Free YouTube Transcribe

Video transcript

I Replaced My CMO With Claude Code (Here’s How)

Max Mitcham · 2,475 words · 12 min read

Want to search this transcript, jump the video from any line, or download it as TXT, SRT, or VTT?

Open in the transcript tool

Full transcript

0:00I have zero idea how to code, but over

0:02the last few months I've built an AI

0:04system that handles pretty much all of

0:07our marketing stuff. Things such as

0:09competitor research, content strategy,

0:11SEO audits, blog creation, all entirely

0:15through Claude's code. It's essentially

0:18like my AI CMO, which is now helping me

0:21run my seven-figure SaaS company,

0:24Triggerfy. So, in this video, I'm going

0:26to show you the skills that I'm using,

0:28how I set up my repo, and how I'm using

0:31the new feature that Claude recently

0:32brought out, which is their Discord

0:34feature, so I can chat with Claude code

0:36while I'm on the go. So, let's dive in.

0:39So, first things first, you will need

0:41to, if you haven't already, create a

0:44repo, which is just a a folder system

0:46essentially. And in that folder system,

0:49the first thing that you're going to

0:50want to do is create a thing called

0:52skills. Now, skills essentially is a

0:55step-by-step function that Claude code

0:58will follow. In fact, and if you're

1:00using Codex or any other tools, they'll

1:01also follow this. They can use skills as

1:04well. Think of it like a mini automation

1:06that you are writing in written word

1:09essentially. The best way to create a

1:11skill is first start by understanding

1:14what do you want to be produced? Is it

1:16things such as SEO audits? Is it blog

1:18creation? Is it social content creation?

1:21Is it competitor research? Whatever it

1:23is, think about what what you want to do

1:25there. And then you want to think about,

1:27okay, how can I go ahead and get amount

1:31of knowledge on what good looks like?

1:34Because skills is essentially training

1:36an agent to say, "Hey, these are the

1:38step-by-step functions that I want you

1:40to follow, and this is what a good

1:42output should look like, and here's how

1:45you get there from A to B essentially."

1:47So, if you look at this one here that I

1:49have for blog writing, you can see here

1:52it's got step-by-step instructions. So,

1:54first of all, it needs to read a brief

1:56from my blogs brief file. Then it's

1:59going to load the Triggerfy context,

2:01which is the Triggerfy knowledge base.

2:03Then it's going to go ahead and write

2:04the articles. Here's the structure that

2:06it should follow. Here's the

2:08architecture that it should follow. What

2:10good looks like, what bad looks like, so

2:12on and so on basically until it then

2:16goes ahead and creates We're using

2:18Linear, so it creates the Linear issue,

2:20which is the blog, uploads that into

2:22there, and then updates the status

2:24tracking of that as well. And you can

2:26see then here we have another skill

2:28associated to blog statuses, things like

2:31that. Another really good quick example

2:33here to show you is social content. So

2:36this is again, how does it write great

2:39newsletters, great LinkedIn posts, great

2:42Twitter posts, or X posts, right?

2:43Whatever you want to call it. And this

2:44one's a big one and again it's what does

2:46good look like, what do the algorithms

2:49look for, how can we analyze a good

2:51hook, so on and so on. So an example

2:54would be is if I was to fire up Claude

2:56here, I could ask it something such as

2:58what makes a good LinkedIn post. And

3:01you're going to see automatically this

3:03is going to go ahead and use that skill.

3:06There we go, because we've basically

3:08created a bunch of hook words, which is

3:10then going to say, "Okay, I found the

3:12skill that that's associated to. I'm

3:14going to read it, and here's what makes

3:16a good LinkedIn post essentially." So

3:18that is skills. Now, how do you go ahead

3:20and create them? Well, maybe you are an

3:22expert, you have the domain knowledge in

3:24a certain area. So maybe it's just

3:26pouring your brain out into Claude. The

3:29best thing for you to do is use Claude's

3:32plan mode. So do {forward slash} plan,

3:34and then you want to say, "Help me

3:36create a skill on then you would enter

3:40in what you want it to be. I want you to

3:42interview me to help really make this

3:46watertight, right?" Something along

3:48those lines. You You want to make sure

3:50that it's interviewing you to really

3:52pull that knowledge out. Now, maybe you

3:54might have domain expertise, but maybe

3:56you're like, "Yeah, I could still

3:57probably do with a little bit more."

3:59Well, here's then where you could tell

4:01it to go ahead and research, go ahead

4:03and search the internet or wherever it

4:05is to also top up potential gaps and

4:08then come back and suggest things to you

4:10and again interview you based on the

4:12research that it's done. So, that leads

4:14me to the next one, which is how does it

4:16do that research? How does it scrape

4:18data? How does it pull information? How

4:19does it find information and all of

4:21that? How do we connect it to things? Um

4:24so, we're going to ask here, "What

4:25tools?" In fact, I'm just going to close

4:27this down here. I'm going to reload it,

4:29get out of plan mode. What tools do I

4:31have connected?" You would have heard of

4:33CPIs at this point, you would have heard

4:34of APIs, and you're now starting to

4:36probably hear about CLIs. Now, why are

4:39people going from MPC to CLI? Well, the

4:43main thing is is that an MPC basically

4:46gives all of this context on how to use

4:49an API. It's almost like a USB on like

4:52this is how to basically interact with

4:53the API. The challenge with it is it can

4:55overwhelm AIs because it injects tons of

5:00context in there, which isn't really

5:02needed anymore. And therefore, the

5:04context window, the size, which is not

5:07actually showing me here for some

5:08reason, but the the context of how big

5:11that session, which is typically 200

5:13tokens, here we're using the Opus model,

5:15so it's 1 million tokens, that's quite a

5:16lot. It will eat into that and you don't

5:18want that. And so, that's why CLIs are a

5:21lot more popular, which is again more of

5:23a friendlier way for an AI to interact

5:26with external APIs. So, you can see here

5:28I've got things such as Trigger Five

5:30setup, which is social listening, social

5:32management, basically. Hey Reach, Clay,

5:35LinkedIn, we've got Linear, we've got

5:36Notion, Discord, Gmail, Feature Base,

5:39Fireflies. Fire Crawl is the one that I

5:41use for all of our web scraping,

5:43crawling, search based, Browser MCP, and

5:46then Claude and Chrome. We've recently

5:47just switched across to Claude and

5:48Chrome, which essentially is the ability

5:52for Claude to actually control your

5:54Chrome browser. Whereas before, we were

5:57using things such as browser base, which

5:59uses Playwright. For those who aren't

6:01that technical, basically means it was

6:02creating a test environment. Now, Claude

6:05can actually have access to your

6:06browser. So, things that you're logged

6:08into, your sessions, your cookies, all

6:10of that, which makes it extremely

6:12seamless experience for it controlling

6:14things. For example, if you're part of

6:16any networks or any groups, you could

6:18have a task set up where it goes ahead

6:21and kind of reads everything in there,

6:23and then updates its own knowledge base.

6:26Which takes me on to the next key and

6:28critical part to this whole thing is

6:30tasks. How do we run them? What are

6:32they? Well, you probably heard a lot of

6:34people talk about things such as cron

6:36jobs. Cron jobs are basically the techy

6:39world's version of something like NA10

6:42as an example. It's basically, again,

6:44just a step-by-step function on how to

6:47execute a script or a series of events.

6:50So, for example, a cron might be every

6:53day go ahead and query this API, and

6:57give me a summary of whatever you wanted

7:00to give you a summary on, for example.

7:02And you can set up local cron jobs,

7:04which basically means you need no tools

7:06for this, but if your laptop is on, you

7:08can say, "Claude, set me up a cron to go

7:10and check my Gmail or my calendar for

7:13the day, and then give me a summary,

7:16send like a message to Slack as like a

7:18summary of what I've got on for the day,

7:19for example." The downside to that is

7:22you have to have your computer on

7:24always. So, then Claude recently added a

7:27new kind of schedule, which you can see

7:29here. If I do forward slash, schedule,

7:31which gives you the ability to create,

7:33update, list, or run schedule or remote

7:35agents that execute via a cron schedule.

7:39So, this basically takes it into the

7:40cloud, which means it doesn't matter if

7:42your computer is on or not, which is

7:46awesome. Or you can get a little bit

7:47more technical and use tools like

7:49trigger.dev or ingest, which gives you

7:51even more finite control, but I won't go

7:53into that. That's a video for another

7:55day. So, how is my kind of repo, which

7:57is my folder system, essentially set up?

8:00Well, we have all the skills which I've

8:02shown you there. Then, inside of here,

8:04we have different folders. For example,

8:06we have a blog automation and a cron

8:09that runs where we are initially doing

8:12an SEO audit. So, the SEO audit

8:15basically is using an SEO skill that

8:18we've created, which once a month

8:20basically goes ahead and runs research

8:23based on the performance of our website

8:25page. So, it uses Ahrefs and then

8:26basically says, "Okay, hey, you know,

8:28this is everything that is going on.

8:31Here is some recommended content

8:33calendar. Here is the pillars that you

8:35should focus on, and then here are the

8:37clusters that you should focus on,

8:39guidelines for that, so on and so on."

8:42So, then what happens is when the agent

8:45fires up to write a blog, it checks the

8:48blog. First of all, it checks what

8:50clusters have we done, what briefs do we

8:52have, and then it kind of works its way

8:54through this report every single day.

8:56And now we're writing five, six blogs a

8:59week purely because of this step-by-step

9:01function, which, you know, this type of

9:03report would have cost thousands for

9:04anyone to produce for us. And then, on

9:07top of that, all of the blog writing,

9:08which totally SEO-optimized and

9:11AI-optimized, as well. Then we have

9:14things such as like daily insights. So,

9:16here basically I have like a folder

9:18system set up, which is giving me

9:20analysis on how is my business doing,

9:22MRR, ARR, churn, you know, what's going

9:25on and things like feature base, which

9:26is where, you know, the like an Intercom

9:28system, things like that. Then we have

9:31our knowledge folder. So, inside of our

9:32knowledge folder, we have what is

9:35Triggerfy. So, how did I create this

9:37knowledge folder initially? First, I

9:39hooked up Fireflies, which is our call

9:41recording system, and I told it to go

9:42and listen to the last 100 calls that we

9:44did and create an analysis of what

9:47Triggerfy is based on those sales calls

9:50that we had done. Then, I told it to go

9:51to our website, ingest all of that.

9:53Then, I told it to go to our help desk

9:55or our help center, ingest all of that

9:57information, and then it created its

10:00knowledge repository off the back of all

10:03of that data. And again, you guessed it

10:04correctly, we've created a cron setup

10:07that basically kind of refreshes that,

10:09right? Because the company is forever

10:11evolving. But then, we also have this

10:12competitor one setup. So, this is a

10:14recent one where we're using Triggerfy

10:17that's monitoring our competitors such

10:19as Brandwatch, and then we're doing all

10:21this stuff. We're having a summary how

10:22they position themselves, how other

10:24people kind of position them, and so

10:27that we have this knowledge repository

10:28on these competitors as well, so that we

10:31can kind of tap into this folder system,

10:33this MD file essentially whenever we

10:36want that to happen as well. And there's

10:38so much more here, right? But like,

10:39there's not enough time to go into to

10:41everything. So, this leads me on to my

10:43last part, which is maybe we're not that

10:46friendly with terminals. So, the

10:48question is, how do we chat to it

10:50through Discord? Well, this is where

10:53things such as plugins come into handy.

10:55So, here you can install, for example,

10:58Slack, or you can install Discord, which

11:01you can see is connected. So, if I then

11:04share my Discord screen here, you'll

11:06then see that I can now go ahead and tag

11:09and speak to Claude from here. Say hello

11:13to everyone. I'm running a demo. Um and

11:17so, you create basically a bot inside of

11:19Discord. You connect it to the plugin

11:22that you've done, and here you are. You

11:24can then see. So, mine one is called

11:26Jinx, for those of you who watched

11:27Arcane on Netflix, League of Legends.

11:30That is me. So, we've got loads, as you

11:32can see here. We've got loads of

11:34different We've got Bella, Jinx, uh Rex.

11:36We also have Taz, Rock. We've got

11:38different agents doing different

11:39specializations, different skills. This

11:41is a video for another day, because this

11:43is a mixture of open claw and clawed

11:45code. But, this is how I'm communicating

11:48on the go with my repo, which is my

11:51marketing repo. So, now I can chat to it

11:53and say, "Hey, let's go write a blog on

11:55this. Hey, can you use a Triggerfy API

11:57and let me know if anyone's talked about

11:59XYZ? Can you help turn this transcript

12:02into a LinkedIn post, into an X post,

12:05and then go ahead and post that for me?"

12:06And it acts as my command center, and I

12:08can chat to it, whether I'm on the go or

12:11just in a simpler way on the laptop

12:14itself. So, if you want to start

12:15building something like this yourself,

12:17the first thing you need is a solid feed

12:19of market intelligence coming into the

12:21system, which is something I think I've

12:23spoken a lot about, which is that

12:25context. That's what keeps everything

12:27grounded and what's actually happening

12:29in your marketing right now. Not what

12:31was happening 6 months ago. That's

12:33essentially what the tool I built,

12:35Triggerfy, handles. It monitors signals

12:37across LinkedIn, Twitter, Reddit,

12:40GitHub, YouTube, podcasts, and so much

12:43more, surfacing what matters to your

12:46agents and then routes it to wherever

12:48you need it to go. So, try it for free.

12:51It's the first link in the description.

12:52And if you want a complete guide on

12:54social listening for B2B marketing,

12:57check out this video next.

This transcript was generated from the captions YouTube publishes for this video. Get the transcript of any YouTube video atfreeyoutubetranscribe.com: free, unlimited, no sign-up.