So What? Setting up Claude Cowork

Well, hey everyone. Happy Thursday. Welcome to So What the Marketing Analytics and Insights Live Show. I am Katie holding up Chris and John. Hello.

It's a message. It's us. Um this week we are talking about setting up Claude Co-work. So Chris and I have been talking a lot about how the heck do we get ourselves out of the weeds of running the business day to day? Because as we've mentioned many times over, we're a very small company.

And that means that we wear a lot of hats, all of us do, and we're all just trying to, you know, keep moving forward sometimes, just not as quickly as we would like to. So the solution is, you know, we could either hire more people, add more headcount, which quite frankly, you know, budgets are tight all around, or we could find ways to automate more and figure out how do we use what we have and make it more efficient. And so one of the things Chris has really been talking about a lot is Claude. So Claude as a large language model has a lot of different things that it can do. And Chris is going to get into it.

One of the things he's talked about a lot is Claude Code. We're not talking about Claude Code today. We're talking about Claude Cowork, but I think it's worth acknowledging there's Claude, the LLM chat. There's Claude Code, which Chris will talk about, and then Claude Cowork, which is what we're gonna focus on today. Do I have all of that straight so far?

So far, so good. All right. That's where I'm gonna tap out now because I don't want to get it more confused. So, Chris, over to you, Chris. All right.

So, a couple of things up front. Claude Co-Work is a brand new product. It is like two weeks old at this point. It is iterating rapidly, and it is in very limited testing. So only people on Pro or Macs plans can use it.

The free plans cannot use it. And it is inside the Mac version, the Mac desktop version of the Claude app. So it's not available on the web, and it's not available on Windows. According to the Claude team, they are working very quickly on a Windows-based version, and they expect to have that out soon, you know, with the capital S soon. Um exactly.

Um, so today we're gonna go over how it is right now. It is going to change. So, for example, in cowork, skills and agents are not formally supported yet. The team uh Boris Cherney over at Anthropec has been saying, Yep, we're working on it as fast as we can. It's very funny.

They actually use clawed code to write co work. And so they're they're trying to add as many things to us as as quickly as they can. Here's why it's important for us to talk about this today. Every other vendor is looking at the Claude ecosystem now and going, yeah, time to get the photocopiers because they've got all the buzz. They've got all you know, you know, valley investors and stuff are all paying very close attention to what Anthropic is doing and going, you know what?

There's a there. Claude co work actually came out of Claude Code because Anthropic was looking at how people were using Claude Code and going, that that's not coding. Like you're planning your vacation. Why are you planning your vacation in Claude Code? And because it has so many tools to do automated work, like go find me flights for this and this, and you could just come back later and is done.

That they're like, what if we made a version that didn't require you to be a gearhead just to get the bloody thing installed? Yes, please. And that is co-work. So let's just uh start digging in. The first thing you're going to need is you will need to download it from Anthropic's website.

Uh if you go, there's a meet Claude, you can see Claude in Claude in all these different places. You are going to be looking for Claude on the desktop. And when you install that app, it looks like this. There's three tabs in the upper left. There's Claude Chat, which is the regular plain vanilla, let's just do the thing.

There's co-work if you're on the Pro and Max plans, and then there's Claude Code, which is their uh cloud-based version of Claude Code, which to add more confusion, there's several versions of Claude Code. If I'm if I'm looking off screen, it's because I am following along with Chris because I just installed Claude for the desktop for myself to learn how to do this. So like I'm literally using this live stream to learn how to do this from Chris live. It's a company. Exactly.

So to orient you, there's a few different things. Obviously, there's the existing chat list here with that we don't have any yet. There's some starter prompts that you can do there. On the right hand side is the status panels that tell you what Claude is working on when it starts working on things. And then there are there's the ability to connect to a local folder on your desktop and then to use connectors.

And so some of the connectors available are things like Claude in Chrome, which you can install. I've installed it on my computer with the big asterisk that it can control Chrome. That means that if you are not thoughtful in how you prompt it, you might be telling it, hey, go um go post some deepfakes on LinkedIn for me, right? If you didn't think that through, it will do those things. The reason why we want to use something like a cowork is is twofold.

One, it has access to our local files, which is such an important thing. And I'll show you in a minute how to connect it to something because we want to also put guardrails around and say, this is the only little sandbox you're allowed in. You're not allowed to go other places. And it can control a browser. Those two things give it the ability to do autonomously a lot more stuff than a regul than the regular Claude chat can do.

We talked about the guardrails uh in more in-depth on the podcast this week, which you can get at trustinsights.ai slash ti podcast, because that was a big question, Chris, that I had for you is okay. So you're installing this stuff on your local machine. You likely keep sensitive data on your local machine to keep it off the web out of Google Drive. So, how do you make sure? And so, you know, obviously, we're just sort of like walking through it quickly, but if you want to get more of a deep dive, check out the podcast.

Or I would also recommend joining our free Slack community, Analytics for Marketers, where uh we can also go into more depth and learn from everyone else what they're also doing to protect their data. Exactly. So we're we're now ready to start doing some of the configuration of this. First things first, it's a good idea to give Claude its own folder per project per task. So don't just give it access to your desktop.

That's a terrible idea. Instead, give it a place on your desktop to work and maybe even give it a folder inside that place. So on my desktop, uh today we want to talk a bit about uh the trust insights newsletter, inbox insights. So I've gone ahead on my desktop and I've created a folder called Inbox Insights Strategy. I'm going to tap the work in a folder here, and it's going to bring up my uh my file browser.

I'm going to go and find my way to the inbox insights folder. And it will say, are you sure you want me to be able to read and write files in this folder? And for this, I'm going to say yes, because I've given you this folder to work in. Please do not just give it your hard drive. Please don't do that.

Yeah, it's just a record. What could go wrong? Yep. And then I have also enabled the clawed in Chrome connector to allow it to control my browser. Uh, because I'm going to show an example of how that might be useful.

Again, with the caveat, you want to be careful how you use it so that it's it's doing things intelligently and not just running amok. It sounds like for those just starting out, maybe wait, maybe a phase two of uh the Chrome permission and just sort of get used to this first before you uh let it take over your Chrome. Look at it you, John Wall and all those nefarious things that you're known for. I know, just let it take your browser and go everywhere it can and start buying random stuff. That sounds like a complete recipe for disaster.

Where did all this stuff come from? All I was just doing was researching, you know, million-dollar homes, and now I'm suddenly the proud owner of five different estates. Whoopsies. Exactly. And a moat.

Remote. That log cabin I've been looking for. Um the next thing we want to do, it has nothing to do with Claude, but this is the part that's gonna make Katie's soul happy. Um, we want to think about the governance inside our Claude folder and how to organize it because you just don't want to have an empty folder, just have pile stuff in there. That's like, you know, that's how my kids clean their room.

They just shovel everything into the closet. I'm like, that's not how to do that, kids. Uh my kids don't watch, thank goodness. So for this recording, I already threw some stuff in here. This is kind of a mess.

What I would suggest doing is having at least three or four folders for every clawed co-work project. And we get this from Claude Code. So I'm gonna create a folder called data that's gonna have data sets in it. And by the way, this is not the way to do it. This is my way of doing it.

You do it your way. I'm gonna have a folder called input. This is where stuff that's not structured data would go, like transcripts for me. I'm gonna create a folder called output, which is where I want claude to put things as it's working on them. And then I always personally have a folder called docs, which is where I put um like rules-based documents.

Again, coming from the coding world, this would be like where my code standards live. For right now, because co-work does not formally support agents and skills, we're gonna create two folders like this. Well, because an agent and a skill, nothing more than prompts. So we're just gonna borrow existing prompts from those things. So we're gonna create a folder called agents and a folder called skills.

So now I've got four of six folders here. Now let's talk about what's in this folder. For today, we want to we want to be thinking about how are we going to. So Katie, actually, why don't you describe kind of what you noticed in our subscriber counts for inbox insights this week and what you want our goal to be. Um, so sure.

If if you're brand new and don't know, we have a weekly newsletter called Inbox Insights. You can go to trust insights.ai/slash newsletter, brand new landing page as of last Friday. Um, you know, we're bringing ourselves into the whatever century this is now. Um, and what we've noticed is that in 2025, our subscriber numbers sort of stalled out and plateaued. And then we did a deep clean and data hygiene of the list itself.

And unsurprisingly, the number went down. So moving into 2026, the total number of subscribers is lower than where we were at the beginning of 2025. And to that, we know from our own data that our email newsletter is one of the largest drivers of engagement and eventually conversion for our services. So it's in our best interest to increase that subscribership, which means promoting it more, writing about topics that people care about, getting more awareness to it. There's a lot of different things that we need to do to hit a target number of subscribers in 2026.

And that is the purpose of what it is we're doing today. If we were to follow the five Ps. The purpose is to increase the total number of subscribers to the Inbox Insights newsletter so that we can continue to see it as a strong channel that brings in engagement and conversion. Right. Uh, which is a fantastic use of Claude Cowork, because one of the things that makes it different than a regular chat is it will build its own to-do lists and try to do tasks if you if you write it out well.

So the first thing we need to do is figure out well, how how do we do this? How do we put this together? And we're going to start with some basics. Uh, the first basic, uh, which we've talked about in past episodes of the show is our ideal customer profile. I've chosen our mid-market growth ideal customer profile from our sales playbook.

And so I've just put that profile in there. It is not the entire thing, not the 135 page document, it's just this one particular segment. Uh, one thing that we've uh started doing is we've started removing anything involving age, a, because it's discriminatory. Uh, and B, because what we really care about is someone's ability to buy things from us. I don't care how old you are, or what generation you are, or what color you are, or where you live, all money is green.

Uh well. Also not a true statement, but that's for a different episode. That is for a different episode. So we have our ideal customer profile. Second, we kind of need to know the landscape.

Like, you know, I I've been in email marketing now for 30 years. One million years. I'm I'm at J I'm in the Jurassic era. Um what's changed? What's new?

And so the easiest way to get that information is to commission a deep research report. So I kicked off a deep research report following the Trust Insights Casino framework, um, saying, hey, inbox insights, is this newsletter? We have this many subscribers, we want to get to this many. We have no budget for paid ads. Zero, none, nothing.

Um, I want to know who is the audience. Uh I want to know like what's happening out in the email marketing world. Here's the sources to find. Focus on 2020 23, 24, 25 if it's relevant. I specifically want to know what are the best practices for retaining your audience today.

What are your best practices for acquiring new subscribers today? What are the mistakes people make when they in trying to retain subscribers? Like what caused you to lose subscribers? What are the mistakes you make for when trying to acquire new subscribers? So those are the four big things.

And of course, you know, I'm not gonna wait 15 minutes. The I went ahead and ran this. The nice thing about um Claude Cowork and many other systems is it can read Google Docs directly, or in my case, because I am old and I like to have things, um, I exported it to a Google Doc and then I dumped it as a markdown file, which is the lingua franca of AI. Um, so that that deep research port now is right here. Uh same exact thing, but now it's just in plain text.

So the AI will have an easier time reading it. When you said that you were old and like to have things, I was fully expecting you to say, and then I printed it out and used a binder clip and took out my highlighter. Like that to me is like printing out everything, printing out all your emails, like that would sort of classify you as old, making things markdown to so it's easier for the AI to read. I don't think that counts. Okay, that's fair.

Um, another question that uh we would probably want to have some research handy on would be, or some information on this, how do we do our newsletter? Right. So one of the things I did before the show uh was I spent and we're not gonna play this. Uh I spent 15 minutes of me just foaming at the mouth about how we do our newsletter, the process, who writes what, the what the major sections are and things. I just recorded it and I turned it just with any transcription tool, transcribed into text.

So we now have a here's how the newsletter works. You'll notice we haven't gotten a claude cowork yet. We're getting all the pieces. What else would be helpful, Katie? To to know if we're going to be doing some strategy about a newsletter.

Probably the existing metrics. So, you know, in I don't know if it's included in any of your background information documents, but here's our current number of subscribers. Here's our goal that we want to hit. Here's the engagement rate. Here's the open rate.

You know, here's the rough makeup of the current, you know, subscriber list, like that kind of because you know, we're saying here's our ideal growth, ideal customer profile, but we need to know who we have today. We need the baseline to if you know Cloud's gonna be like, that's cool. Those are two different audiences. So good luck with that pivot. We don't know that.

Exactly. So you are 100% right. And the easiest way for me to do that was to just take screenshots of our marketing automation system for the last full year of issues. Like you know, how many people was it sent to? What was the the red rate?

Screenshots are just fine because Claude can read images. The other thing that would be helpful is maybe to see how much of our website traffic comes from our inbox insights newsletter. And the place to get this would be uh Google Analytics, right? Yep. So if I go to Google Analytics, and let's go to Google Analytics here, and I go to Explore Hub and I go, I don't know what I'm doing.

Crap. Makes for a great live stream. It makes for a great live stream. Because I have the Chrome connector turned on, I can say I need some help building an exploration in Google Analytics to find out how much website traffic I'm getting from my inbox insights newsletter, which has a session source of inbox dash insights newsletter. Help me build this exploration to see what traffic I get to my website, trustinsights.ai, that's the host name for every day from January 1st, 2025 onwards.

If I put that in as a prompt, and then I say, use the Chrome connector to connect to my Google Analytics as the instruction, Claude will say, okay, I understand what you're what you're trying to do here. I'm going to uh fire up your Chrome. Yay, there's your Chrome. I found it. And it's um it's going to try and figure out where our Google analytics is.

So it's Claude is doing this. You're not my hands are here's my hands. I'm not doing anything at all. Did you just see my blood pressure go up? Right.

This is how it starts. I mean, I get it. I do. But I'm also like, oh, whoa, whoa, whoa. Yep.

So it's now going to the explorations. Now I ran this. This is not fast. This takes this takes it about 20 minutes to go that whole thing and stuff like that. Um, so I'm gonna I'm gonna pause Claude.

Um, but what it came up with, okay. Stop that. Um, and you can see by the way, the up this click up here says Claude MCP. That tells you this window is being controlled by Claude. So uh you will know that it is working on something like that.

Oh this is what it came up with. It, not me. This is what it came up with, it's which is correct. This is the correct way to get this data. But it okay, please continue.

It clearly said host name contains trust insights AI, session source meeting contains inbox insights. The value is sessions, 500 rows by date, and you have this, and now my my thing would be just just download this again. I have stuff. First use case. If you have a piece of software and you don't know how to use it, but you know what to ask for, you can tell Claude Cowork, take control of my browser and start doing the thing.

And you can at any point you can go to the cloud room and just hit stop, stop, stop, stop. You're like emergency stop, like I just did. But if you if you uh have say Adobe Analytics or Adobe Launch or Salesforce or Hubspot or whatever, and you're like, I don't know how to do this. Claude Cowork, walk me through how to do this. So it's now it's gone from watch me do it and and critique me to not just do it for me.

Um, as long as you know what to ask for. And I think that that's the key. And so uh Kelsey has a question. Uh, what about in your insights, which is our podcast? Would you also want podcasts involved for this particular task?

Uh no, because we're focusing specifically on growing the newsletter subscriber list. We will offline repeat this process to figure out how to grow the podcast subscriber list or the podcast downloads. Um, and so they're both important assets, but for the purposes of this exercise, we want to specifically grow the newsletter subscriber list. Exactly. And I did mention in my prep recording that we do mention the newsletter on the podcast and of course on the live stream and stuff like that, so that Claude will know what all the pieces are.

So I have now my Google Analytics data, which is there. I'm probably gonna rename this Google Analytics so that because Claude, as it reads file names, can understand the context of things. Um another thing, actually, uh Kelsey had asked about earlier, no, it was last week was well, what about like G E O A E O I A I O G and X E O. X E O'S. Um, what about that?

Yeah. To which I said, yeah, that seems like a reasonable thing to ask about. Um, so what I did was I uh using our own homegrown XEO tool. This is not publicly available, by the way. Yes.

Uh uh, I went through and I extracted, I asked it to asked Gemini, hey, I'm interested in a newsletter about marketing and AI. What you got? And so I spit out the trace file, which contains um everything that Gemini asked itself about this problem, this problem and all the search terms and queries. And so that's gonna go in as an ingredient because if we're talking about growing our subscriber base, things like what topics should we talk about? What are what are people looking at?

What is Google saying it recommends? And are we aligned with that? So that's another ingredient. And then the final ingredient is um we had Kelsey working on an idea for how do we how do we make our marketing more fresh and and relevant to a broader audience? So we had her put it in the five Ps because we love the five P's around here.

And and there's some great ideas in here. So that's gonna be an ingredient too so we're gonna put all these things in together to Claude Cowork to try and have it sit down and think things through so our next step is governance we've got data what kind of data where does the data go my subscriber data um I'm gonna put in the data folder since that is numeric data right so that is our open rates our our subscriber basis our Google analytics data I'm gonna put our uh our research data all this into the input folder because these are all inputs that we have come up with in some fashion and I'm gonna put our what we currently do in the docs folder. Now that is that is me personally you can put this in anywhere you like maybe you want it all the docs folder or maybe you all want wherever um I'm gonna delete the newsletter recording video because we don't need that at all. So now I have things in sort of a sensible uh layout I might just for clarity um I'm gonna put uh my screenshots in a folder called Modic so that Claude understands oh this is from the Modic marketing automation system again the more clear I can be with labels the better the final thing we have to do now is we have to come up with a recipe we have to think about how how do we want to do this and so coming up with a recipe for this oh I forgot the agents whoops I should put some agents here I want as Chris as you're pulling that out I don't want I want to acknowledge that we are more than halfway into the show, and we haven't touched a large language model really at all. I mean, you started to use Claude uh to pull the browser data from Google Analytics, but in terms of what we're intending to do, we're still just getting set up, we're still doing requirements gathering, we're still getting all of those pieces.

And I think it's important because in order to have any system, it doesn't matter if it's a large language model or CRM or whatever, you have to do this work. This is the stuff you can't skip over. This is what is going to set you apart from everyone else generating mediocre things. Mm-hmm. And so our recipe now is going to be um what we put in as the master prompt in Claude Cowork for it to go and do its thing.

So I'm gonna bring up a new text window here. How do we do this? How do we put together a good Claude Cowork recipe? I'm gonna go out on a limb here and say, you know what's a great framework for this. Let's see how quickly I can find it.

Hold on. This one? Uh-huh. Got it. So we're gonna start by saying you are an email marketing expert.

Your task today is to help us build and refine a marketing strategy to grow the inbox insights newsletter from its current position of 33,000 subscribers to 40,000 subscribers by the end of calendar year 2026, which is 11 months away. So that is our purpose, right? Uh very clearly. In fact, I'm gonna even label these. So we're gonna go do purpose, people, process, platform, and performance.

And just because we are all old and can't see anything, uh, let's see if we can make this font a little bit bigger here. You mean my trifocals aren't doing the job? No, it's fine. There we go. All right.

So for people, um, inside of our folder, right? What do we what do we have? We have things like an ideal customer profile already provided, right? Or we have uh we should have, yep. So let me just compare.

In fact, let me close the side window here and bring it like this so that we can compare. Um to help accomplish this task, you're gonna need to focus on our ideal customer profile, which you can find in the input folder in mid-market icp.md. You'll also find some additional ideas for expanding our audience if appropriate, in, and this is where we'll put in the 5P analysis that Kelsey did, uh, which is this one here in the input folder. Okay, so that's the people. Any other people, Katie, that you would you would want to have for this recipe.

Um, would we want to have the baseline metrics of well? I guess that's more, I don't know. That's gonna be one of the things that I'm gonna do I guess performance is more of the baseline metrics. But no, in terms of people, I think no, I think that makes sense. I think we have the you know, here's our ICP, and then here's additional expansion ideas.

So let's figure out how this all works together. Exactly. So next process. You're going to review the docs folder to understand how we currently do our newsletter now. You'll then review the remainder of the files in the input folder.

The B2B newsletter growth strategy document gives you an assessment of the landscape and the patterns and anti-patterns to pay attention to for building a thriving newsletter. We're going to take from there, so we're going to say you'll also find analysis by Google Gemini of what our audience typically is searching for and related search terms that you can use for understanding how they think. Claude Code, which is what this is based on, knows how to read files. You'll review our existing performance in the data folder. There is a Google Analytics CSV file in that folder that shows you the web traffic from our email newsletter over the last year by day.

There is also a folder called MODIC, which is our marketing automation system, and you will see two screenshots in there of our newsletter's current performance. And then in platform, we're going to say you have two different agents that you can use to review your work throughout the process. One is called Co-CEO Katie, which is an avatar of our CEO and can provide the business sense needed for evaluating a decision. The other is an agent called VOC, which stands for Voice of the Customer. Combined with the mid-market ICP, the VOC will speak for the customer and help keep us grounded in what our customer actually cares about.

So those are those are the agent pieces. We're also going to say you also have a skill available in the skills folder called fact check. Use the fact check skill as you're evaluating reading documents to make sure that any assertions, claims, and statistics are valid. You'll first check the existing data, and then you can use your web search tools with a sub agent to do additional validation. All right, and finally, the last part of the recipe is say what we want as the output.

Like what is it, what is the output look like. Once you've gone through the entire preparatory process, you're going to think through what is the Trust Insights inbox insights newsletter strategy today. How well is it working? Do a full QA assessment. What's good about it, if anything, what's bad about it, if anything, what's missing from it, if anything, what's unnecessary, if anything.

And then, based on those evaluations, go ahead and start building a refined, unified strategy, tactics, execution, and measurement plan that will get us from 33,000 subscribers to 40,000 subscribers. So that's our performance plan, is what what Claudia's going to do. So this is our recipe for co-work. Any questions so far, Katie? Yes.

I know. Surprising. Overview. But one of the things that we do is we offer the newsletter in a few different formats. So we obviously we have the modic email.

Uh, I record it for YouTube, we turn that into a podcast. We also put it up on LinkedIn as a newsletter. But then there's also the dissemination strategy once the new newsletter is completed, like, you know, we want it to go up on LinkedIn, not just as the newsletter, but hey, did you know we have a newsletter? You know, Kelsey's been now putting it up on Instagram. Hey, did you know we have a newsletter?

And then part of her recommendation is that we put together uh we expand our social footprint into other platforms. Hey, did you know we have a newsletter? And so I'm assuming all of that's included, but because I didn't actually see it, I just needed that uh validation. You are correct. And that is in the 10 minutes of foaming at the mouth I did.

Um I even talk about what UTM tracking codes we use. Um, we have you know sporadic evergreen posts, promotional tools, how we do all the stuff is in that big file. Because to your point, we want to give Claude or any LLM as much information as possible. Now there's two ways we can go ahead and do this. We can either just copy and paste this and hit go, and that is perfectly fine, or um one of the other things that we can do is we can create this as a recipe file.

Um, and the recipe file just goes into the folder. Now you might say, well, why? Why do you need a recipe file if you've if you've already built uh the prompt? Because I forget things. And so if I have the recipe file written down, then the next time I do this, when Katie says, hey, it's Q2, how about we reevaluate our strategy and go, what did I say?

Well, it's also really just a best practice for scalability is if you do things in a more consistent way and repeatable way, you know. One of the things we know about LLMs is they're really good at those consistent repeated tasks, which help you be scalable. And so I could imagine, you know, you could probably hook this up to like an N8N or some other agenc and be like, kick this off, you know, once a month and evaluate what the heck we're doing. And you don't even have to be there. You just suddenly get this like file delivered to you of like, here's how your newsletter's doing, here's what you're doing, here's what's not working, here's some new ideas.

Exactly. So I dragged in the recipe file, and Claude says, Hey, I noticed you use the uh 5P framework, which again, love that. Um, it says this is a standard substantial project involving research and analysis and strategy development. How would you like me to approach it? Full deep dive phase approach or start with assessment.

I say, let's full deep dive. Let's go all in. Go for it. Um as John would say, what could go wrong? Um so what you can see now is it's already putting together its to-do list here.

Uh and it's got a to-do list of 12 different items. And you can see it's got a scratch pad. This is where it takes notes uh as it goes through this process. It says, and you can uh see each of the steps it's doing, reviewing the files, reviewing key documents and parallel. It has now launched multiple agents to start reading, and it's you can see it's assembling its notes on each of the files as it reads through them.

And is that all stuff that we as humans can then access to see the notes that it's taking to start to learn how it's quote unquote thinking about it? Yes, you'll be able to get that once it's done. Sure. Yeah. I don't want you to start pressing buttons and exploding it.

I actually want the results of this analysis. This by the way, this might take some time. Um just just fair warning. But what it's doing now. So the two it's it's done the to-do list.

It's now looking at the modic screenshot files. Uh let's see. Older campaigns, oh 2026, email sent status. Let's check the fact check skill, then proceed with the analysis. And so now it's reading our skills.

It got to a point where it was like, I don't have high confidence in the information you provided. I need to do it, go and do some fact checking, um, which is pretty darn cool. It's it says I'm gonna run four different agents. I'm gonna fact check your your research, which okay. But also helpful because we're humans and we make mistakes.

Oh, yeah. We're also uh we're also biased humans to be all of us, collectively, John especially, um, to what we think is the right answer. So, you know, using tools like this, I feel like is a good way to pull yourself out of it and not say my agenda is this. It's here's what's best for the voice of customer. That's something I struggle with.

And this is why I really enjoy these tools, because it's allowed me to take my own personal here's what I think we should do out of it. I let the data drive a lot of those decisions. Yep. And you can see as each of these agents run, they have their own decomposable task list. So you can say in the B2B strategy claims, it has kicked off six different web searches to try and figure out if what is in that deep research document from Google is actually true or not.

Um it was not for this project, for a different project. I had Claude, I was using Claude code, and I had it use the fact check skill on a piece of deep research. Claude came back and said 27% of this document is fabricated from a Gemini Deep Research thing. It went out, it checked Gemini's sources, which are in the document. It read through each of Gemini sources and said 27% of this deep research is fabricated.

And I'm like, awkward. I like how Claude is throwing shade at Gemini. You know, when the LLMs start fighting, you know that you've completely lost control and you've entered into John Wall's best slash worst nightmare. This is it, virtual virtual war. They're actually punching each other behind the screen here.

Exactly. Uh so let's see what it's done so far. It's gotten through uh these items not to do list. It it looks like it's consulted co-CEO Katie already. Um, and now it is on the comprehensive growth strategy part to of the plan, which is step 11 in our plan.

So while this is finishing up running, and then we can go over the results. Let's talk for a second about cost. So using these kinds of tools for the most part are not free. They cost you something, they cost the company something. Can you just walk through, Chris, at a high level?

If someone were to watch this live stream today, go download, you know, Claude Desktop, what kind of expenses should they expect? Like what can they do for free? And then where do the costs kick in? So Claude, co-work and code all bill against your weekly allotment of uh server usage at Anthropic. So Anthropic has uh like even a little meter, if you will, of um of your usage of its software to say, like, hey, this is this is how much um you've used.

So if I I'm gonna pull, I don't want to touch what's in uh uh co-work. Um so I'm gonna open up Claude Code, but the same thing is available inside your your Claude desktop application. You can type in usage and we'll show you. For this week, I've used 60% of the compute budget that Anthropic has allowed us, right? Just since we started, I've used 5% because this reset at noon.

So I've used five the Claude gives you a five-hour window of how much budget you're allowed to use within any given five-hour period, and then a weekly overall budget. So just on what we've done so far, I've already used 5% of it, and we haven't gotten a result yet. This is on the Max 20 plan, which is the most expensive plan. If we were using Claude Pro, which is the baseline entry level plan, we would basically be at the five-hour usage mark now. So you would be able to get through one of these every five hours of what we've done so far.

Um of course, it says you can add a credit card and we'll bill you API costs if you go if you need to to you know desperately need to finish a project. I have not done that. Thank you. I have I have uh read through many complaints on Reddit and other places where a 30-minute session inside Clawed Code can cost you $5,000. Oh I mean, first, I would figuratively slaughter you.

And second, I would be begging John to find us more business to cover those costs. Many a tech company crushed by overage. This is a recurring theme. But I think that, you know, it's definitely worth acknowledging because they don't have to be, but these systems aren't upfront about what that looks like. They're just like, oh, by the way, you did the thing, here's the bill, versus, hey, you're coming up on your usage.

Do you want to like maybe pause here? It's like, no, you go ahead, you do all the things. John, I'm gonna sucker you into this whole conversation about how you're gonna include your sales and you're gonna improve everything. And oh, you've been at this for a few hours. Great, here's your bill.

Careful. Like that's pretty much how it's so. I feel like it's worth acknowledging. You know, it's not a buyer beware, it's a buyer, keep your eyes open and make sure you, you know, again, use the free version to like test out all the features and push the buttons and do the things. But to do this kind of work, you want to really be thoughtful about the projects that you're launching so that you don't, you know, basically, you know, you don't have me reaching out to you, tagging you in everything with a WTF.

What happened? Yep. Because those of us who control the purse strings do not like surprises. So Claude has finished after about five minutes of thinking this through. It says, Here's all the things that I work through.

And of course, with any of these things, you can tap through and then see what logic it it uh went through. It's data analysis, it highlighted some fact-checking results, like some things that are true and are not true, which agents it consulted. And ultimately, what it spit out was a word document. Thanks, Claude. I should have asked.

Please put this in markdown, but you know what? Here we are. Um so inside our strategy document now. We have an exact summary of the core challenge, strategic thesis, retention is the new acquisition and the zero budget environment with the team before. The most efficient path to 7,000 subscribers is activated the existing 33,000 as a distribution channel while simultaneously fixing discovery gaps and operational efficiencies.

Of course, um it said, for example, um in the D in the Gemini trace, which I looked at, um there were some quotes from me in Zapier's email marketing blog that made in, but not trust insights. So it's like, hey, you need to follow the recommendations to make sure trust insights is creating more content about the agentic AI things that that people are asking about. Uh I didn't specify a list hygiene policies. Well, we do have one, which I just did not narrate it. So our next revision would be for me to start narrating.

Oh, by the way, I forgot to tell you, we send this out on Wednesdays. Of course, this is gonna be a Wednesday spike, duh. Um, and here's what our list hygiene policy looks like. Here's how we clean our list. Um execution layer monthly, partnership swaps.

Um, so and that goes through here's your current state assessment, your competitive landscape, um, your QA assessment. What's good? What's bad? What's uh missing, referral program, no referral program, execution of layer content, templates, workflows, how-tos. One of the big things they highlighted in both the um Gemini trace as well as the best practices is a lot of the people who are talking about the same things we are as much as it annoys me and corrodes my soul.

Um they're like, hey, copy paste this prompt for instant chat GPT results, right? So they're suggesting have more giveaway content in the content itself that people can take action on immediately. Um execution layer content, progressive profiling. Uh so we forgot to tell we did we do surveys, um, so that uh is something we would need to revise. Partnerships, AI content optimized structure, open rate obsession, get over it.

I wouldn't say that I'm obsessed. I think Claude's putting words in my mouth. In there that is coming from the screenshots of Mautic, where open rate is the most featured message. Oh, sure. And there.

Um 48 hour LinkedIn delay. Uh, that was one of the recommendations from the deep research was make sure your email newsletter goes up before two days before it goes out on LinkedIn so that the subscribers have an incentive. And it's saying that doesn't seem like it's necessary. So it was fact-checking the deep research to go, nah, that that doesn't make any sense. And then trying to be everything to everyone, don't do that.

Voice of the customer, reality check, right? So the customer response. Um prompt libraries, webinar access, things like that. Cool. Um warning signs, more than one email per week, unsubscribe trigger.

Yep, which we know. Um, data content creep feels like a bait and switch. That's from the here's the free webinars you have to register from in the content of the newsletter itself. Uh, interactive polls feels like engagement theater that's benefits the sender, not the reader. That's fair.

We um guilt trip re-engagement emails, guarantee unsubscribes. That's from the deep research. It goes through the math, and then okay, here's your execution plan, list hygiene, audit everywhere I'm found and make sure that uh trust insights is in my bio everywhere it's found. We already do a pretty good job of that. We need to uh make sure that we can do that with them some search.

Do a referral program for inside the newsletter itself. Do some uh deliverables within the newsletter itself. So every month have something that people can get that they can only get from the newsletter. So you have to say stay subscribed to get the uh AI marketing prompt library or the uh mid-marketing mid-market marketing budget template, uh, and so on and so forth. And then the partnership stuff as well.

And then scaling it up. Um activate or grow upulse for your Evergreen newsletter uh posts, test substacks indication, automate jobless sourcing uh as an efficiency. And so those are the its tactical things. So this is what it came up with after you know a decent amount of sharing. There's some revisions that need to be made, and there's some additional feedback we need to give Claude to say, oh, by the way, I forgot to tell you this, this, this, and this, I forgot to do this, and then you modify the recipe and say, Claude, run the recipe again.

And that's the beauty of Claude Cowork is instead of having to redo a prompt chain, don't say, I can literally just give it the revised recipe, hit go, walk away, and come back in 15 minutes, and I've got a V2 of the strategy. And I know we went through it quickly, but I think I'm a little underwhelmed because I and maybe this goes into the revised recipe. One of the things that I'm looking for is where do we get those new subscribers from? So it's saying re-engagement or you know, keeping the existing subscriber base. That's great.

But in terms of, you know, we defer to LinkedIn because it works really well for us. But like, are we supposed to be on other social channels saying, hey, did you know we have a newsletter that talks about these things? I feel like that recommendation, those tactics. Maybe I wasn't clear about what I was looking for from the performance, and we can revise that recipe. But I've and again, it's my own bias, and I'm acknowledging it.

I feel like the newsletter itself, the content is solid. It's because we get that feedback from readers every week of like, hey, this was great, hey, this was helpful, that kind of thing. But like, where do we let new audiences know, hey, we have this thing, it's established, people like it. So you would absolutely include that as guidance to say the acquisition side needs to be beefier. Like you would you would uh revise that recipe.

And again, that is the beauty of Claude Cowork, because I can say, change that recipe, rerun, and I'll come back when you're done, right? Or I provided another recording. So now there's three recordings in this folder here of stuff. Go do it and I'll I'll come back when it's done. And so it now acts as a true agent that goes off and does something and comes back.

I don't have to you notice when it was running, at no point other than giving the permissions for the folder initially, did I have to go back and babysit this thing? It would just kind of off doing its thing, and I can be doing other stuff like this live stream. That's why that's where this is valuable. Now imagine starting to take it and say, okay, well, great. Now that you, you know, maybe you have an execution plan for that evergreen scheduling.

You might say, okay, I agree with you. I need some evergreen post content. I'll log into my Agora Pulse account and Claude, here's you're gonna schedule it for these days and times. Here's the files in the folder. Here's the recipe.

Go off and schedule this for me in Agora Pulse. And then you walk away and and you know, come back and it's done. Now you should check its work and maybe the first couple times watch it carefully to say, like, you know, Claude, that's the wrong, that's you know, that's not the right account to be doing this in. But you can see how it leads to, I'm going to take away this task that requires no executive function, no planning organization um decisions of problem solving, you know, scheduling social content does not require that. Uh, I'm gonna get this execution task and get it off my belief.

Um, so one of the questions that I had, and I asked this again on the podcast, Trust Insights are AI slash TI podcast. Why would someone use this instead of the web version of Anthropic Claude the chat and basically just upload all of those files, give it the same recipe you gave it? Like, why would someone want to do this desktop version versus the web version? Um, obviously they're using chat and not co-work. So, you know, can you just at a high level just walk through why one is better than the other or they're the same or whatever?

It does a very good question. So there's two reasons that you would do this. Number one, you want this autonomous agent-driven version, the regular chat doesn't have that. You'll notice when at if we scroll back up here, you can see the times that it kicked off multiple sub steps, like doing all these web searches by itself, you running the co CEO by itself. Because it was in the recipe, I didn't have to do that.

If you do that in the desktop version in the regular chat, you have to manually do each of these steps. Now you can, and that's totally fine if you're a person who wants to just do that copy paste it manually. However, the other thing, this is the big thing. Every time one of these subprocesses run, it runs in its own short-term memory. All AI has what's called context windows.

Claud's is 200,000 K, which is about 150,000 words. In this folder, we have so much content and and and it's there's such a large data set that if you did this in one chat, Claude would give you the I'm out of memory about halfway through. Because it would not be able to keep in memory all these documents. When these agents run, they run and they come back with a short summary of what they did. And that's what that allows the orchestration side of things to keep things on the rails because it doesn't have to try and keep everything in memory all the time.

You can see in its scratch pad and its working folder, it knows where those documents are and it can search within them and get references, but it doesn't have to try and remember everything all at once. Gotcha. So it's it's really just a matter of your working style. If you want to give it piece by piece and sort of refine things as you go using the web version of chat is totally fine. But if you have all of the pieces, if you did sort of what we did and get everything together, get all the data, all the requirements, get the recipe, and then just let it go run in the background, then co-work, the desktop version of Claude is a better option, but they both connect to the same account.

So this is our client services account. So would I be able to see that this data in the web version? Uh not on the web. You'd be able to see it in your desktop version of uh Claude, but not on the web because co-work does not exist on the web. But you can see the chats.

You can see the chats were not showing on this live stream for uh obvious reasons. Um because you know, John does very private conversations within our Claude chat. Yeah. No secrets. The other thing is repeatability.

Yeah, when you use the chat, you would kind of have to repeat that process. If you do it the way we did it here with that recipe, just start a new chat with a new recipe at the FQ2 and say, Claude, rebuild our our, you know, here's the new data, put the new data in those folders, rebuild our email marketing strategy for this quarter and walk away, come back after lunch, and it's ready for you to review. And so there's an aspect of reusability with these agents that is fantastic that you don't get in regular chat. So I feel like this is a particular product that we could probably go on and on about, do different use cases. But I do want to acknowledge we're coming up on time.

The viewers have been with us this entire time, which we appreciate because it sounds like it's probably a really interesting topic to them as well. So our next step internally is to take what was the output from this particular instance, refine it because it's not ready for us to use yet. We still have more work to do. But what we would recommend is people, you know, watching this, trying to figure out how do I automate more things. This is a great option if your company has access to Claude.

Exactly. And you have a Mac to run the desktop on for today. In the next couple of weeks, hopefully the Windows version will be out. And then I would say if you have someone who's much more technical, they should run full on clawed code. Claude cowork is for people who don't want to touch coding stuff, totally fine.

It gives them some of the capabilities of clawed code with none of the cognitive burden. So and I know people in our Slack community, Analytics Marcus have asked us for a clawed code version of a walkthrough, and maybe that's a topic for another live stream. But uh, yes, definitely go ahead and get started. The one last thing I will say is you doing the clawed cowork version that we did today, I can then put all those documents, the inputs and the outputs inside Google Drive, and everyone else can work on them at the same time, which again is is can be done with the chat, but it's a lot easier when it's all in the file system. So, John, man of many words, what questions or thoughts do you have about what we just did?

Yeah, I mean, it's you know early stage for this thing, but the thing is, you know, it does have 10 possible things to try, and that's really the end result, right? Is you need to just look at that list and say, okay, we're gonna take a shot at three of these and put them out there and just test them and see how it works. That's the and a lot of times coming up with that list of like which way do we go next, that's the hardest part because you kind of have no idea where it's gonna go. So, at least if you have some direction that's based on some kind of research, you know, you're you're saving yourself a lot of time and you're way better off with your testing than just you know what Randy the intern came up with last week. On that note, I think we're gonna see you all next time.

Thanks everyone. Thanks for watching today. Be sure to subscribe to our show wherever you're watching it. For more resources and to learn more, check out the Trust Insights Podcast at TrustInsights.ai slash TI podcast and a weekly email newsletter at TrustInsights.ai slash newsletter. Got questions about what you saw in today's episode?

Join our free Analytics for Marketers Slack group at TrustInsights.ai slash analytics for marketers. See you next time.