Podcast: Play in new window | Download
What if AI didn’t just assist your work – but actually did the work for you? AI trainer, Eliot Prince, breaks down how Agentic AI is moving beyond prompts and into real execution, helping business owners automate tasks, streamline workflows, and get meaningful results faster.
What Agentic AI Looks Like in a Real Marketing Workflow
There’s a moment that happens with every new technology where you stop reading about it and start using it. For me, that moment with agentic AI came when I pointed Claude Cowork at my downloads folder, which contained tens of thousands of files, and told it to organize everything. An hour later, it had sorted everything into labeled folders and even flagged which files I could probably delete.
No code. No technical setup. Just a conversation and a result.
That experience sent me down the path that led to my recent conversation with Eliot Prince on the Agents of Change Podcast. Eliot is a UK-based AI trainer and curriculum developer who helps business owners use AI practically and systematically. He’s also the person whose YouTube video gave me the confidence to try agentic AI in the first place.
What we covered in our conversation went well beyond file organization. Eliot walked through three specific use cases that show what agentic AI can do for marketers and small business owners right now.
What Is Agentic AI, Exactly?
Before we get into the use cases, it’s worth clarifying what “agentic AI” means in practice. Traditional AI chat tools like ChatGPT or standard Claude are conversational. You ask a question, you get an answer. Agentic AI goes further.
As Eliot explained, Claude Cowork pairs the familiar chat interface with the coding capabilities of Claude Code running under the hood. That means it can write scripts, scrape websites, create files, navigate your browser, and solve problems on its own, all without you writing a single line of code.
Anthropic built Cowork after noticing that people were using Claude Code not to build software, but to automate their everyday business processes. So they wrapped that power in a user-friendly interface. The result is a tool that can act on your behalf, not just answer your questions.
Use Case #1: The $5,000 PR Study for Under $10
This first use case is the kind of project most small businesses would never attempt due to cost. Eliot described building a full PR study, the kind a PR firm might charge $5,000 or more to produce.
Drawing on his SEO agency background, Eliot knows that getting press coverage requires newsworthy data and compelling angles. Traditionally, that means hiring a firm to conduct surveys, analyze data, write the report, and craft press releases.
Here’s what he did with Cowork instead: he gave it a directive to scrape over a thousand job listings from LinkedIn, analyze the data, create charts and spreadsheets, write a full report, and then produce a press release in his brand voice. Three prompts. About an hour of work. Under $10 in AI costs.
The important caveat, and Eliot was emphatic about this, is that you still need to check the work. He referenced a major accounting firm that got in trouble with the Australian government for publishing AI-generated content with hallucinated citations. His approach: treat the output like work from a talented intern. Let it do the heavy lifting, then verify before anything goes out with your name on it.
Use Case #2: An AI-Powered Social Media Content Machine
The second use case Eliot walked through was using Claude Cowork as a social media content creator. This is where the concept of “training” Claude on your brand gets interesting.
There are two layers to it. The first is tone of voice. You feed Claude examples of your writing (blog posts, LinkedIn posts, transcripts, website copy) and ask it to analyze how you write. Eliot said Claude came back and identified that he writes with “dry, British wit humor and cynicism,” something he never explicitly told it. You then refine those observations, adding rules about what you want and don’t want.
The second layer is visual branding. In Claude, you can set up what are called “skills,” which are repeatable processes with custom instructions and supporting files. Upload your logos, fonts, brand colors, and brand guidelines. Claude will then apply those consistently to anything it produces, whether that’s a LinkedIn carousel, an invoice, or a social media graphic.
Eliot’s workflow: hand Claude a transcript, let it find the best angles and hooks, write posts in his tone of voice for each platform, and then generate branded PDF carousels for LinkedIn. No Canva, no Word documents, no back-and-forth with a designer.
And here’s where it gets even more powerful. Cowork has a scheduling feature. You can set it to run every Monday at 9:00 AM, for example, to pull in trending topics, cross-reference them with your recent content, and generate a week’s worth of social posts. As Eliot pointed out, that replaces the complex Make.com or Zapier workflows that people spend more time building than using.
Use Case #3: A Full Technical SEO Audit
Eliot ran a technical SEO audit on a website using Cowork, and described what happened.
He gave Claude a website URL and a simple prompt: “Can you do a technical SEO audit on this website?” Claude started browsing the site, checking HTML, testing UX elements to see if buttons worked, and pulling page data. Then, without being asked, it navigated to Google Page Speed Insights, ran the site through that tool, and pulled the results into the report.
It also connected to Google Search Console to pull additional data. Eliot noted that these are tasks he used to do manually, bouncing between tools that cost $150 or more per month.
But here’s the part that really shifts the equation: after identifying pages that hadn’t been updated since 2022, Cowork could research current information and draft page rewrites in his brand voice. That’s not just auditing. That’s doing the deliverables. As Eliot put it, that could represent months of SEO work, ready for review by morning.
The Expertise Multiplier Effect
The thread running through all three use cases is that your expertise matters more with AI, not less. When I asked Eliot whether someone without an SEO background would get the same results, his answer was clear: the best AI users aren’t treating it like a search engine. They’re feeding it their frameworks, their SOPs, their templates, and their professional judgment.
AI can take you from beginner to intermediate without much effort. But getting to expert-level output requires the kind of knowledge that only comes from experience. Knowing what good SEO work looks like versus bad SEO work, knowing which data points make a compelling PR angle, knowing how your audience responds on LinkedIn versus Instagram. That’s the human layer that makes AI output genuinely useful rather than just well-formatted.
As Eliot put it, it’s the difference between someone who has read all the marketing books and someone who has been in the field for 30 years. The books give you theory. Experience tells you what actually matters.
Getting Started: Simpler Than You Think
If you’re curious about trying Claude Cowork, Eliot’s advice is simple: download the Claude desktop app (you’ll need a paid plan). Switch to Cowork mode using the toggle at the top. Point it at a folder on your computer and give it something simple to do: organize your screenshots, rename files, sort a messy downloads folder.
That small first step teaches you how the tool thinks, how it communicates, and how it solves problems. From there, you can start handing it bigger tasks: brand training, content creation, audits, research.
The investment of time up front is real. Just like training a new employee, you’ll need to iterate, give feedback, and refine your instructions. But once it’s dialed in, the leverage is significant.
Your Next Move
Here’s what I’d suggest if this episode resonated with you:
- Download the Claude desktop app and open Cowork mode.
- Start with one simple file organization task to learn how it works.
- Feed it examples of your writing and ask it to analyze your voice and style.
- Pick one recurring marketing task (social content, reporting, audits) and build a workflow around it.
- Check out Eliot Prince’s YouTube channel and weekly newsletter for step-by-step AI recipes that walk you through the setup.
The gap between “that sounds cool” and “I’m using it” is smaller than you think. And the people getting the most out of agentic AI are the ones who brought their own expertise to the table and let the tool do the heavy lifting.
Transcript from Eliot Prince’s Episode
Rich: My next guest is a UK-based AI trainer and curriculum developer who helps professionals and businesses owners use AI practically and systematically. He’s trained hundreds of business professionals across industries. He gives away full AI systems prompts and step-by-step recipes through his YouTube channel, which is how I discovered him, and his weekly newsletter.
Today we’re talking real-world, pressure tested Agentic AI with Eliot Prince. Eliot, welcome to the podcast.
Eliot: Hey, Rich. That was such a good intro. I don’t think I’ve ever had a fully perfectly bio intro like that. That was a blessing.
Rich: Well, you know, I always like to start with the facts and then throw in my own stuff. Because I did discover you. I was doing some research into Claude and Claude Cowork, and your video popped up on my YouTube channel, and I watched it and I thought it was brilliant. You actually gave me the confidence to do my first Agentic AI work, so thank you for that.
And then you learned about me because I talked about it on LinkedIn, and I tagged you in it because I wanted to give you credit for it. And that’s where this beautiful podcasting journey began.
Eliot: Yeah. I owe you another thank you then for shouting about me on LinkedIn. Which, I don’t go on there a whole lot these days, but when I do, it’s nice to get a little surprise in there.
I’m interested to know, what was the first thing you tried out with the Coworker gen stuff?
Rich: There were three things that I attempted. And I don’t know for sure which was the first one, but two popped to mind.
One was, I had it go through everything in my downloads folder, which was literally tens of thousands of documents, and it organized it into about eight different folders for me. And then I’ve been meticulously going through it in small batches to get rid of these things that were either very old or duplicates and all this sort of stuff.
And the AI even had some really good rules about like, this is probably stuff you can delete. This is stuff that’s related to your podcast. So it helped me prioritize on where I wanted to spend my time, and that was just a really nice use case.
And then I took the idea of renaming all the screenshots that are on your computer. And instead, I have all these Adobe stock photos on my computer. And so I had Claude Cowork go through all of them, determine what the image was, then rename them based on what was in the image as opposed to Adobe stock 17 different digits jpeg. So both of those were really helpful.
Eliot: It’s funny that, I guess two types of people that when I show them those really simple ways of using this stuff and people are like, “Great, we are using AI to rename screenshots.” And they’re not very happy with me for showing such a simple use of AI.
And then there’s the other people which are probably more busy and have businesses to run and they’re like, yeah, I would never sit down and organize my screenshots, but I know there’s some really useful stuff in there. So, thank you.
Rich: Yeah, absolutely. And that was my feeling. It had gotten to the point with Adobe where I had over a thousand stock images on my computer, and I knew full well I was never going to go through all of that. So it was something I wanted to do, but I wouldn’t have even handed off to an intern because it seemed like cruel and unusual punishment. But Cowork had no issues with it. So there we go.
Eliot: That’s a nice way to punish a staff member then, isn’t it? You’re like, do something.
Rich: So Elliot, I know that you started or at least were for a long time, an SEO guy. So what led you to change your focus from SEO to AI?
Eliot: To be honest with you, it sounds hyperbolic, but I remember thinking this the day that ChatGPT came out and I tried it, because I tried so-called “AI things” for writing and trying to make my, like, obviously SEO has a high volume content game a lot of the time, so how can we speed this up, how can I give it a transcript and get it to do stuff just didn’t work.
The moment I started using ChatGPT I was like, ooh, what I do today isn’t going to exist in even a couple of years. Like even now, a lot of the work that I would have been doing or do can be handled by AI pretty spectacularly. And that’s only something that’s happened over the last few months. So there were two things.
I was like, okay, how do I make my life easier? What processes can I run through ChatGPT? At that time, that meant I didn’t have to spend a week writing my reports for clients, which was just basically framing data and presenting it in a nice way. Or the other part of that is SEO and AI are so interlinked because people are using it to surface information, and that means I need my clients to show up when someone’s looking for that piece of information or a solution to their problem.
So there’s two ends there. Using it for my internal processes and figuring out how this stuff kind of presents information under the hood to work it as a marketer.
Rich: All right. Now we mentioned that we’ve got some examples of ways that we can use Agentic AI. But before we kind of dive into them, you’re at a cocktail party, somebody hears something about Agentic AI. How do you explain what Agentic AI is to somebody who’s not in this world?
Eliot: Oh, that’s a good question. Without being able to sit down and show them, I would say that I am able to give Claude, or ChatGPT for some people, just a transcript of a call like this, and just say, go, and it will go through that transcript. It will find the best angles for content or marketing or sales, or for whatever you need it to use. It will then take that. It will be able to write. That idea in my brand voice, using my exact expressions, the way I write and format stuff. That stuff tailored to a particular algorithm for LinkedIn or Facebook.
If you’re still using Facebook and even produce a fully branded LinkedIn carousel with my logo in my brand colors, in my brand font, ready for me to use. So it’s essentially acting as a whole social media content designer, just ready to go. All I have to do is give it a transcript and say, ‘go’.
Rich: And that’s just one example, and we’ll be kind of diving into that today. So yeah, I discovered you, we’re going to be talking about Claude Cowork, primarily. If people haven’t used it or they only know Claude as an LLM, how does Claude Cowork differ?
Eliot: Yeah, so the traditional area of Claude is similar in the look and feel and the features as ChatGPT. It’s sort of a traditional LLM chat mode. Now, Claude Cowork has only been around about six weeks now, so it’s still very new.
But what Anthropic have done is they’ve paired the fundamental, enjoyable UI of using a chat LLM with the power of something called Claude Code. And Claude Code is the AI that’s really great at coding. It’s the best one out there of all the different models. And what was surprising about the way people started using Claude Code was they’re not actually starting to use it to build apps and software. They’re actually using it to write scripts and automate processes to make their lives easier, to automate stuff in their life and business, rather than building actual software like you think a software developer would.
Philanthropic saw this, or the makers of Claude saw this, and they said, ah, we’ll take that under the hood, this ability to basically write code and scripts to do anything you want, and we’ll pair that with the enjoyable user and sort of the way that people are used to interacting with these LLM Chatbots. We are going to push it all together so that you’re no longer constrained by the file types that ChatGPT or Claude is allowed to put out or the way they’ve developed it, so you can tell it what you want. It’s able to actually write code under the hood without you even seeing it to automate or do whatever you want to do in your business processes or your marketing processes.
Rich: Alright. And from my experience, to use Claude Cowork you do need to have a paid version and you do need to download the desktop app. But then once you’ve done that, there are three tabs at the top. There’s the code, which I haven’t touched yet. There’s Cowork, which we’ll be talking about today. And then there’s just Simple Claude, which is like that chat bot that we’re used to working with.
Now you gave me a bunch of different things that you’ve already done in Claude Cowork, and so I just want to kind of go through a few of the ones that popped out at me. And the first one was a $5,000 PR expert, I think is the way that you phrased it. And you described building this full PR study with data charts, report, even a press release all for under $10. So what does it actually look like step by step? Can you walk us through that process?
Eliot: So this is something that throws back to my SEO agency days. Because when you’re doing SEO, you need to get backlinks, you need to get mentions around the web. The best way to do that is getting featured in newspapers or the in… what would you call the internet version of a newspaper now? A news website being talked about around the web, on social media. So you want to create newsworthy angles.
So one of the ways to do that is you run research and studies to find interesting data points that make headlines. Now usually, even at an SEO agency, we would outsource that to a PR firm to go and actually spend a long time either conducting surveys, doing the research, getting data together, analyzing the data, and then actually sitting down to work out what does this all mean? And then writing the PR stories and angles and hooks and everything like that.
This is thousands of pound, thousands of dollar’s worth of work easily. You could have people charging $5,000 for one of these reports and dissemination of everything. Because there’s quite a lot that goes into that when you think about it.
Now with Cowork, because it’s able to write Python scripts and everything under the hood, and I’m able to connect to, in this example, I wanted to analyze over a thousand job listings in the UK, and I wanted to pull all that, scrape all that information off LinkedIn. I wanted to analyze all the data, and I wanted to write my PR report and then my press release, and distribute it within my brand, like guidelines and everything.
Now, it took me an hour to actually send Cowork off to say, “I want to scrape LinkedIn”. It wrote scripts, it went and scraped LinkedIn for me. It ran into some problems. It was able to problem solve and fix it, and then get all that data, scrape over a thousand job descriptions. Which is a mega task in itself to actually do that task. And then it was able to analyze, create, put all the data into spreadsheets, create charts, create all the takeaways and analysis. So we’ve got the full report written. Or the full outline of the study.
And then because I’ve got it set up with my brand guidelines, which I was just saying, my logos, my fonts, plus it knows how I like to write it, was then able to take all that information in that report and produce a press release, all through, I would say, about three prompts.
Rich: Wow. On a project like that with all that data, is there a way that we know it’s reliable, or is it just in this particular use case, because it was job descriptions that appeared on LinkedIn, we could trust the data that came in?
Eliot: I mean, it depends on your industry and your reputation. If you’re not too nervous about being called out and blowing your reputation or getting something wrong like that, then have at it.
But honestly. We treat any of this AI stuff like an intern, or if I’ve got a group of interns running this stuff. I’m not going to let it let an intern send out a branded report and press release without having me check the information, because it’s my reputation on the line.
So when you see a lot of these new headlines, news headlines. And I saw, I think it was, I don’t want to mention the brand, but one of the big accounting firms got in trouble with the Australian government because they had links or citations that were just to dead pages and hallucinations and stuff, because they didn’t do their due diligence and check it. So yeah, make sure you’re managing your AI properly and doing your due diligence with it.
But because it pulls all the data down and it gives you the exact job descriptions, you can go and run a command F and say, “search for that particular term that it said only had mentioned five times”. I can check that really easily. So you should be doing your due diligence on it.
Rich: And so it’s like, even though we’ve got Claude Cowork saving us reams of work, that there is still some human judgment and oversight that you at least would put into this before you’re willing to put your name and send it out into the world.
Eliot: Yeah. Like I’m not having it automate and send off a press release without me checking what that actually means. I think we still need, well, we definitely need still that human taste, that human judgment, that guidance on what’s coming out.
In time, once you learn to write the right guardrails and start to trust what it can and can’t do, and working with certain elements, then sure, you can be a bit more confident that stuff’s coming out.
But again, it very much depends on what that is. If it’s internal communication in a company where you’re like, well, it’s not that important if we don’t quite get the number exactly right. If it’s a court filing or press release or client work, then hey, let’s double check what’s going on here.
Rich: Absolutely. If somebody listening thought this was brilliant, and I’m sure there are people out there who do, and they wanted to try it on their own, what’s one or two things that might trip them up? What are some of the places where there may be more friction than you might expect?
Eliot: I’d say Windows or PC users might run into a few more errors, because Cowork is a newer platform and it was sort of developed more for Mac. But it does work on PC, but I see a lot of errors coming that way.
The other places are, I think for me it’s still figuring out what it can’t do. Because actually Claude Code behind the hood is actually good at figuring stuff out. So I don’t run into too many issues that I can’t figure, unless it’s something crazy complicated.
And I’ll give you an example. I was trying to go back to the scraping example of that again, because it was using an API through another connector, connecting to another scraping platform that was doing it. I said, “This is the most efficient way to do it. Can you do it like that please” And it went off and did it, but for whatever reason, it just wasn’t working. There were rate limits or things like that.
Now, within Claude Cowork and the Claude infrastructure, they have things like Claude in Chrome, which is an agentic web browsing feature. So Cowork said, “Okay, I can’t actually access the API here, I’ve got a work around. I’m going to go and use the Claude and Chrome extension, and I’m just going to browse the web using your web browser to find the information it needs.”
So actually, you’ll find that you don’t have to be super techy. You don’t have to be a crazy developer or something. You can communicate conversationally and figure out a lot of problems as you go, would be really what I’ve found.
Rich: And I would agree. Because I definitely am not a coder. But I do remember, actually before I found your video, I did try one thing and it was to scrape all the articles off my dad’s website so that we could do a test in Notebook LM of a chat bot where somebody could talk to all of his articles at once, hundreds of articles. And he gave me the code and told me what to do, so that I could basically do all of that even though I had no experience at all.
So I do find that even when I reach a point where I don’t know what I’m doing, Cowork can often talk me through the next steps anyway.
Eliot: I mean, that’s the secret hack a lot of the time. I get a lot of questions, and I was like, did you not ask the AI what to do next?
Rich: Right, exactly. Yeah.
Eliot: But I think the other big thing that people might find frustrating is it doesn’t work out the box. You do need to put some time into actually, again, treating it like an intern. If you’ve got an intern that’s whip smart, it’s got all the book knowledge, it knows everything in the world and you plug it into your business, it’s not got that context or those operating procedures of how you like to operate.
So you might find there is a steeper curve of getting it set up and getting it working how you want it to work, and learning all those little nuances that take a bit of iterating training. I didn’t get it quite right. Okay, let’s go back. This is how I want you to do it next time. That’s the big learning curve, I think there.
Rich: Yeah. I think just like with training an intern or a new employee, there is an investment of time if you want it done right. But especially if you’re getting it to do something it’s going to do several times, it absolutely speeds things up.
Now, the second one that I wanted to talk about is the social media manager. So you talked about using Claude Cowork as a social media manager, and you trained Claude on your brand, and then you turned a single piece of content into multiple social posts. So first question is, what does it mean to train it on your brand?
Eliot: So there’s two types when I talk about training on your brand. First is tone of voice so it can write like you. That involves giving it evidence, like examples of how you write, whether that’s blog posts, even transcripts are great for capturing a tone of voice. Giving it as much information as possible, asking it to analyze, break down, and give you back a sort of one pager of how you write and what you do and don’t do.
For example, for me, it’ll come back and say, “Eliot writes with dry, British wit humor and cynicism. So I didn’t tell it to write that, but it could see that in my writing. Add British spelling and then you can go through and say, “I don’t want you to do this. I do want you to do this and give it some instruction that that is one of the core tenets.”
Now, the other bit that most people don’t realize is in Claude you can train what are called skills, and these are repeatable processes. They’re a set of custom instructions with supporting files. So if you go into Claude, either the normal chat or Cowork, and you give it your logos, you can upload the files into Claude or point it at that folder directory. Give it your logos, you can give it your fonts, your brand colors. Even the thing that people ignore from their designer, their brand instructions about how their logos and colors should be used, and then they immediately go mess it up on Canva.
Those are hard written rules that you could give Claude and say, “This exactly how you use these colors with these fonts.” And you just say, “Can you extract all the information, extract all the logos out of these files?” Create a Claude skill that when I ask it to apply my company branding, it applies it to whatever I want to apply it to. So in this example, I have it set up for invoices. So I was working this morning, someone emailed me, “Hey, can you invoice me for this?” And I just copied the email and said, “Claude, make me that invoice.” And it comes back with perfectly templatized with my logo, colors, fonts, everything out the door. I don’t even have to touch a spreadsheet or an invoicing system. It just goes.
But in this example, it does exactly that same thing. But all I ask it to do is take a transcript, and I have a nice prompt that looks for interesting angles and stories and hooks from my natural language and write that in my tone and style of how I like my LinkedIn posts from previous examples. And then say, “Also, make it into a PDF carousel for LinkedIn”, and off it goes and works. I don’t need to touch Canva, I don’t need to touch Microsoft Word.
Sure, there might be a bit of editing out the other end just to give it that once over, but it’s like having a social media content creator or a copyrighter underneath you to say, “Hey, we need this. Can you take care of it?”
Rich: And since we’re talking specifically about social media posts, as you’re training it up, like we get the brand kit in there, let’s say. So now we want to train it up to speak in your voice. Would you put up all of your writing styles, or would you just recommend putting in social media post examples from Eliot Prince in this case?
Eliot: Personally, I give it a pretty widespread… I’ve written a lot of SEO blog posts, so I got millions of words to choose from in terms of just generally when I was firing stuff off, a lot of LinkedIn posts. As I said, transcripts are great for capturing a natural tone of voice, which people often don’t have if they haven’t written a lot online.
You could just find some cool transcripts that you’ve got recorded. Even if we’re talking about business stuff rather than personal brand and you want that, you could get all the copy of your website, especially if that’s been professionally copy written. That’s absolute gold. And I’ll mark up the files with what each piece of information is from and then within.
My custom instructions or in the skills that we’re building, Claude will actually be able to identify and, and present different, slightly different tweaks on tone for different social media. So it’ll know that I’m probably a little bit more relaxed in email than I would be if I’m maybe writing website copy versus the way I like to write my LinkedIn post.
So you can differentiate that if you are like, I just want to focus on LinkedIn, that is my platform. I have stuff specifically set up for YouTube because that’s my number one platform and I want it to be super dialed on that. So it depends how dialed you want to go on one platform or another. But if you just want one encompassing voice, then have at it. Yeah, let it rip.
Rich: All right. And so you talked about using a transcript. Was that because if you’ve done a video for YouTube on how to use Claude skills or whatever, you can just hand it the transcript and it’s going to generate all those sort of social promos? But can you also use it for things like, I’ve got a month’s worth of content and I need you to go find out what are the trending topics for this month, or what are the big holidays this month that a company like mine should be talking about, and pull in that stuff too, to create an entire social media calendar?
Eliot: In terms of your transcripts, are they stuff you’ve already recorded in the past?
Rich: Well, like for Agents of Change, we have transcripts, and obviously that’s a great source of content for social media. But if I was running an HVAC company and I wanted to pull in things about spring cleaning your HVAC system or things like that that might be relevant, could I also have Claude not just be taking transcripts from my podcast but also pulling in relevant things that hopefully my audience would be interested in learning about on Facebook or Instagram or LinkedIn?
Eliot: Sure. I mean, it’s got research features in there. As I said, it can browse the web. So if you’ve got a wealth of knowledge about cleaning HVACs, but you want to theme it against whatever’s trending or whether like spring cleaning, then Claude can help you do the research of what does spring cleaning mean versus what are we doing with an HVAC. But you’ll find it goes off and does research and adds any extra context that you need automatically.
I can’t think of what it was I was doing the other day, but it just went off and started researching some numbers for me. It might have been in this sort of PR study that I was doing about the job market, and it just went off. I didn’t even ask it, and it went and found 2026 data about what the job market and how many people are hiring in the UK versus other countries or stuff like that to help inform more stuff.
So sure, you can use it to not just repurpose but re-angle and find trending topics. You know, I’ll send it off to YouTube to find whatever anyone’s talking about this so we can see if I can spot a content gap.
Rich: Nice. And I guess one of the things I think about when I think of Agentic AI, is it can kind of go beyond the browser or beyond the platform. So in this case, and I’m not suggesting that this is what you should do, but could you also have it after it’s created all this content, just go ahead and schedule it on these platforms as well? Or is that a little bit beyond its capabilities?
Eliot: So I haven’t tried that actually, but my inclination would be yes. At the very least, as we said, it can work around. So I know that, go back to LinkedIn again. LinkedIn has a schedule feature built into the platform, so it could browse to your LinkedIn, go and use, like it can drive a tool.
For example, I don’t know if we’ll get onto SEO, but I use it to drive Google Analytics, build things into Google Analytics for me, Google Search Console. So at the very least it can drive a tool to set this up and you can train it to actually, if it can’t quite get there to take the steps.
Or on the flip side, Cowork has a scheduling feature in there, so you can schedule it to run every Monday at 9:00 AM. Which is bigger than it sounds. Because we’ve gotten so used to, or a lot of people got so used to, I’m going to make this crazy NA10 and make.com workflow with these beautiful step-by-step processes. We’re going to plug some AI into it, and you spend more time setting up your make.com workflow than actually using it. And then it breaks down because something changes or an API key changes or something, because Cowork can have loads of different connectors to different apps.
Plus, it can work its way around problem solving. And you can schedule it. You can schedule it to go and scrape or get information on the trending topics. Bring that all down, synthesize that with your brain dump from last week that was sitting in your notes, and go and create a notion page with your script for the day or your social media posts for the day. So that scheduling feature within Cowork is the real golden nugget there.
Rich: I actually think that’s incredibly powerful. Because I know that often AI when I ask, it has no idea about the day or the time unless it goes and does the research. So the fact that I could schedule in, it wasn’t even something that had had entered my brain. But that’s brilliant, and I’ll have to play around with that.
Eliot: It’s getting better at knowing where it is and what day it is, to be fair.
Rich: Alright. And the last topic I did want to touch on was the SEO angle of things. You have an SEO background. I love SEO, now GEO, visibility, whatever you want to call it.
So one of the things that you had mentioned is that Claude can audit an entire website and return a prioritize action plan in minutes. So in a situation like that, what is Claude Cowork actually doing there?
Eliot: Well, I actually was running this, because the thing is, one week it won’t be able to do something and the next week it can magically do something.
So I was actually running a test before I came on here to see like, what can we do now? And I didn’t give it super detailed instructions. I just wanted to see its capabilities. And I gave it a website and I just said, “Can you do a technical SEO audit on this website?” And it went off and started using its web browser, started using its connectors to go and get all the HTML on the website, do a sort of drive through and use the UX to see do these buttons work or not. Which is great. I was like, cool, it can actually go through my whole website and check this information.
What actually surprised me then, was it started to navigate and drive other tools. So it went to Google Page Speed Insights, which is a place where you can go and put your website link and it will give you Google’s analysis on how fast your website loads and all the technical stuff behind it. It went and drove that tool without me asking, and pulled all the data and information from that and put it into my report.
Plus, as I said, it can go and drive into Google Search Console. These are all things that I used to do manually. Like, I’m going to go to Google Search Console and get this information, or I’d have to hook it up to $150 a month minimum SEO tool and actually just get this sort of data down. And it was able to just do it all, bring the report back.
And now I’m thinking, not to jump to too many tools, but Perplexity computer I was testing this – and I think Cowork can do this now. As I said before, I’ve got it set up with my brand voice. It’s just done a technical audit and seen these 24 pages or blog posts haven’t been updated since 2022. I’m going to research the most up to date information and I’m going to pull all that based on each page, and I’m going to run page rewrites or update little sections of this page and data automatically in your brand voice. So it’s actually not just auditing, it’s actually doing the deliverables ready for me to go and send to a client or action.
Or again, we’re checking everything before we send anything to a client, but ready to just action. So that is for me, that is months of work. I could issue a year’s worth of SEO work for a client and go to bed and wake up in the morning and the intern’s job is done.
Rich: And that’s fantastic. And that’s powerful. It’s almost a little bit scary.
But I’m curious. When you started doing this, you have an SEO background. How much of your knowledge did you put into these prompts and this particular activity, versus how much did you, like if you were doing something without any background knowledge and just asked, “Hey, can you put together this for me?” I guess what I’m asking is, did your experience with SEO give you the framework to create a better tool than somebody without knowledge would’ve been able to create?
Did AI enhance what you already bring to the table and just speed it up? Or is this something where even if I don’t know anything about a subject, I’m still going to get the same amazing results that Eliot Prince might get?
Eliot: Here’s the big mistake people make with AI is they think it should know everything. Because that was sort of the promise. And just as back in the day you’d be told, don’t trust Wikipedia, everyone can edit Wikipedia so the information on there is wrong. The same can go for ChatGPT but worse.
Firstly, it’s trained on God knows what information around the internet. You have no way of checking if it’s right half the time, unless you are using certain research tools.
And two, if you’re not specific about what you want it to be working on with which information, then you don’t know what angle, what information it’s actually pulling to make its decisions.
So the best users of AI are not using it like a fancy Google to do the research for them. They’re giving it , whether that’s in a particular chat or in a project or in their file structure, they’re giving it the exact information that they want it to use and work on. They’re giving it the structures, the templates, the body of information of how they operate or what they want it to use to work on.
Sure, you can supplement that with a bit of extra factual research. So for me, I can look at an SEO audit and tell you pretty quickly whether it’s a load of rubbish or not. Because that’s half the battle of SEO, is who’s actually doing good work and who’s not. So the ability to know what’s good work and what’s bad work is critical. Because otherwise, you fall into trap of everything looks great when it comes out, any AI tool. Because it’s nicely structured. It’s kind of well written sounds, sounds right. So knowing what is good and what isn’t good work is helpful.
And then to layer on top of that, I’ve got all the spreadsheets, the frameworks, my SOPs. So when I’m doing this audit, I’ve got a spreadsheet, which is, here’s everything we want to cover. Here’s also even a website where you can order a process to do that. It can follow that step by step better than a human, because a human gets lazy and checks a box that they haven’t actually checked.
So yes, is the answer. Or in my opinion, as an SEO specialist, I would do a better job driving an AI tool than a beginner.
Rich: And that’s been my experience too, when I see people using AI and I’m really impressed with what they’re doing. They didn’t start at square one. They came with a specific base of knowledge, with life experiences and a premise or a perspective that was not the general one.
And you said something in the answer, you said something about AI knows everything and that’s the problem. Because AI knows this average general information about it, but it’s the Eliot Princes or whoever it may be, an SEO or whatever your skillset is that has those unique experiences, that all of a sudden takes it from an everything all in one tool to a very specialized tool.
And I think a lot of people don’t realize that they can get from beginner to intermediate, no problem. But they can’t get to expert because they don’t have that experience, has been my experience so far.
Eliot: I think you would compare it as well to someone who’s been in any industry, who’s been in a particular type of marketing or whatever field it is for 30 years, versus someone who’s read all the books in marketing. Like you don’t actually know until you’ve done the thing and run into the problems and understood why that doesn’t matter.
Like in theory that’s really important, but when you’re actually out there in the wild, that little thing that you learn in a book doesn’t make any difference. So that experience in anything you do is invaluable.
Rich: Getting back to the SEO thing, although I enjoy this conversation. You mentioned that you were building things in GA4 and Google Search Console and some of the other things. What exactly are you using? How are you connecting Cowork and those tools? Are you uploading screenshots or reports from those tools, or are you giving them access to those tools? And what does that part of the job look like for you?
Eliot: Every which way you can imagine, really, depending on what I’m trying to achieve. The frustrating thing about Cowork from an SEO perspective, or Claude generally, is it doesn’t have direct MCP or API direct connector into Google Search Console, into Google Analytics. Whereas Perplexity is the only other tool I’ve found that that does that. So that’s a frustrating frustration that it can’t just go and pull data out there.
So that either leaves me like, well, I’m just going to download 5,000 rows of data from Google Search Console, and I’ll just give it to Claude to work through and just figure out what the hell’s going on here.
Spot patterns. AI’s great for pattern spotting, trend spotting, and things like that. But as I said before, it can work around and it can control my browser. It can genetically control my web browser. So instead of me trying to figure out how to build a Google Analytics 4, which I’ll leave my opinions on GA4 at the door, but it’s a nightmare to try and work out how to do anything.
I can just send off Cowork to kind of build my report or whatever I need in there. The other trick there is if it’s a tool that it’s struggling with, you can just get all the documentation about how to use the tool. So I could give it all the documentation on GA4, put it in a project and say, here’s everything you need to know. And it will go off and find it.
Often, I don’t even do that because it’s clever enough to go and find the documentation on a website. It was trying to use an API the other day wasn’t working. It went and read all the documentation about how to set up the API and figured it out for me. So the Claude and Chrome add-on, which you can use manually, or you can drive it through Cowork. Cowork is an absolute treat for all that sort of stuff.
Rich: Awesome. If somebody’s listening today and they’re like, well, that sounds great. I’d love to try it, but I’m not sure what to do, what would you say are their first steps to kind of understanding and using Agentic AI or Claude Cowork specifically?
Eliot: Claude Cowork specifically, it can be a bit confusing, frustrating. Because I think one thing we haven’t mentioned here is it actually can work with the folders on your computer. So people kind of jump in and they go, whoa, whoa, whoa, confused. Like, what’s this folder versus this?
So step one, and it sounds simple, but someone said to me the other day, this was one of the biggest breakthroughs they’ve had recently. Download the Claude Desktop app because that’s going to give you the full feature set. You’re not just going to be locked into Claude Chat, you’re going to get all the other features.
Then once you go into the Cowork mode, the toggle in the top middle, select ‘Cowork mode’. You need to point it at a particular folder on your desktop. So it’ll ask you what do you want to work on? And it will ask you to pick a folder. And as you said at the beginning, the easiest way to understand how this works is to get it to do a really simple task on your desktop.
So go and point it at your screenshots, for example, or a big file of images that you’ve never bothered to look at, or you’re like, I know there’s good stuff in here, but I’ve got to rename it. Go and get it to look at all those. Just ask it, “Can you go through all these images? Can you work out what they are? Put names on them so I can identify easily what they are from the name and put them into grouped folders so that they’re nicely filed away.”
And just learn to do that really basic step, and then you’ll start to see the power of what we can start doing here.
Rich: Awesome. Eliot, if people want to learn more about you and maybe some of the things you’re doing, where can we send them online?
Eliot: YouTube. Just search Eliot Prince on YouTube or on Google, you’ll find me. So I have training, you can find all the Cowork 101 basic stuff that Rich found me doing on YouTube.
And then I also run sort of a weekly newsletter, which has all, I call them ‘AI recipes’. So you can get step by step processes of all these sorts of things I talk about, step by step with all the prompts, how to do them, everything. So even if you’ve no idea what to do, that’s where I help people to start and actually get you using Agentic AI.
Rich: Awesome. Eliot, this has been great. I’m so glad that we connected and you were able to come on the show and share some of the ways just to get our brains thinking about how we might use Agentic AI. Thank you again for your time today.
Eliot: It’s been so much fun. Thanks, Rich.
Show Notes:
Eliot Prince’s work focuses on helping business owners use AI in practical, repeatable ways that save time and improve results. With a background in SEO and systems thinking, he’s trained hundreds of professionals on how to build AI-driven workflows that actually work. Check out his YouTube channel and weekly newsletter for step-by-step instructions and more useful insights.
Rich Brooks is the President of flyte new media, a web design & digital marketing agency in Portland, Maine, and founder of the Agents of Change. He’s passionate about helping small businesses grow online and has put his nearly 30 years of experience into the book, The Lead Machine: The Small Business Guide to Digital Marketing.