Full transcript
0:00Welcome to the deep dive on how I'm
0:02building agents to enable my business to
0:05scale beyond anything that I could have
0:06ever done before pre- agents. Now, my
0:10journey is scaling Triggery, my company
0:13from 0 to 10 million using agents and
0:16keeping the team lean and mean. And
0:18today, we're going to go into one of the
0:19tools that I've basically uh been using,
0:22which is NA10. And I'm creating what we
0:24call a social agent. Now, Triggery gets
0:27a lot of brand mentions uh across
0:30LinkedIn, Reddit, YouTube, various
0:33different basically places. And I want
0:34to create a social agent that will
0:36notify my team anytime a critical
0:39mention happens on one of these social
0:41platforms. So, for example, is it a post
0:44where it talks about, you know, ease of
0:46use or difficulty? If so, the product
0:48team needs to be alerted. Is it a post
0:51where it's talking about buying signals
0:53or churning signals? if so sales or
0:55customer success team needs to be
0:57notified. So let's have a look at this
1:00agent high level to start with. So first
1:03of all I'm taking a web hook uh coming
1:05in from triggery which is a social
1:07listening uh tool and we're then using
1:10deepseek here to classify the post
1:13itself give me a certain kind of output
1:16before it basically does uh whatever we
1:18want to do with that data. So taking it
1:21a step over here, the first thing that
1:22I've done is I've gone to triggery which
1:24is a social listening tool and I'm
1:26basically monitoring uh for these
1:30particular keywords. I'm using a or
1:32operator and I'm using uh triggery or
1:36triggery.io and this is then sending
1:38data via a web hook through to this
1:41particular agent that we have here. Now
1:45this agent is essentially reading the
1:49post which comes via this content and at
1:53the moment it's saying okay hey well
1:54this is the department and if you're
1:56watching take a screenshot you can use
1:58the prompt um and then it's telling me
2:01the reason why and so what I'm doing
2:03here is I'm actually using a specific
2:06output format and the output format is
2:09very very simple it's basically uh I'm
2:11telling it it's going to be the
2:12department which would be something like
2:14that and then the reason which would be
2:16whatever it wants to give. And so the
2:18output then for me is product and the
2:20past mentions. Now, I want to actually
2:22take this a step further today. And so,
2:24we're going to edit the prompt here, and
2:27we're going to say um a few things. Uh
2:31if the
2:34department
2:35is marketing,
2:39uh add and we're going to we're going to
2:42say,
2:45uh add the email. And then we're going
2:48to
2:49do we're going to do a madeup one,
2:51right? Just for the purpose of this.
2:55Info@triggerfy.io. If it's
2:57product, add the email
3:01um productify.io.
3:06And if it's uh
3:08sales
3:10email, now you would obviously want to
3:12replace these with your own kind of uh
3:15you know emails and your own
3:17departments, right? And so what I then
3:19want to go do is I want to update the
3:21JSON here. And so I'm going to add this
3:24into here. And the reason that I'm going
3:28to do
3:30this is we could then build a workflow
3:34out the back end of this to actually
3:37email that particular individual. So
3:39let's test this and see if the results
3:41come out the way uh I'm hoping that they
3:45do. So in theory it should give product
3:49um uh and then the email. Yeah, which it
3:52has. Okay, perfect. So I'm just going to
3:53add that into the set field. And so what
3:56I'm also going to do is I'm going to go
3:57back to here and I'm going to add the
4:00uh LinkedIn URL that this particular
4:04person uh has. Right? So there's the
4:07particular post that the person did. And
4:10I can also add the content here as well.
4:14And so now what I could do is I could
4:17actually go and then
4:18basically send a uh
4:22email to the particular product owner of
4:26that feature. So for example, let's say
4:28you've got multiple product owners. I
4:29could then basically send that to that
4:32individual and the individual could be
4:34plain text saying like you know like hey
4:36you know John whatever it is
4:39um uh you could say hey John this uh has
4:45just come
4:47up a post talking about and then you
4:51could add like the actual um thing
4:54itself here post talking about if you do
4:56an expression uh post talking
5:02about
5:05that. I wanted to flag
5:09this for you in case you needed to
5:15jump on it,
5:18right? URL. And then you could add the
5:20URL into
5:22here. Uh we'll go up to here. The URL
5:27post URL. There it is.
5:29[Music]
5:31Um, and you can maybe even add the
5:34reason. Uh, wanted a flag for this for
5:37you. You could say reason
5:44being. Add that in
5:47there. Reason being user feedback about
5:50usability issues, which is valuable for
5:52the product improvement in case you want
5:54to jump on it. There we go. Nice. you
5:57know, best whoever, right? And then what
6:00you'd also want to do because
6:02uh urgent social action or maybe uh you
6:07could then actually uh CC people into
6:10here as well. So we could do like a BCC
6:11or a CC. Uh and this could be your
6:14social media manager, CMO, demand gen,
6:16you know, growth, whoever is responsible
6:19for social content at your company. But
6:22this is highly valuable workflow for
6:25individuals
6:27uh or companies that get a lot of
6:29organic mentions across multiple
6:31different social channels and is
6:33allowing you to then action things uh
6:35quickly. Now if you get so many you
6:38perhaps don't want to uh notify someone
6:41every single time. So what we could do
6:45is we could actually rather than sending
6:47an email, we could either score it on
6:50the severity of how important it is to
6:52reply. If it's high, then we email. Uh
6:55if it's low, we store the information.
6:58Or we could just store the information
6:59in general in like an air table, a G
7:01Sheets or somewhere similar and then
7:04have an agent at the end of each month
7:06or each week uh summarize the posts for
7:10the marketing product, finance, sales
7:12department, whatever is relevant as
7:14well. And maybe we'll build that agent
7:16in another
7:18video. But this is how we're using uh
7:21social agents in my live prototype that
7:23I have running. I'm actually sending it
7:25through via Slack and sending it through
7:27to the relevant user and the relevant uh
7:30Slack channel. Both options, whether
7:32it's email or not, are are fine um and
7:36work really well. So, we'll share the
7:39the the JSON template for this. Um but I
7:43think you should be able to get started
7:44and build it uh relatively
7:45straightforward and simple.