So What? Introduction to Claude Code

Katie Robbert – 00:34

Well, hey everyone. Chris got me dancing to the theme song. Happy Thursday. Welcome to So What, the Marketing Analytics and Insights live show. I am Katie, joined by Chris. John is in small town purgatory. He does not have internet today, so we will carry on without him.

Today we are doing an introduction to Claude Code. Last week, in case you missed it, we did an introduction to Claude Cowork, which is different from Claude Code, which is different from Claude chat. Now again, I like to reiterate these just to make sure I’m speaking correctly.

Claude chat is what most people are using. You open up Anthropic’s Claude, get the chat box, and say, “Today I want to write a post on this,” and it says, “Great, happy to help,” and it gives you a post. That’s the chat feature.

The cowork feature is available if you download Claude desktop. Then you can access files from your local machine. If you have proprietary information, things that you want to keep more secret, or if you just have a lot of information and you want to access skills and other agents that you build into your system, you have all those prompts. That’s what Claude Cowork is good for. If you missed that episode, go to the TrustInsights.ai YouTube channel, go to the So What playlist, and you’ll find it.

This week we’re talking about the third component, which is Claude Code. Chris, you’ve been using this for a very long time. I personally have not dabbled into it. My coding extends to updating the HTML code on our website, but that’s about as far as I’ve coded.

Personally, as I’ve talked about before, I managed developers for a long time. Development and coding doesn’t hold a lot of interest for me. I think my time is better suited doing other things. Chris and I have decided that’s fine. I’m happy to support him with the coding and he’s happy to do the coding. So Chris, let’s talk about Claude Code and what we need to know.

Christopher Penn – 03:00

Well, first let’s make things really complicated. There are three versions of Claude Code. There is Claude Code on the web, where Anthropic has a little thing on Claude.ai and you can do Claude Code work there. You need to have a Git repository, which is a form of version control software, to do that.

There is Claude Code built into the Desktop app, which is essentially almost identical to the one on the web. That can do things similar to Cowork and access folders on your desktop.

The most powerful version is Claude Code in your terminal or command line environment. That is the full-powered, can-do-just-about-anything, has-all-the-bells-and-whistles version. That is the one that I use and it’s the one that most people use. Anthropic realized when they were watching people using their tools that everyone is using it to do non-coding tasks like planning vacations because it’s such a smart environment.

Today we’re going to talk about the command line version of Claude Code. We’re going to leave the other two off because honestly, if that’s what you want, use Cowork. That’s the easiest way to put it.

Katie Robbert – 04:32

Before you get too far, I want to pause and talk about GitHub for a second just to give the overview. You talked about how it’s a repository for your code. In software development, one of the best practices of coding is to check your code in and out so that you’re not coding against a live environment.

Code is basically just a series of if-then statements in a text file document. They say, “If this happens, then do this.” There’s a very specific language to make all of that happen. When Chris says GitHub is a repository, you put your code there. When you want to work on it, you check it out or make a copy, make your modifications, and then check it back in. That protects the code, protects version control, and makes sure you’re not updating things on your live environment. We don’t want to bring down all of StreamYard in the middle of a live stream.

Christopher Penn – 05:46

GitHub is a website and a company owned by Microsoft, though the Git protocol was invented by Linus Torvalds, the inventor of Linux, 30 years ago. Git can be installed on pretty much any computer and everyone using Claude Code or similar should use it for everything.

Version control software is basically unlimited undo. When I was working on the unofficial LinkedIn algorithm guide, I used Git with Claude Code and said, “Let’s start a checkpoint.” I make some edits and then I check it back in. If I go, “Oh crap, I deleted something that I didn’t mean to,” I can go to Claude Code and say, “Hey, bring this back.” It will go through the Git repo, find it, and it’s back from the grave.

Any content—images, video, audio, text—if it’s on your computer and it’s important, Git is another form of backup and you should use it.

Katie Robbert – 07:07

I wanted to make sure we acknowledged what it was. As we’re talking about software development, having an understanding of these other tools is important. You can write code in any system, but if you have no place to put it, save it, or protect it, that is something you should definitely consider for your own sanity.

Christopher Penn – 07:35

And governance. The first thing you’re going to have to do is install it. Claude Code is a command line application, which means you need to be able to use the command line to install it.

We’re not going to spend a whole lot of time on how to use that in this show, because in our Analytics for Marketers Slack group—which is free and you can join at TrustInsights.ai/analyticsformarketers—I’m going to be putting six prompts into the chat there.

What you do with these prompts is you go into any AI system you like and drop in the one you want. There are prompts for Macs or Windows, depending on your level of skill: Beginner, Intermediate, or Advanced. Beginner is “I don’t know what a command line is.” Intermediate is “I know what it is, but I don’t really use it.” Advanced is “I live in the command line.”

Drop the prompt for your skill level and say help. It will walk you through interactively every step of the way and hold your hand to get set up.

Katie Robbert – 09:03

And that is TrustInsights.ai/analyticsformarketers. If you’re already a member, you’ll see it. If you’re not a member, it is free to join.

Christopher Penn – 09:15

Once you install it, this is what Claude Code looks like. You go to a new terminal, type in the word “claude,” and you’ll see this. It is a terminal program and now it’s up to you what you want to do. Think of this like a brand new window; the language model is waiting for you to do something.

The first and most important thing are slash commands. These are basically the buttons because there is no UI. Slash commands like /skills allow you to check your usage. You need to have a paid plan to use this. This will tell you how much of your quota you’ve used. For example, on the Trust Insights Claude account, we have used 83% of our weekly usage.

This is a good thing, by the way. It’s use it or lose it on the usage. We pay the $200 a month plan for Trust Insights. If this usage level was consistent throughout the month, Katie, guess how much we would be spending in API calls? $2,700 a month.

Katie Robbert – 11:35

Yeah, no thank you.

Christopher Penn – 11:37

The Claude plans are a great deal. One of the more useful commands is /model. This lets you choose which of Anthropic’s three models you want to use. In general, if you are doing stuff like summarization or basic document writing, Sonnet is pretty smart and very efficient. You’ll use up less of your quota using Sonnet than if you use Opus.

Opus is the highest level plan; it is the smartest, the most expensive, and it chews through your quota the fastest. The last one is Haiku, which you really only want to use for things like summarization because it’s extremely fast and dumb by comparison.

Katie Robbert – 12:38

I am really glad there is a transcript for this video because you lost me a little bit there. That’s a lot of information and I’m already feeling a little overwhelmed. The one thing that I got was the slash command, which is a very basic function if you use software like Slack.

I’m trying not to be intimidated by trying to see what resonates. I understand different model types in other contexts, and I can see on your screen it gives an overview of which one to use. That’s helpful. I’m ready to see what it can do.

Christopher Penn – 13:39

We can’t do that yet because there’s still a bit more prep work. Claude Code supports plugins. If you type in /plugin, it’ll take you to the plugin marketplace. It comes installed with Anthropic official plugins, which are all free. Be very careful if you install any marketplace other than Anthropic’s because it’s other people’s code. They are basically installing code on your machine.

There are 50 different plugins you can choose from. The only one I would suggest most people will get use out of is one called Superpowers. It has built-in tools that can help you do brainstorming, review stuff, and improve things. It’s smart enough to know in a conversation, “Oh, you’re doing some brainstorming. Let me help you with that.”

Katie Robbert – 15:18

Is there any way to make this terminal screen a little bit bigger? It is really hard to read. That’s better. Now I can see things like Atlassian Superpowers, explanatory output style plugin, Notion. That’s really helpful to know. But Chris, to your point, just be aware that there could be a cost associated or it may not be an official plugin.

Christopher Penn – 16:20

If you are a design shop, there’s a Figma plugin. Claude can then control Figma and help you make designs. If you use Jira, you can connect your Claude Code directly to it. I talked about Superpowers already; Notion is in here, and Slack is very popular. There is also Asana.

The last two things that Claude Code has that are not available in other systems are agents and skills. An agent is something you make that is specific to the way that you do things. One of our agents is the Voice of the Customer agent. This takes an ideal customer profile and makes it available as an @command. You can say “@voc, using the Trust Insights ICP, evaluate this document.” It will return an audit saying, “As your customer, this sucks,” or “I would love to buy that.”

Probably the one that people who watch our live stream know is the virtual co-CEO. This is a virtual version of Katie. Anytime I’m working in Claude Code, I can go “@co-CEO,” and it will say, “I’m here to help,” and then immediately trash my plans.

Katie Robbert – 19:03

I like to think of it more as providing necessary structure to what could possibly be an unwieldy shiny object.

Christopher Penn – 19:17

Exactly. Think of an agent like an app. You can also access skills, which are like macros or a predefined routine. One of my favorites is called fact-check. You can say, “@claudecode fact-check this document.” It will run a skill I built that looks at supporting documentation. If it sees you making claims that aren’t there, it will search the web for credible sources. It then produces an audit report saying, “Here are the sources grounded in truth and here’s the crap you made up.”

If you find yourself giving the same prompt over and over again, that is an opportunity to turn it into either a skill or an agent.

Katie Robbert – 21:11

Last year we talked about MCPs, the Model Context Protocol. My understanding was that it was the connection between an LLM and third-party data or tools. It sounds like this could replace the need for building a separate MCP, or is this an MCP?

Christopher Penn – 22:07

An MCP is an API for AI, a specific way for it to talk to a specific data source. A lot of the things shown in here as connectors or plugins are actually MCPs because they connect. I am building one right now called Co-Blogging, which is an MCP that lives inside the Trust Insights website.

When it’s done, Katie, you will be able to say, “Hey, go to the services pages and make these changes for me.” Claude will understand that it needs to connect to the Trust Insights MCP to do those things. If whatever you’re doing has an API, Claude Code could potentially write the code to talk to the API directly or help you build an MCP.

Christopher Penn – 24:12

What do you do with this thing? First, you’ll want to learn how to navigate to folders on your computer inside of a command line. Please do not just give it access to your hard drive.

For today, I have created a folder called Live Stream. I started Claude Code inside that folder, so it has access only to what’s in here. I create my same eight folders that I personally like to start with for governance. Some of the easiest use cases start with connectors.

I connected it to Asana and said, “Add to my to-do list: I need to make a LinkedIn post to promote Katie’s new AI Readiness Assessment.” Inside my Asana folder on my desktop, I have a template called my daily briefing. When I get in the morning, I open Claude Code and say, “Give me my daily briefing.” It looks through my Asana tasks due in the next three days so that I can plan ahead. I can talk back to Claude and say, “Make these changes to Asana,” and it will adjust it.

Since I installed the Asana MCP inside of Claude Code, the gigantic mess in my Asana is gone. It’s all organized. I even found a way to list all my tasks that don’t have a time estimate. I turn on the voice memo app on my phone and start reading down the list, dictating how long each will take. I put the transcript into Claude Code and it updates Asana. If Katie wanted to use the workload feature, she now has realistic estimates. I don’t have to click 80,000 times; I just dictate it.

Katie Robbert – 27:33

If putting a Claude Code connector on top of our project management tool gets you to use it, I am happy as a clam. I feel like these tools open up the possibility to meet people where they are versus trying to force them to use things that don’t align with their working style. Someone is asking: “Do you find that Claude overestimates time?” Chris, I think you are the one underestimating time.

Christopher Penn – 29:04

Claude dramatically overestimates. If you say, “Claude, do your best guess,” it might say a task will take seven hours when it’s really a five-minute task. You have to meet it in the middle.

One of the big cautions is that executive function—planning, organization, decision making—is something you want to be doing. If you let AI do the high-level thinking, you are de-skilling yourself. Your company won’t need you because they can just use Claude instead. Let it do grunt work like updating Asana, but you should be the one deciding when to do a task or how long it will take.

Claude Code is also better than Google at deep research. I use the Trust Insights CASINO framework, which you can get at TrustInsights.ai/casino. I gave Google Gemini a research task to tell me how to build micro-dramas—90-second, bite-sized videos. I wanted a manual. It came back with 5,700 words of filler content.

I took the exact same CASINO format into Claude Code. Because it has its own built-in web search agents, it can run 50 agents at a time. It spat out a report that was twice as long as Google’s and followed the requirements exactly. If you need strict guardrails on output, you should switch to the Opus model because it’s the most precise. Then you deploy the fact-check skill. When I did this on the Gemini report, 30% of it was hallucinated.

Katie Robbert – 34:09

It makes sense now why you like Claude Code so much and why you don’t like managing people. You don’t like that gray area of, “Well, I told you how to do it, but I did it this way.” If you’re someone who has that precision in mind, Claude Code gives you that back.

Christopher Penn – 34:50

I told it, “You’re going to perform the research task in this file and output it to the output folder. Use the brainstorming skill to develop a plan.” It loaded the skill and asked me questions. Then it kicked off two dozen web search agents that brought back data, synthesized it, and wrote the report. It gave me exactly what I wanted.

The hardest thing for people is thinking through what the “recipe” looks like. My recipe is the CASINO prompt framework. That prompt is a thousand words long because I took the time to think through what I wanted.

The final use case we’re going to cover is coding. Claude Code is the best coding tool available right now. You don’t have to be a coder as long as you have the vocabulary to describe what you want.

Last week, Katie and I had a 45-minute conversation about an idea she had. The requirements document plus the transcript were the basis for Claude Code. I said, “Let’s build a full product requirements document.” One of the secrets to Claude Code is that the more chances you give it to think, the better it does. It asked me questions like, “If Katie wants this, what about this?” and I said, “Ask your co-CEO.”

Katie Robbert – 39:05

This is live now if anyone wants to try it: TrustInsights.ai/ai-readiness-assessment. The thing that worked is that we went into this knowing exactly what we wanted. I didn’t say, “This has to be the interface,” because that’s not as important as the purpose. I used the 5P Framework. I handed Chris a fully baked document including the scoring mechanisms and how things are tallied.

Christopher Penn – 40:33

I gave Claude the CASINO framework and it built a rules file with the patterns and anti-patterns we were going to use. I coded none of this. It then went and just built it.

As a marketer, why would you do this? AI can summarize a static document like a blog post, but the web doesn’t need another summary. Claude Code enables interactive content. When an AI crawler looks at that page, it doesn’t fill out the form or get the assessment at the end. That is why you would want to use Claude Code—to bring an idea for interactive content to life.

Katie Robbert – 42:47

I wanted it to be something interactive, not just a Google form. I wanted it to feel like part of the website.

Christopher Penn – 43:33

On the back end, Katie wanted to know how to get this into HubSpot. HubSpot has a gazillion APIs. Claude can capably write code for APIs. We gave it the instructions and credentials, and it built that plumbing for us. No one has to copy-paste data; the system does it because Claude Code built it.

Katie Robbert – 44:28

Documentation for APIs exists and many companies don’t gatekeep it. If you have that, you can give it to your LLM and say, “Help me make sense of this.”

Christopher Penn – 45:00

Last example: I had Claude Code write a trashy romance novel because it was the holidays and we had unused quota. I had a writer’s bible and a blueprint. It generated the overall outline, then chapter by chapter, then wrote each chapter.

I told Claude Code to launch up to 50 agents at a time. It launched 32 different agents and wrote each chapter simultaneously. Instead of waiting eight hours, it banged it out in 15 minutes. Then 32 editor agents launched and edited each chapter. The agents part of Claude Code allows you to scale way more than Cowork can right now.

Katie Robbert – 48:08

Next time, I request a murder mystery about a pushy person following breadcrumbs. I’ll let you know if it’s any good.

Christopher Penn – 48:51

Duly noted. To wrap up, Claude Code is the third piece of the ecosystem. We recommend the command line version. It operates locally but connects to just about anything. You need to understand slash commands, plugins, agents, and skills. Critically, do your homework before you turn it on. It will do what you ask, even if what you ask is stupid. Do the planning, and then set Claude Code off.

Katie Robbert – 49:57

If you want to gauge your AI readiness, go to TrustInsights.ai/ai-readiness-assessment. It was a great use case because we are a small company and don’t have the budget for external contractors. This is getting us to where we need to be.

Christopher Penn – 50:42

I encourage folks to give it a try. The prompts for getting it set up are going to be in the Analytics for Marketers Slack group in about 10 minutes. Thanks for watching, and we will talk to you all on the next one.

Katie Robbert – 51:14

Be sure to subscribe and check out the Trust Insights podcast and our weekly email newsletter. Join our free Slack group at TrustInsights.ai/analyticsformarketers. See you next time.

Would you like me to extract the specific CASINO framework instructions or the list of Claude Code slash commands mentioned in this transcript into a separate guide?