YT YouTube2Text

YouTube2Text → Transcripts → Marketing Agents Masterclass (GROW your startup)

Marketing Agents Masterclass (GROW your startup) — Transcript

by Greg Isenberg · 8,932 words · 1,258 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00It's true. Marketing agents are the new
0:02coding agents. Just like coding agents
0:05were such a big deal and people are able
0:07to create software on demand, deploying
0:09marketing agents are so important
0:12because you're able to get customers on
0:14autopilot. [music]
0:15So, how do you actually set them up?
0:18What do they look like? Well, this has
0:20got to be my most requested episode in a
0:22long time. I bring back Cody Schneider
0:24and he shares all the sauce how you can
0:27use codeex or claude code to to build
0:29these. What are the other 20 tools that
0:32you need for the marketing
0:33infrastructure in order to deploy these
0:35marketing agents? And by the end of this
0:37episode, you're going to get your
0:39creative juices flowing around some of
0:40these growth tactics that are going to
0:42help you stand out, that are going to
0:44help you get customers, so that whatever
0:46it is you're building, you don't have to
0:47worry too much about traffic, you don't
0:49have to worry about too much about
0:50revenue, and you can focus on building
0:53an incredible product uh while your
0:55marketing machine is [music] running.
0:58Enjoy the episode.
1:02[music]
1:07Welcome to Greg Eisingberg's podcast
1:09called Sip Baby. I'm your co-host or
1:12guest today. Not co-host. I'm never the
1:14co-host. I'm Cody Schneider. I'm going
1:16to be your guest today. And today I'm
1:18going to teach you how to build an AI
1:20agent that does cold outbound both on
1:23email and on LinkedIn. This is based off
1:25of the comments from last video. If you
1:27want to learn other goto market motions,
1:30you need to comment below right now. Do
1:31it right now. It also helps us for the
1:33algorithm, so you're supporting the show
1:35and it keeps the lights on here.
1:37>> Welcome to the show, Cody. Marketing
1:39agents are the new coding agents. We
1:42only shared one marketing agent last
1:45episode, but the people aren't satisfied
1:47with one. So, you needed to come back
1:48on. You came back quickly. And by the
1:52end of this episode, you're not going to
1:53share one endto-end marketing agent,
1:56right? You're going to share two
1:58marketing agents, how people could set
2:00it up. So by the end of this episode,
2:02people can go stop the video and
2:06actually go set this up and actually get
2:08customers to their vibe coded startup.
2:10Right? This is exactly what I'm
2:12promising you today. You're going to
2:14have two of these in the wild. I'm going
2:15to teach you everything that you need to
2:16know. I'm also going to share all of the
2:18tools that you need. There's no
2:19gatekeeping here. I despise people that
2:21do this. Don't buy a course. Literally
2:23DM me. I'll teach you anything. I'll
2:25just make a public video for everybody.
2:27So let's do it, G.
2:28>> All right. Let's run it. Awesome, man.
2:31All right, so today we're going to build
2:33a system that basically monitors
2:36LinkedIn posts of influencers within
2:39your niche, within your category, and
2:41then it's going to go and extract the
2:43engagers from those posts. Um, and then
2:46we're going to do what's called a
2:47waterfall enrichment to find the emails,
2:50uh, and even potentially the phone
2:52numbers of these people so that you can
2:54then go and do an outbound motion to
2:56them. um doing cold email and then also
2:58uh doing LinkedIn DMs. So that's what
3:00you're is going to happen and then I'm
3:02going to uh teach you how to basically
3:03have it so you can have a agent that's
3:05wired up to both of those inboxes like
3:07managing those inboxes say for example
3:09answering questions or trying to push
3:11them into like booking a demo with you
3:12as an example. So uh yeah man that's
3:15that's really it. Uh the I don't know if
3:17there's any other like specifications on
3:19the high level. I think the only thing
3:21to mention with this is like the
3:22strategy around this. So right now cold
3:25email is getting decimated. Reply rates
3:28are down. Everything is down. Actually
3:30every marketing channel is down right
3:31now. Let's be re let's be real. Uh the
3:34reason is just because like AI slop is
3:37flooding the zone and it's becoming just
3:39red ocean everywhere. Um but the way
3:41that we have found that you can stand
3:43out is you have to look for signals or
3:45triggers that basically show that people
3:47are hand raising um saying hey I want I
3:50want this thing. I have an interest in
3:51this thing. Right? And a great way to do
3:53this is with these LinkedIn uh uh
3:56engagements. Uh they're basically when
3:58they like content that is a a hand raise
4:01or a signal that I am interested in this
4:03you know uh specific thing and from that
4:06we can use that as a way to measure okay
4:08is this my target customer that I'm
4:10trying to sell to and not just like
4:12their firmographics or their
4:14demographics or their psychoraphics
4:15which is like what we would
4:16traditionally use for for outbound. um
4:19this is specifically like no they they
4:21have a propensity or an interest in this
4:23topic and we are going to go and now get
4:26in front of them. Okay. So how do we
4:27actually do this and this is an exact
4:29strategy that we implement for you know
4:31the companies that we're working with.
4:32So I'm going to teach you that right
4:33now. So let me screen share and I'm
4:35going to walk through it. So the first
4:37thing uh that you're going to want to go
4:39to do is literally go to LinkedIn and
4:42find uh influencers within your
4:44category. So, last episode we talked
4:46about AI for WordPress or AI WordPress.
4:49And so, I'm just going to use this again
4:51as an uh the example, you know, like
4:54target demographic that we're going
4:55after. Um, so on LinkedIn, what I would
4:58go do is I would go try and find people
5:00that are talking about WordPress uh
5:02development potentially. Um, let's see
5:05what comes up with that uh development.
5:11And I would try to find posts. This
5:13might actually be a terrible category.
5:15So I we might have to explore something
5:17entirely different, but um I would try
5:19to find posts or creators that are
5:21talking about uh these specific topics
5:24like on a daily cadence, right? Um so
5:27like this like again just for this
5:29example today, this is probably going to
5:31be like a lot of like just not good
5:34signal. So a better way to look at this
5:36is like uh we'll say we'll do AI for or
5:39AI marketing, right?
5:41Um, let's see what comes up and we're
5:43going to try these find these posts
5:45here. So,
5:46>> and what makes a good search? Like why
5:48why was AI for WordPress not good and
5:50why is AI marketing better?
5:52>> Yeah. So, [clears throat] it's really
5:54just like is the content that's being
5:57served what your target customer would
5:59be interacting with? Like that's what
6:01you're trying to get down to here,
6:02right? So like how I would be going
6:04through this and honestly I use the the
6:06for you page of all these algorithms is
6:08so good now that it's like it's going to
6:10show you the content that's relevant
6:12right like this is literally an exact
6:14[laughter]
6:15like perfect like perfect example first
6:17one that comes off it's like awesome
6:19people trying to do some type of video
6:21editing for obvious it's probably for
6:23marketing everybody that's potentially
6:24engaging with this is like a target
6:26customer right so I would say okay cool
6:29I'm going to find these creators and
6:31then I'm going to build a spreadsheet
6:32sheet of all of them, right? Like all of
6:34these people um that I'm going to try uh
6:37that I'm going to source these leads
6:39from. So, right, I would build this
6:41spreadsheet out and we'll just do a
6:43handful of these like from my own feed.
6:46It can even be business accounts and I
6:47think this is the thing that people
6:48don't realize like if there's business
6:50accounts um that the people would be
6:53interacting with that would be your
6:54target customer, that can work as well,
6:55right? So, it can be literally Clay. Um,
6:58and we're going to do the posts from
7:00Clay. And, uh, we'll just keep going
7:03down on this. So, MCP, it's probably too
7:06broad.
7:07>> And you're doing this manually. Like,
7:08you're not using agents to do this. Why?
7:10>> I I wouldn't even typically the company
7:13knows who is interacting. Like when
7:15we're working with a business, right?
7:17They know who their like their target
7:20customer is interacting with, right? So
7:23you can all you need is typically like
7:2510 10 to 20 of these and you have more
7:27than enough to be able to like source
7:30the lead volume that's necessary to
7:31actually make this like a viable
7:33channel. Um you I'm using the feed here
7:36because like what it's going to show you
7:38is what is most like relevant to you. So
7:42it's probably going to be stuff that's
7:44uh you know in your the niche that
7:46you're in. But you can also use the
7:48search for this as well. We used to do
7:50this where we'd like do the search and
7:52we'd find the trending posts from that
7:53period. In reality, it's like there's a
7:55handful of outliers within any niche and
7:58everybody is engaging with those handful
8:00of outliers. If you just monitor those
8:02outliers, you're actually going to get,
8:04you know, 80% surface area coverage for
8:07that entire industry. You don't need
8:09more than that, right? Uh or or it's
8:11it's it it's just like the marginal
8:13return of trying to go for all of it.
8:15It's not it's not there for for that
8:17system. This is the same idea with um we
8:19do this a lot like we try to solve
8:20entropy pro this entropy problem with uh
8:23within like ads, paid ads in particular
8:26where it's like if you just have the
8:27agent like go in this loop, it'll just
8:28kind of make the same ideas over and
8:29over again. How do you solve for that?
8:31Well, you find human creators like 10 of
8:33them on Instagram and you track the
8:36content that they're they're publishing.
8:38You look for the outliers and then from
8:39that you you typically can get signal of
8:41like, oh, here's this new hook format or
8:43here's this new topic and I can just
8:45pull that. I can remix that and that's
8:47the way to do this. So, all right, I
8:49find a handful of these these these
8:52companies. Um, and then from that what
8:54I'll go and do and just for the sake of
8:57uh uh you know the example today, we'll
9:01use Louise as an example. So, we'll say
9:03everybody that interacted with this
9:06post. We're going to use this post as an
9:07example. So um once I have these people
9:11I need to use ampify and I will find the
9:15actual one that we like.
9:17>> And what's ampify for people who don't
9:18know?
9:19>> Yeah. So ampify is a scraping API. So I
9:22can use a single API key and then I can
9:25use it to scrape LinkedIn. Um I can use
9:28it to scrape uh uh Twitter. I can use it
9:30to scrape all of these different
9:31channels. So, it's a way for me to get
9:33data into the context for my agent so
9:36that it can have, you know, awareness uh
9:38and and have that context for it to make
9:40decisions on or make content based off
9:42of, etc. So, okay. So, the one that
9:44you're going to want to use or the one
9:46that we like, we've worked with him like
9:48a decent amount because it's the most
9:50stable connections. There's tons of
9:52these and the challenge with Ampify is
9:53finding good ones that are actually um
9:56uh like being monitored and being
9:58maintained. And so this uh uh this guy
10:01API maestro has a ton of these for
10:03LinkedIn. You can see all of these here.
10:05It's all of these different functions
10:06that you can do. So how appy functions
10:09is you get an API from ampify and then
10:12this enables for you to be able to have
10:15your coding agent like cloud code or
10:17codeex call from uh ai or call the app
10:22through the appy to one of these
10:24endpoints that are here. So, uh, for
10:26example, you can do this post scraper.
10:28Uh, for the one that we're going to do,
10:29it's going to be engagements. So, let me
10:32find that. Uh,
10:35post reactions on LinkedIn. I believe
10:38this is it. This is exactly it. Yep. So,
10:41post comments and then post reactions
10:43are the two that you're going to use.
10:45And what this enables you to do is
10:47everybody that has engaged with that
10:49post. So, the post that we are just
10:51looking at here. So, everybody that's
10:52interacted with this and commented on
10:54this, we're going to be able to pull
10:55this out. And I'm going to show you how
10:56you can actually do this in uh Cloud
10:58Code right now. So, I'm just going to
11:00spin up a terminal real quick. And let
11:02me reshare my screen. And so, I have
11:05that uh I have that Appify API key um in
11:10uh already saved locally within the
11:12directory that I work out of for uh all
11:16of my growth work. And if you don't know
11:17what I'm talking about here, I have a
11:19whole video on my uh channel that's
11:21basically a crash course into how to do
11:23this. It's called go to market
11:24engineering or marketing engineering. It
11:26will walk through the entire setup
11:27process. Takes about 10 minutes. But
11:29basically, this ampify API key is shared
11:31here. And I've already written this
11:33script. I had the agent go and read how
11:35do I use this endpoint to pull out all
11:38of the post and comments information,
11:40the all the people that have interacted
11:42with this. So I can give it this post
11:44URL and I can say extract the engagers
11:49using the ampify API key
11:53and it's going to go and run that
11:55process for me. So this is how I would
11:57go and build this automation or build
11:58this agent as I would basically take
12:00this code and I would deploy it into the
12:02cloud and I would say okay on a daily
12:04cadence I want you to check for net new
12:07posts. So that is where I would look at
12:10the profile posts. So this is the
12:13profile post scraper. So I would extract
12:15the post urls from this person,
12:18right? So every net new post daily is
12:20getting extracted and then from that I'm
12:22then extracting the engagers
12:26using that API endpoint as well. Right?
12:29So right now as you can see the duped by
12:31public profiles there's 63 raw and it's
12:34about to pull all of those contacts out.
12:37So once I have those contacts this is
12:39this is done man like game over.
12:42As long as you have the LinkedIn
12:43profiles, you can go and find the email
12:45addresses of them. You can find the
12:47phone numbers of them. You can find
12:49everything that you need on the cold
12:50outbound. And I'm going to show you that
12:52right now. What are the tools to
12:53actually go and use to do this? Um, so
12:56let me just show you though again just a
12:58uh the final completion of this. And
13:00what makes this an agent versus a
13:03marketing automation?
13:05>> Yeah. So the agent component of this is
13:08that it is running on a cron job daily
13:11and then you're going to have an agent
13:13that's later on we'll have it responding
13:15to the inbox and this is this blurry
13:18line right like what is an agent people
13:20ask me this every sales call and the
13:23answer to all of this is like it's how I
13:26think about it personally is it's
13:28something that's doing a job to be done
13:30right so the job to be done here is
13:32finding leads and outbounding to those
13:35leads and then responding to those leads
13:39as they're like asking questions or
13:41again like driving them deeper into the
13:42pipeline. Um in reality though g like
13:45what is a market like what is a
13:46marketing agent? It's it's code. It's
13:48maybe some thinking loop and it's a live
13:50data stream, right? That that is really
13:52how like this functions. And the thing
13:54that you can make, you know, extend this
13:56further with is like what you're who
13:58you're outbounding to. Um you want it to
14:01basically do an ICP fits or is or or a
14:04target customer segment fit. So before
14:06it even does this enrichment that we're
14:08about to do, you would be like, "Okay,
14:10agent, research this person and the
14:13company that they're at. How many
14:14employees do they have? All of these
14:15things." And then based off of what we
14:19find, if it fits this customer profile,
14:22like it's you're going to have the agent
14:23basically think through that, right?
14:24Using an LLM, if it fits this customer
14:27profile, then it goes into this
14:29enrichment. Then we're actually going to
14:30cold email them. So that's where that
14:32thinking loop could potentially be here
14:33as well. But really the the blurriness
14:36between all this I think about it as
14:37software anymore like to be transparent
14:40like everybody the thing a different way
14:42to say this is like everybody tried to
14:44put God in a box and give it access to a
14:45Facebook ads account and we realized
14:47that is not the right way to do this
14:49whatsoever. The right way to do this is
14:51like what was the human doing? They were
14:53running this very specific process with
14:55like media buying. They were researching
14:57ad creative angles. They were making new
14:59ad creative. They were testing the new
15:01ad creative and then they were like
15:03pruning the losers, promoting the
15:05winners, right? Like that is what the a
15:07top media buyer does. Okay, how do we go
15:08and make a piece of software that does
15:11that exact same thing? So when you hear
15:13agents like really just think software
15:15with potentially a thinking loop like
15:17you shouldn't be paying a different like
15:19way to think about this and this is
15:20something I'm obsessed with right now.
15:21You should not be paying anthropic. You
15:24should not be paying Chad GPT to do an
15:26API call. You should be paying them to
15:29make the software that uses CPU to do
15:32the API call. Why are you paying this
15:34tax on tokens every time that you're
15:36trying to do this marketing activity?
15:37That's ridiculous. Build the software
15:39that does the solution for you, not
15:41tokens burning every time that you're
15:42trying to do the action. So anyway, um,
15:46okay. So we've got these LinkedIn URLs
15:47and what do we do with them now? So
15:49we're going to do what's called a
15:50waterfall enrichment. And so we're
15:51basically going to use these LinkedIn
15:53profiles to go and find the email
15:55addresses and then the phone numbers of
15:56these individuals. So how do we do this?
15:58The first thing that we're going to use
15:59in a tool stack is called getleads.io.
16:02Um so this is a database of uh it's
16:06basically they aggregate all these B2B
16:08contacts and you can access it via their
16:11API.
16:12um the emails that we don't find within
16:15git leads, we're then going to use
16:17something or we're then going to
16:18waterfall down to something like Apollo.
16:20Um and then you could take this even
16:22further down into something like
16:23Origami. It's another tool that we have
16:25been using and experimenting with. Also,
16:27their team is just doing awesome work.
16:28Like Finn and his whole team is
16:30incredible. So anyways, for git leads,
16:32let's go back to our uh u uh our
16:35terminal right now. So again, this is me
16:38hands on keyboard doing the process to
16:40teach it to you. But everything that I'm
16:42doing right now, this is all just going
16:43to be code under the hood. And once it's
16:45code, I can deploy that into a cloud
16:48system. As long as it has the necessary
16:50data that it needs and the necessary
16:52access that it needs, it can go and run
16:54this operation autonomously. And then
16:56you're just there basically jockeying
16:58the agent or modifying the system.
17:00Right? So we're building a system here.
17:01So from here um what I would then go do
17:04is say use the get leads API
17:08uh to uh find the emails and phone
17:12numbers
17:14>> and like dumb question.
17:15>> Yeah,
17:16>> that's legit like f you know like
17:19it's not gray to get these people's
17:21emails. It's like fully legit.
17:24>> It is fully legit to get these emails.
17:26um what you do with those that's where
17:28things uh like from a compliance
17:30standpoint change. You can cold email
17:32technically in the United States. You
17:34can also add people to a email
17:37newsletter um to be and be can spam
17:40compliant. There's like tons of you like
17:43things you basically have a checklist of
17:45things that you have to do. With this
17:46said though, um like this is one of
17:49these like on the cold email side and
17:51the contact lookup. Um you're basically
17:53just buying data from a data broker
17:54which is is legal, right? That that is
17:57accessible. So these companies how they
17:59do this is they basically are buying all
18:01these lists and then aggregating them
18:03from all these different data brokers.
18:05That whole piece is it's a whole other
18:07shady network. But this uh like what
18:09we're talking about here, you know, on
18:11the spectrum of like black hat to white
18:13hat is pretty far on that white hat
18:14side. So
18:16>> cool.
18:16>> Yeah. I mean, I don't think anyone
18:18would, you know, mistake you for a
18:20lawyer also.
18:21>> Oh, totally. Take this with a grain of
18:23salt, you know, and and like there's
18:25also different compliance rules within
18:27the United States.
18:28>> Your own research.
18:29>> Exactly. Exactly. Within, you know, the
18:32United States versus uh like the EU has
18:34totally different compliance pieces.
18:36>> Exactly. Um but with that said like the
18:40uh you know the finding of people's
18:41information and then like reaching out
18:43to them uh there you c you can do this
18:46basically is kind of the high level but
18:48again this I I we don't have time today
18:50to go into all the the specifics about
18:52like all this the finite details here.
18:55So once I found this um each of these
18:58individuals and then the emails um from
19:00there what I'm then going to do is
19:03validate these emails. So I would send
19:05it to a software called millionverifier.
19:08So million verifier um enables me to uh
19:11basically check if the email is good,
19:14risky or bad. Um you know more technical
19:17terms would be uh like good, catchall,
19:19um you know risky etc. Um the the
19:22reasoning for this or the reason you
19:24have you want to do this is the emails
19:26that come out of these providers. So out
19:28of git leads, out of Apollo,
19:32out of Origami. I think they do some
19:33checks like a little bit deeper though.
19:35So you I don't know much as much about
19:37this, but I know for sure with get leads
19:39in Apollo, it's like do the second
19:41verification. You're basically only
19:43wanting to send cold email to valid
19:46emails because if you send to invalid
19:49emails, you're going to basically just
19:52run into deliverability problems. And
19:54probably right now you're asking
19:56yourself like, "Okay, cool. Well, how do
19:57you send these cold emails? I'm going to
19:58show you that in a second, so bear with
20:00me. So, we've done that waterfall
20:02enrichment. We found the emails. We
20:04found the phone numbers. And when I say
20:06a waterfall enrichment, what is
20:07happening here is we're taking that list
20:09of 50. And just to use this spreadsheet
20:12as an example, so say we have, you know,
20:1450 that we have uh 50 LinkedIn URLs that
20:18we found and on git leads, maybe we only
20:22find, you know, 32 emails
20:25of those people, right?
20:27So that next cohort, so those other 18
20:30that are left, I'm then going to send
20:32those 18 to Apollo. So of those 18 that
20:36I send, maybe I only find 10.
20:40And then those eight, that's when I
20:41would send that to something else like
20:42Prospio or Origami or these other
20:45enrichment tools. And the reasoning
20:46behind this is you're you're starting
20:48with what is the cheapest, most accurate
20:50and then moving your way down into the
20:53more expensive uh uh validation tools.
20:56Um but from this you can pull out
20:59basically from a list like you know this
21:00is the way that you get to uh you know
21:03an 80% fine rate etc. And you can chain
21:05as many of these together as you want.
21:07Um it just you know depends on your
21:09budgets that are available etc. There's
21:11also aggregators of this like Origami as
21:13an example like aggregates this
21:15waterfall for you. So you can just send
21:16them a LinkedIn profile and it's going
21:17to like waterfall through the options
21:19that are available. Um, okay. So the
21:22other other thing to throw in here that
21:24will be valuable to your team is a
21:26software called Lead Magic. So this is
21:28one that I we use a lot for like mobile
21:30phones in particular. Um, but same
21:32strategy here. Uh, it's just basically,
21:34you know, another enrichment tool. Uh,
21:36but specifically on the phone number
21:37side, we we've used it a decent amount.
21:40So once I have that contact information,
21:43I now need to go and actually build this
21:44outbound motion. So on the cold email
21:46side first, how do we go and do this? Uh
21:49we need to buy inboxes. So a couple
21:51different ways to do that. I can use a
21:53tool called inbox kit. I can use
21:55instantly AI's pre-built uh uh like
21:59emails that you can buy from them. Um or
22:02I can use uh a company called Hypertide,
22:05which is the partner that we use and we
22:07work with. they are some of the best
22:09info in my opinion. So when you're
22:12buying these emails, uh you're buying or
22:15you're really what you're doing is
22:16you're buying inboxes and domains that
22:18are burner domains that enable you to
22:22send cold email
22:24um not from your core domain. And the
22:26reason that you have to do this is so
22:28that you don't burn the deliverability
22:30of your core domain. So what do I mean
22:32by that? If you send from you know your
22:34exact domain um and uh you know say we
22:39send 10,000 cold emails from that um we
22:42will nuke the deliverability of the
22:45business URL the actual domain that we
22:47use to you know run our company right
22:50you don't want to do that so typically
22:52what you want to do on the marketing
22:53side is have this se have this
22:54separation so you have domains that are
22:56for your cold email you have domains
22:58that are for your email marketing you
23:00have domains that are for your
23:01transactional marketing so This would be
23:03or transactional email. So this would be
23:05email that's being sent directly from
23:07the product to a customer. Imagine like
23:09a password reset as an example. And then
23:11you want to have your business you know
23:13domain email which is what your team
23:15actually uses to run the company etc. Um
23:18so with Hypertide as an example um we we
23:21have a partnership with them. So it's a
23:23little bit different but uh we can send
23:25about 10,000 cold emails. Uh just to
23:27give a a um you know kind of the cost
23:30breakdown here. We can send about 10,000
23:31cold emails with them for about $100 a
23:33month in infrastructure costs on the
23:35inbox side. Um it's about the same for
23:38majority of these. So inbox kit as an
23:40example is very similar pricing. They
23:42also run like sales all the time. So
23:44look for those on the domain side. So
23:46you basically buy the domains and then
23:47you're paying a subscription to have
23:49these inboxes hosted for you. And then
23:51on instantly side u you can typically
23:54get started with this $97 a month tier.
23:57in total, you know, out the door to get
23:58going on this, the infrastructure cost
24:00can be in that range of about $100 to
24:03get started or sorry, about $200 to get
24:05started for the sending uh software and
24:07then also the inboxes. So again, just to
24:09reiterate this because I know I've
24:11talked through a lot, I'm pulling the
24:13lead list from LinkedIn. I'm finding
24:16these people. How do I know that these
24:18are people that I want to reach out to?
24:19It's because they're engaging with
24:20content that I know my target customer
24:22would be interested in. And so these
24:25people are basically hand raising that
24:26they are would potentially be my target
24:28customer,
24:30>> which is insane by the way.
24:32>> Right. Which is insane to
24:34>> find this, right? [laughter]
24:36>> Yeah.
24:36>> Yeah.
24:37>> It's impossible to find this. Um and so
24:40the uh so I'm finding these people. I'm
24:44then doing a waterfall enrichment to
24:46find all of their uh contact
24:48information.
24:50And then once I have their contact
24:51information, I need to actually be able
24:54to send to them. So I'm getting inbox
24:55infrastructure to be able to send. And
24:58then I'm sending with a platform like
24:59instantly. And then on the LinkedIn DM
25:02side, what I'm sending with is a
25:04platform um like hey reach.
25:08Another one that we like is called Bot
25:10Dog.
25:12Um both of these have APIs. Um, but what
25:14these enable you to do is basically uh
25:17do uh LinkedIn DM campaigns um from
25:21these accounts. I also know people that
25:22are just like using LinkedIn DM or sorry
25:26LinkedIn inmail for this and seeing
25:28incredible success right now uh using
25:30this strategy. So again just throwing
25:31out all the strategies that are
25:32available. Um so this is how you can
25:36build this pipeline right now. How do
25:39you actually like have an agent that is
25:41managing that inbox? So looking at
25:43instantly as an example, they have an
25:45API
25:48and that API
25:50allows for you to monitor and manage the
25:53entire account. So you can have an agent
25:55that's literally writing copy for each
25:58individual email or person that you're
26:00contacting or reaching out to um and
26:03writing those variables and then that
26:05can be basically pushed into instantly.
26:07So this happens outside the platform
26:08gets pushed in. But the bigger thing
26:10here is they also have web hooks. So
26:11when a positive reply happens, you can
26:14send that web hook confirmation back to
26:16your agent that's hosted on some type of
26:18cloud server and that agent you give it
26:21basically um like a base prompt, right,
26:23of like you're the goal like here's all
26:25the context that you need and your goal
26:26is to try to get people to schedule
26:28demos on you know this this link, right?
26:30It can manage that inbox, answer
26:32questions, push people deeper. But the
26:34thing that gets really fascinating and
26:35really powerful with this G is like it
26:38can do these follow-ups like months
26:41later, right? So it's like okay like
26:43also like every six months, right? I
26:45want to pro it grow program that in to
26:48like re reereach out to these people
26:50that went cold. I can also plug it into
26:52my scheduling application like Kalanley
26:53or like Cal.com.
26:55I can give the agent access to see okay
26:58did this person that we reached out to
27:01can we did they actually schedule a
27:03discovery call did they actually you
27:05know produce the action that we're or
27:08you know make the action that we're
27:09trying to optimize for and so from this
27:12you can basically build this like SDR in
27:15a box right that is again finding new
27:18people for you based off of the
27:20engagements that they're interacting
27:21with on social finding the emails
27:24actually writing the emails, deciding if
27:26this is a good ICP fit, and then sending
27:29that to these sending platforms and then
27:31managing the inboxes of those sending
27:33platforms. And again, when I say agent,
27:35right, like when I'm saying, oh, it's
27:36managing this inbox, it's literally just
27:39code under the hood, right? It's code
27:41under the hood with an LLM attached.
27:42That is an agent. Like in this context
27:45here, you don't have to over complicate
27:46this. You don't have to have God in a
27:48box managing an email inbox. Be a very
27:51simple setup to actually produce this. I
27:53also get asked this question a lot like
27:54do you need use like some agent
27:56framework under the hood it's like a lot
27:57of the times you don't need it it's just
27:58bloat you can just have a very simple
28:01like a very simple solution for these
28:03finite problems right it doesn't have to
28:05be this over complicated or
28:06overengineered thing so anyways happy to
28:08answer any questions about that or dive
28:10deeper on any of this again it's hard to
28:12show code so I I didn't really do that
28:14today of like this is how you do it but
28:16what you need here basically the final
28:18piece is you need to set up a server so
28:20use something like a railway or this is
28:21what we do at like graft right? Is like
28:23we have the data pipeline warehouse and
28:26then the server to deploy these agents
28:27to that's like off of the live data
28:29streams. But yeah, happy to answer
28:30questions. D
28:31>> I mean to be clear, you're you know
28:33you're using a harness like Cloud Code
28:35or Codeex to actually build out all of
28:39the thing. But this the hard part is the
28:42strategy around you know who you're
28:45going after, why you're going after
28:46them, what's your tool stack that like
28:49what's amazing is you just like outlined
28:51here's all the tools that you need to
28:53get like set up then it becomes okay I
28:57have to go into you know that's what
29:00people are talking about software
29:01factories like we're all in the software
29:03factory business now right because we're
29:06just going and we're spitting up stuff
29:08like this the software to actually go
29:10and complete these tasks.
29:12>> Absolutely. I I think the thing that we
29:15are like focusing on like so to say like
29:19a good way to think about this is like
29:20if you can build it in cloud code and
29:22like have some type of local system that
29:24you're running, you can probably deploy
29:26that to a server somewhere, right? And
29:29have that run on an hourly cadence or a
29:31daily cadence or whatever that ends up
29:33looking like. The challenge ends up
29:35being how do I set up the infrastructure
29:37that's necessary for the agent to be
29:39able to do this right and the the the
29:41solution is like the open source
29:43solution as an example like we talked
29:44about this on the last call use
29:45something like a uh with click house to
29:48get like create your data pipeline and
29:50your data warehouse so you have that
29:51data stream for the agent to make those
29:53decisions and then you have to have some
29:54server and like when I say server what
29:56do what is that right for the
29:57uninitiated it's just a computer that is
30:00on [laughter]
30:02all the time somewhere else that you're
30:04putting code onto, right? I think this
30:06software factory thing is super
30:07fascinating as well. Like like really
30:10it's funny. This is how I'm thinking
30:11about marketing now. Like marketing is
30:13just code like a like when I generate
30:16[laughter] a banana image. Like that's
30:18just a JSON prompt under the hood. Like
30:20when I make like you know seed dance AI
30:23avatar videos that's just like an LLM
30:26that like scraped Reddit like read some
30:29things wrote a script and then we it's
30:31just an API call that's happening to Kai
30:34AI to generate that image with like okay
30:36here's how you chain this together to
30:37make it into 30 seconds every everything
30:39now like in and my co-founder this is
30:42his firm belief like Max always says
30:43this he's basically like the only agent
30:46is a coding agent actually [laughter]
30:48everything else is this software that's
30:51being made by the coding age and I think
30:52this is like this paradigm shift and
30:54like something that we are obsessed with
30:55like why are you paying tokens for
30:57things that can be just code that is
30:59running on super cheap compute you don't
31:01you don't have to have like inference
31:03every time that you're doing this action
31:05only use inference when you need it and
31:07this is kind of this like differing
31:08viewpoint that I think you know
31:11everybody's just like oh token abundance
31:12I'm going to token max I'm like I'm
31:14actually totally like probably the
31:15opposite of that like why it just feel
31:17it is wasteful like do the thing that is
31:20the simpler thing that has less
31:21likelihood of breaking. Like if you have
31:23Hermes try to run your Facebook ads,
31:24high likelihood it might just like
31:26absolutely nuke the account, but if you
31:27have it run based off you you build a
31:30piece of custom software for yourself
31:32that's running based off of a system
31:34that a normal human like a real human
31:36would run totally different, you know,
31:37outcomes that you're going to get from
31:39that that are probably higher quality.
31:40So,
31:41>> okay. Do we have time for a second
31:44marketing agent demo flow? Yeah, I can
31:48talk through um I just did this for
31:52[laughter]
31:54um I just did this for my team. Um I I
31:57don't know if that'll be super
31:58interesting actually. I mean you tell me
32:00we basically we're like okay how do we
32:02at scale make social content on LinkedIn
32:05for like the entire team and like so we
32:08have them like basically we're
32:09interviewing them we take the
32:11transcripts we pull out the insights the
32:13insights get written into the posts the
32:15posts automatically get scheduled to
32:16their LinkedIn accounts using a tool
32:18called ordinal uh MCP
32:20>> yes stop like yes this is interesting
32:22because a lot of people I mean a lot of
32:25people might have heard you know listen
32:27to this cold cold email approach
32:29approach or cold reachout approach and
32:31are like m I want to go the organic
32:34route. So like what's what's an example
32:36of setting up a marketing agent in an
32:37organic route and and can you break that
32:39down for us?
32:40>> Absolutely. Yeah, I'll do the LinkedIn
32:41one because it's super topical and like
32:43we've had a lot of interest in this
32:44lately by companies which has been
32:46pretty fascinating. They're using this
32:47with like their sales teams like they
32:50want, you know, their seven person sales
32:51team to be posting daily. How do they
32:53actually do that and make unique ideas?
32:55So um this also pairs with the cold
32:58email. I'll talk about that as well. Um,
33:00but yeah, just to run through the
33:02process. Uh, super simple. Um, it's like
33:04literally record a conversation like
33:07this. Like I have a a a weekly call like
33:10one-on-one with like the people that
33:12we're doing this for in the or just like
33:14tell me everything that like you've
33:17learned in the last week. I just
33:18basically interview them, have a
33:19conversation, right? It doesn't have to
33:21be anything like you don't have to have
33:23any focus. It's just like what are the
33:25things that that jumped out at you after
33:27being in these sales calls or whatever
33:29your job is. You can do this for like
33:30technical people as well at the
33:31organization. You can do this for
33:32everybody. And I imagine this is how the
33:34large like the real companies are doing
33:36this. There's no way that like everybody
33:38at like a lovable [laughter] is writing
33:40the content that's going out across all
33:42of the accounts. Maybe that's happening.
33:43But um I think what's more likely is
33:46that there's somebody behind the scenes
33:47that's orchestrating this. It also
33:49doesn't have to be an interview. It can
33:50just be sales calls or internal comms.
33:52Like Alex Lieberman as an example has
33:54been talking about about this a lot
33:56where they're they're basically sourcing
33:58like so much context is happening within
34:00their notion within their codebase
34:02within their their Slack. We see this as
34:04well, right? You can use one of these
34:06agents to query those data sources,
34:10right? Like query the sales channel um
34:13or query the gong transcripts and that's
34:15where you can pull these insights from.
34:16And honestly, a lot of the times you
34:18find that it's really inside like it's
34:20really good content that's trapped in
34:22there. Like these ideas like for
34:23example, a customer had uh you know, a
34:25customer said that or a potential
34:26customer said this and it was like why
34:29they didn't buy the product and that can
34:31turn into an unbelievable piece of
34:33content um that you can extract from. So
34:35you get source material. Why do you have
34:37to get source material? The reason is
34:39because if you go and you try to just
34:41have the agent like think about this,
34:44you're like, "Write good LinkedIn
34:45content." [laughter] It's going to be
34:47the most mid thing you I mean, it's
34:49you're going to waste the person's time
34:50on the other side, right? Um or you're
34:52going to get flagged for AI slot by
34:54LinkedIn's new feature that just
34:55released this morning. Um the the better
34:58way to do this is source this from real
35:00human conversation because that's where
35:01these original ideas are coming from.
35:04Um, another example of this is like
35:06literally this podcast. You could
35:08extract all the insights from the
35:10transcript and that can be used as
35:11social content. This is like a strategy
35:13I use for myself. But it doesn't have to
35:15be just your own. It can be somebody
35:17else's as well. It can be, you know, a
35:19podcast with Naval. It can be whatever.
35:21It can the source material can be
35:22anything. But the system that you create
35:24is some type of source material that's
35:25happening on, you know, some type of
35:26cadence. And then from that, I'm I'm
35:29building basically this writing and
35:30scheduling process. So, what I I'll walk
35:32through now how to actually like do
35:34this. Um, so take that source material.
35:37You're going to do an API call um into
35:41uh you know some LLM as an example uh
35:44for this. Like you could I mean we've
35:46even used just like uh Claude Sonnet as
35:49an example and it's probably good enough
35:50on the writing side. Um and then once
35:53you have that those written posts,
35:55you're then going to go and use
35:57scheduling tool. We like Ordinal for
35:59this. um they're a partner of ours as
36:01well. Um but it allows for you to have
36:03multiple LinkedIn accounts connected to
36:05it and then they can also interact with
36:07each other which is amazing. Um but you
36:10can through their API or their MCP
36:13schedule these posts to each of the
36:15individual accounts
36:17and then Ordinal also has and I could
36:20just go into this actually show you um
36:22Ordinal also has uh the uh analytics
36:25data that pulls in from your LinkedIn
36:27post there as well. So we can see the
36:29breakdown of like which content is
36:31actually performing well. So it has the
36:33analytics of the multiple accounts. You
36:34can actually see the breakdown of the
36:36individual posts and that data stream
36:38can go back to the agent so that it
36:40understands okay this is what's getting
36:42impressions. This is what's doing well.
36:45Let's go do more content like when it
36:47does its cycles of writing that can
36:49influence the next round of creative. So
36:51topics like this perform better based
36:53off of the source material we pulled.
36:55How can we snowball or remix? use those
36:57specific words snowball or remix to have
37:01it go further, right? And this is where
37:02the LLM is thinking on top of that data
37:04stream. And when you look at like what
37:05is happening here, like what does the
37:07social media manager do? I actually
37:09think the social media manager job like
37:11full stop. It's it's [laughter]
37:13I think it's already dead, but let's
37:16won't get into that. If you're listening
37:17to this, please learn how to make and
37:19manage content at scale across multiple
37:21accounts. um it's with agents cuz that's
37:24going to be I think that's the real meta
37:25now is like how can a single person
37:28manage you know 10 20 100 accounts
37:31across all of these different channels.
37:33Um but when you look at what a social
37:34media manager did previously like a good
37:36one that was actually excellent
37:38excellent at their job is they would
37:40prospect for ideas. They would make
37:42content about those ideas. They would
37:44publish it. They would look at the data
37:46to see which got the most impressions
37:49and then they would turn that into a a
37:50recurring content calendar where they're
37:52like, "Okay, I'm just remixing this
37:54these same ideas over and over again."
37:56If you look at my Twitter like post as
37:58an example or even my LinkedIn, it is
38:00the exact same thing remixed every 90
38:03days like full stop. That is all that's
38:06happening. And that when you get enough
38:09information like a big enough corpus,
38:11you have you basically understand what's
38:13already going to go viral. Like I I have
38:15these posts that I've literally used for
38:16the last two years. Every time I post
38:18it, I know it's going to go viral. I
38:19can't post it every day. You post it
38:20every 90 days, right? And that's how you
38:22can go back into this cadence. And so
38:24again, have this mentality of I'm
38:26prospecting for ideas. I'm prospecting
38:28for winners. Once I find those, I'm
38:29trying to use those as as often as I can
38:33because I know that that's what's going
38:35to work. That is what the audience is
38:36resonating with. And this is this
38:38applies to product as well, right? Like
38:40when I think that a lot of first-time
38:42founders, they they spend time thinking
38:44about like I'm trying to get the market
38:46to buy this and in reality it's like I'm
38:49try the the the pros at this is like
38:51what does the market want to buy? Can I
38:52build it and can I sell it to them?
38:54Right? Like that is actually how you
38:56start a business. And it it for some
38:59reason it's this this flipped thing
39:01where they're like, "Oh, I'm trying to
39:02invent a new idea." I don't want to
39:03invent a new idea at all. Well, I want
39:05to be like, what do people want to buy
39:07that currently like they can't buy and
39:11can I go and figure out this the way to
39:13build that thing? And then I know I can
39:15sell that back to them. I know it's the
39:16market is going to be receptive to and
39:17you need to think about content in the
39:19same way where like what is the content
39:20that the market is currently receptive
39:22to and by mining that content from other
39:25sources that has already had a viral
39:26moment. This is a way to leaprog that to
39:28identify that and then you're going and
39:30you're putting your own spin. You're
39:31putting your own, you know, angle on
39:33this. So anyway,
39:34>> lot of thoughts there. Um, agreed on the
39:37social media manager is like that role
39:40is dead or it's evolve. It's going to
39:43evolve like it's going to evolve into
39:45the social media agent man manager. So
39:48you're going to need to be able to spin
39:50up agents so that you can create a bunch
39:54of accounts on the fly that
39:55systematically creates content like you
39:57have. Like you get millions of
39:59impressions a month, free impressions.
40:02actually the platforms are paying you
40:04which is insane to do it. It's insane.
40:07And
40:08>> I get paid to build lead pipeline. Like
40:10think about that.
40:10>> It's crazy.
40:11>> And like I I it's so funny, man. I'll
40:13talk to like founders or like you know
40:16large like people that that run bigger
40:18companies and they'll they'll be like
40:21why are you why would you would you
40:22invest in social? And I'm like look at
40:24the earned media. Like if you were
40:25paying for those impressions on
40:26platform, for example, on LinkedIn, it's
40:28like $22 per thousand impressions is the
40:31average, right? So like every post that
40:34you get, even with an account that's
40:35like 500 followers, you can get a,000
40:37impressions. That's like $20 that you
40:39just like put into your pocket for free,
40:41right?
40:42>> But but it's it's so there's the earned
40:44media side and then there's also like
40:46the platforms pay you. Like YouTube
40:47literally pays you to do marketing for
40:50late checkout. Like what the what the
40:53hell? [laughter]
40:53>> It's crazy. It's crazy. And then, you
40:56know, for the people who are like,
40:57"Well, I don't want to do a personal
40:58brand." Makes sense. What Cody is
41:00suggesting is like have people on your
41:02team have these personal brands. And if
41:04you don't, and by the way, I'll give you
41:05a piece of sauce. If you don't want to
41:06do that, another really uh smart thing
41:10to do with agents creating content for
41:11you is creating theme-based pages or
41:15topic based pages. So, for example, my
41:18good friend uh Julian Shapiro, you know,
41:21he had a company, a growth agency called
41:24Demand Curve.
41:25>> Absolute goat, by the way. His blog is
41:27incredible and that's what I came up on.
41:29So, I'm just like one of
41:31>> I actually grew up with Julian.
41:33>> No, did you really? That's amazing.
41:34>> Yeah, he was like my name.
41:35>> He was like a farm now or something,
41:36right?
41:37>> Yeah. That's awesome.
41:38>> Yeah. So, I need to get him on the pod.
41:40that uh Julian being the smart guy he
41:43is, it's not like he created a uh X
41:47account that was slash demand curve. I
41:49mean maybe he has that, but he actually
41:51created an ex account called at Growth
41:54Tactics.
41:56So he's creating content on this growth
41:58tactic page. People interested in growth
42:01tactics follow it and then they learn
42:03about his agency and his products,
42:07right? That's social media company. And
42:09like again, it doesn't it could be I
42:11mean there's the ones that are my
42:12favorite are like Chase passive income.
42:13I don't know if you've seen this.
42:15>> Yeah.
42:15>> Um they're doing it more as a meme page,
42:17but like you can use like this attention
42:19that you can garner for free as a way to
42:21drive inbound for whatever whatever it
42:22is that you're building. It doesn't have
42:24to just be you. It can be this like
42:26anonymous thing that is still providing
42:27value that you're aggregating and you
42:29know organizing for the internet, right?
42:31So I'll leave it there. I don't know.
42:34>> Really,
42:36>> uh, you know, impactful marketing agents
42:40that you just broke down. Um, I wish we
42:43had 40 hours together and we did like a
42:46crazy comment below. That's the only way
42:49I come back. That's the only way he'll
42:51have me. All right. So, you have to do
42:53this. You have to comment what you want
42:54to learn. I'll take I'll teach you
42:56whatever you want. It can be how to
42:58build social media agents like for Tik
43:00Tok clouds. It can be like, "How do I
43:02actually run a paid ads account?" It can
43:04be anything that you can imagine. It can
43:06be direct mail. I'll literally walk you
43:09through how can you send direct mail at
43:10scale by scraping Google Maps. You name
43:13it. How do you advertise on TV and
43:15what's the meta there? Uh like how do
43:17you get cheaper clicks on LinkedIn? I I
43:20can break down any of that. So, I
43:22appreciate you, Cody. We'll see you in
43:24the comment section. Like always, I'll
43:26include links for where to follow Cody
43:28on the internet in the show notes in the
43:29description. Can I shout it out? She
43:31give me give me the opportunity.
43:34>> Go for it.
43:35>> Hell yeah. Go find me on Twitter,
43:37LinkedIn. That's where I'm the most
43:38active. And if you want to deploy these
43:40exact agents that I talked about today,
43:42go to graph.com. Uh we have both the
43:45platform solution for this and also we
43:46forward deploy software engineers to do
43:48these actual implementations on our
43:50platform. We would love to help you. If
43:51you're a fast growing company, that is
43:53who we're seeing the most success with.
43:54So thanks for having me, G.
43:56>> God bless you, Cody. I'll see you next
43:58time.

About this transcript

This page contains the full transcript of Marketing Agents Masterclass (GROW your startup) by Greg Isenberg, generated from the public captions YouTube serves with the video. The transcript has 8,932 words across 1,258 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.