people it's looking for, but I'm just 03:19.850 --> 03:23.090 going to have it return all for for this so that we don't get bogged down. 03:23.090 --> 03:26.450 And that's going to be fine up until a reasonably large number. 03:26.570 --> 03:28.410 Uh, so that's how it's going to work. 03:28.410 --> 03:33.930 It's going to get all of the people, and that should work just fine for this. 03:34.210 --> 03:38.610 And now we want the agent to output structured outputs okay. 03:38.650 --> 03:40.610 So I open up the AI agent. 03:40.610 --> 03:41.690 Here it is. 03:42.010 --> 03:47.050 I am going to first give it a system prompt so we know what it's going to do. 03:47.330 --> 03:48.650 Here we go I'll paste one in. 03:48.690 --> 03:52.570 You're provided with a potential name of a person in the Pipedrive database. 03:52.570 --> 03:55.650 You use your tools to see whether that person is already in the database. 03:55.650 --> 03:58.490 If there are duplicates, you pick one reply with the details. 03:58.490 --> 04:02.370 You find a deal will be created in Pipedrive as a result. 04:02.650 --> 04:03.450 That's super clear. 04:03.450 --> 04:06.890 So now we need to require a specific output format. 04:07.130 --> 04:08.370 Let's go back. 04:08.370 --> 04:14.970 Now we've got this here and we're going to say structured output parser and define the output we want. 04:15.010 --> 04:15.850 And here we go. 04:15.890 --> 04:20.050 I'm going to have it be a JSON object with four fields found. 04:20.050 --> 04:25.730 Whether or not the person was found in the people, the id the person ID goes there, the company, 04:25.730 --> 04:29.810 the company name goes there, the name, the person's name goes there. 04:29.970 --> 04:32.010 Very simple, very clear. 04:32.250 --> 04:34.610 That is going to be our structured output. 04:34.610 --> 04:39.730 And then based on that output we can then create the records in Pipedrive. 04:39.770 --> 04:40.250 Okay. 04:40.290 --> 04:43.690 So from what comes out of here let's put in an if node. 04:44.570 --> 04:47.970 If node, because we know that that this is going to come up with an output. 04:48.130 --> 04:50.250 Uh, let's make this an expression. 04:50.810 --> 04:58.250 And we know that what comes out of the previous cell dollar JSON output dot. 04:58.370 --> 05:01.410 And then one of the fields is called found. 05:01.610 --> 05:04.370 And if that's true then we have found a record. 05:04.370 --> 05:05.810 If it's false we have not. 05:06.250 --> 05:10.330 So now to not make the mistake I made before we go Boolean is true. 05:10.570 --> 05:11.490 There we have it. 05:11.530 --> 05:14.090 JSON output is true. 05:14.450 --> 05:15.770 That's the Tesla we want to make. 05:15.810 --> 05:17.730 We are either going to go true or false. 05:17.890 --> 05:24.330 If we go to true, then what we want to do is we want to carry out a Pipedrive node, a node action. 05:24.330 --> 05:26.850 We want to be to create a deal. 05:27.130 --> 05:28.450 Create a deal. 05:28.810 --> 05:29.690 All right. 05:29.690 --> 05:32.930 So what's the deal title going to be. 05:33.290 --> 05:41.370 Well, we'll have the deal title be this Jason name which is the name thing coming from from the the 05:41.370 --> 05:44.010 AI agent followed by the word deal. 05:44.170 --> 05:48.330 So if the person's name is John Smith, it will say John Smith deal. 05:48.330 --> 05:50.730 That will be the expression we'll use there. 05:51.090 --> 05:53.090 Uh, and we can associate it with something. 05:53.090 --> 05:54.130 Let's associate it. 05:54.170 --> 05:58.570 We choose between either associating a deal with an organization or with a person. 05:58.570 --> 06:00.330 Let's associate it with a person. 06:00.330 --> 06:02.330 We have to provide a person ID. 06:02.610 --> 06:03.850 Well, we have one of them. 06:03.970 --> 06:08.890 An expression uh, the expression should be dollar JSON 06:10.130 --> 06:12.970 output.id. 06:13.130 --> 06:18.130 That was the field coming out of our AI agent that is populated with the ID. 06:18.570 --> 06:27.130 So now we have called it John Smith deal, and we've linked it to John Smith's ID, and John Smith is 06:27.130 --> 06:32.930 the person who was found by the AI agent and outputted in the structured outputs. 06:32.930 --> 06:33.730 Right here. 06:33.770 --> 06:35.050 You see found is true. 06:35.090 --> 06:41.250 That's the ID and we've taken, uh, their, uh, their name and put the word deal at the end of it. 06:41.850 --> 06:42.450 Okay. 06:42.730 --> 06:45.330 So this let me tidy it up for you. 06:45.370 --> 06:46.050 There we go. 06:46.330 --> 06:52.090 So this should create a deal for a person if they are in the database. 06:52.650 --> 06:53.690 Time to give it a try. 06:54.050 --> 06:56.410 Just before we give it a try I did some tidying up. 06:56.410 --> 07:02.530 I just added to stickies and then I added a couple of push notifications just so that we we are kept 07:02.530 --> 07:05.730 informed about what's exactly is going on. 07:05.890 --> 07:08.130 Uh, make sure my phone is not on silent. 07:08.290 --> 07:09.010 Uh, okay. 07:09.050 --> 07:13.210 And with that, we can now give it a try and have some, some visual to it. 07:13.210 --> 07:18.050 Okay, so I bring up the chat and what I'm going to say is if we look in Pipedrive, let's look at one 07:18.050 --> 07:19.850 of our leads, Rob Carter. 07:20.130 --> 07:30.780 Uh, let's now come back here again and say, uh, record that Rob Carter has become a deal. 07:31.500 --> 07:32.940 And let's see what happens. 07:33.580 --> 07:35.260 Off it goes to our chat model. 07:35.260 --> 07:37.700 It's reading people from Pipedrive. 07:37.980 --> 07:39.700 It's done something and I got it. 07:39.780 --> 07:40.260 Did you hear that? 07:40.860 --> 07:44.180 I got a push notification and it created a deal. 07:44.380 --> 07:49.580 And if I go back to my leads inbox here, there is a section on deals. 07:49.740 --> 07:55.220 And there we have Rob Carter in there as our deal. 07:55.220 --> 07:57.460 We have our first deal in Pipedrive. 07:57.700 --> 07:59.020 That was pretty easy. 07:59.260 --> 08:00.580 Uh, excellent. 08:00.580 --> 08:01.340 Onwards. 08:01.460 --> 08:02.060 Okay. 08:02.100 --> 08:07.060 So now what we're going to do is turn this into a subagent, remove this chat received and add a new 08:07.060 --> 08:08.140 trigger node. 08:08.420 --> 08:12.460 And in the past we've, we've put in here when executed by another workflow. 08:12.500 --> 08:15.300 I'm going to do it differently this time for reasons that will become clear. 08:15.300 --> 08:18.020 But I'm going to have it be on webhook call. 08:18.020 --> 08:24.260 So the way that you trigger this is by hitting a URL, and we'll hit it with http post as we did before, 08:24.500 --> 08:27.780 which means we get these URLs that you can hit if you want to do it. 08:27.980 --> 08:30.780 Um, and, uh, this all looks fine to me. 08:30.780 --> 08:31.700 I think everything's good. 08:31.700 --> 08:33.460 We'll have this one respond immediately. 08:33.460 --> 08:34.140 We won't make it. 08:34.180 --> 08:35.660 Wait and then respond later. 08:35.660 --> 08:38.140 We'll just have it respond back immediately. 08:38.340 --> 08:40.100 And that's what we will do. 08:40.140 --> 08:43.020 That's going to be connected in here and in here. 08:43.020 --> 08:44.060 We just have to come in here. 08:44.060 --> 08:45.660 It's no longer expecting this. 08:45.700 --> 08:47.300 It's now needs to be defined. 08:47.300 --> 08:49.260 Below we'll have an expression. 08:49.460 --> 08:52.020 And so an expression needs to be in double curlies. 08:52.060 --> 08:55.820 And you may remember that it's it's got to be dollar Jason. 08:57.060 --> 08:59.020 That's the name of what comes in. 08:59.260 --> 09:06.180 And we'll assume that name that will be populated with name, which is going to be the name of the person 09:06.180 --> 09:08.060 in the in the Pipedrive database. 09:08.500 --> 09:11.260 So that is, uh, is what we're after. 09:11.260 --> 09:17.140 And that has now set this up to be a, uh, ready to be called as a webhook. 09:17.380 --> 09:21.620 And what we can now do is save it as already saved and publish it. 09:21.620 --> 09:27.060 And that will now be published out there waiting on its webhook that we can take later. 09:27.460 --> 09:28.260 Excellent. 09:28.500 --> 09:30.700 Now, on to the next Subagent.