Stop Starting From Scratch

Written By Andrew Foxwell | Co-Founder of Foxwell Digital

Andrew Foxwell & Will Sartorious open with a quick roundup of what's new in AI before diving into the main interview. They cover GPT-4o's image generation model beating Gemini in a 15-ad cookoff, a new Alibaba animation model called Happy Horse for animating static ads, Claude's new design tool, a Canva feature that separates layers on AI-generated ads for editing, and a joint venture between Anthropic, Blackstone, and Goldman Sachs that signals a shift toward AI-augmented service businesses.

The interview with Grant Hushek focuses on practical AI infrastructure for teams. Grant is an AI consultant specializing in education, adoption, and implementation. The bulk of the conversation covers skill files and markdown files, what they are, how to build them, how to share them across teams, and how to iterate on them. Grant explains how he connected Fathom (his note-taker) to HubSpot to auto-update CRM records every two hours, and he shares his three-iteration framework for building high-quality skill files. The episode wraps with a debate on Cowork vs. Claude Code and a practical screen-share walkthrough Grant does live on the call.

Key Takeaways

  • Why this one file type is the single most important building block for getting repeatable, high-quality AI output in your business.

  • Why you need to be sharing AI skill files with your team so that everyone produces the same level of output.

  • The first four markdown files every company should create when getting started with Claude.

  • Why should you be talking to Claude instead of typing when building your "About Me" and voice & tone files.

  • How this three-iteration framework turns a bad first skill file into one you'd actually be proud to use every day.

  • The real difference between Cowork and Claude Code, and the one that is actually right for your team.

  • The Process Grant used to automatically update every CRM contact record after every call.

  • Why giving Claude a PDF template tanks the quality of the output it produces and the file type you should use instead.

Tangible Links from the Show: 

To learn more about Grant and his team At Grant Bot head here https://x.com/GrantHushekgrantbot.co

To Connect With Andrew Foxwell reach him here Andrew@FoxwellDigital.com

To connect with Will Sartorious DM Him Here https://x.com/will_sartorius

To Connect With Thomas Moen DM him Here https://x.com/thomasmoen

To learn More about The Foxwell Founders Community and the converstations, like this one being had go here: www.foxwellfounders.com

Full Transcript

(00:03) If you work in D2C and you use AI and you're wondering what the F is going on every week, this is your podcast, the AI D2C WTF podcast, your home for tactical tips, strategies, and ideas that you can implement right now in your AI workflows to make your brand or agency more money. Well, welcome to the AI D2C WTF podcast.

(00:26) And let me just say by starting, this podcast was really for helping you get tactical information about how to use AI better in your D2C e-commerce agency or brand, right? I mean, that's basically what we're trying to do here, isn't it, Will? Yeah. I mean, very simple. How to make your stack, whether it's generating ads or just your stack in terms of your operations better through AI.

(00:54) So one of the things that we have is we're interviewing someone for most of the episodes that we have, and we have a bunch of episodes we've already recorded. So be on the lookout for those as they come. This is the first one, but we're going to try to, as much as we can as well, talk about what's new in AI at the time that we've recorded it.

(01:11) There's just changes by the hour, right? In the recent weeks that we're talking about. I mean, certainly we could talk about the new model on ChatGPT or what are we calling it? GPT2 or whatever it's called for image generation. So what do you think about that, Will? And then what are some other things that have happened recently? Yeah.

(01:30) I was sort of bearish on GPT's image modeling and just GPT in general for quite some time, but I think it's sort of twofold. One, the GPT2 image model sort of blows Nano Banana, the Gemini Google model out of the water in almost every stat. I ran a bit of a cookoff in 15 different ad types and GPT2 won all of them. The caveats are it's a bit more expensive, but ultimately, you know, in terms of product fidelity, in terms of matching your fonts, in terms of matching your colors, it's pretty unstoppable.

(02:08) And obviously we're going to continue seeing these components changing, but I would say bar none, this is the model to be using if you're going to be generating static ads. And then subsequently, if you want to be animating those ads, the newest model from Alibaba, Happy Horse, I think is what it's called.

(02:28) You know, who comes up with these names? God only knows. But it is, it's really quite good. You know, Seed Dance from TikTok was sort of reign supreme for a while, but now sort of the Happy Horse model is great for taking static ads and animating them. I think so that's good to know. That's a that's a change that we've seen. What other new announcements have you seen that have sort of hit you and been like, oh, wow, that's that's something that's really cool.

(02:53) I mean, one that I saw that I thought was really cool is sort of the operating system. It seems that Jacob has put out that a lot of people are curious and using and it allows you to spool up agents. I'm not really sure what it does 100%. Maybe you can help me understand that. That seems to be something a lot of people are talking about.

(03:11) Yeah, I'm honestly not too, too familiar with Jacob's setup. Yeah. So he put out this thing that's it's called HQ and it allows you to share knowledge, skills, workflows and API access across the organization. So everyone works at the same foundation. Nice. OK. I mean, this sounds super promising. I would say something we struggle with internally and I get on client and consulting calls and all the time is where should I be putting on my markdown files or where should I be putting all my agents? You know, right now they just sort of live locally on my computer.

(03:42) How do I share them with my team? This is definitely a pretty massive solve. I can't say that I played with it, but it sounds pretty promising, I would say. Awesome. Yeah, I think it's interesting. I mean, I think a lot of that is like the next phase of AI seems to be the organization of where we put all this.

(03:59) And so that clearly makes a lot of sense, which is pretty cool. Outside of that, Claude, our best friend lately, anything that Claude has has put out that you've been like absolutely loving or utilizing recently? Yeah, I would say sort of twofold. Obviously, the Claude design came out. You'll hit your limits pretty quickly.

(04:17) The limits very fast. Yeah, very fast, very fast. But we built our new website on Claude design. It took me a weekend to do totally nailed brand guidelines. I would say for one off projects like a website, it's great for iterative processes. So you prefer to use Claude code, especially with CRO landing pages.

(04:38) Claude code is still bar none the best. If you combine Claude code for Cell and GitHub, you can spin up landing pages really rather easily. The other thing that is, and I'm sure most folks listening to this may or may not know about, is on Canva, they dropped a tool that allows you to upload AI generated ads. And what it will do is it will take each layer of the ad you created and separate them.

(05:06) So you can move elements around, you can change your copy. Folks have had issues for ages, you know, editing AI images. And so like that has been a massive unlock for us, you know, for, you know, for allowing our editors to, to clean those minor things. Be more flexible with it. Yeah, nice. Cool. Totally. And then I think, you know, the most exciting announcement probably unrelated to marketing was sort of the announcement of the JV with Anthropic, you know, with Blackstone, Goldman, et cetera, sort of saying that it's going to be spinning up this

(05:39) service oriented consulting service to ultimately implement AI within enterprise businesses. And I think folks have been talking about this for a while and saying that service augmented with AI is going to lead the pack in terms of company types. More, we had SaaS for a while, but now it's service plus AI. And this almost sort of confirms that hypothesis that there is a lot of demand for this and both Anthropic and OpenAI, you know, OpenAI is about to announce it as well.

(06:16) Doing this almost confirms that, like, if you are able to effectively better augment your agency or your brand with custom tooling with AI, you are going to be at the forefront of this, of this bubble, if you'll, if you want to call it, even call it that. Yeah, it's an, it's an exciting time. I agree that, you know, you going through this as a person that's learning and listening to this podcast and soaking up new knowledge is, you're at the leading edge.

(06:43) And so good for you for, for continuing to follow along and yeah, seeing the big player validated is huge, right? So the more that you can help people navigate this, the more that you can experiment yourself and be able to augment the service offering that you have is also a huge with this. So it's a, it's a fun times.

(07:00) Our first guest that we have on the podcast is Grant Holschick, who is really interesting. I mean, we talked about a ton of Claude stuff in this episode. We talk about a ton of the structural ways that he's building things. Thomas is also in this episode outside of a cafe in Prague, drinking a glass of champagne, which is just like, so you can tell by his vibe, the questions are quite good.

(07:22) So let us know what you think, Andrew at foxwelldigital.com. I'm always here to receive anything from you and these episodes are going to keep coming, right? So if there's people you want us to interview, there's people you want us to talk to, we're happy to, and we're really excited to, to dive into it.

(07:36) We go into these episodes with a lot of curiosity and that's our goal. So without further ado, let's take it onto the interview. Thanks for coming on the show. We appreciate it. Yeah. Thanks so much for having me. I thought you were going to try to say my last name and I was like, let's see how this goes. Nope.

(07:54) I thought about it and then I didn't. And then I didn't. So I'm just going to leave it for now. How do you pronounce your last name? Let's just see. Hoshik. Hoshik. Okay. I always thought there was an L. So this is why this is the problem that I have because I've said it wrong to other people. Hoshik.

(08:12) So Grant Hoshik on the podcast and we are glad to have you. By the way, so just like a 30 second brief intro, what do you do with AI now? Yeah. I'm an AI consultant, education, adoption, implementation. And what that means is that I am actively processing transcripts, processing client needs, processing my team's communication into something coherent.

(08:37) Right. So that could be making skills for other people, making skills for myself, making scheduled tasks. It runs the absolute gamut. Well, so just, you know, you see a lot of implementations in business. And my first question is, what are some of the biggest unlocks you've seen for folks in like the last 30 days that you're that, you know, you feel like every time you show the super person, they're like, oh, that's an insane idea.

(09:02) I think it's a lot less sexy than people will hope to hear. But really breaking down the fact of like, what a skill file is, when you think about repeatability and leverage in an AI world, it all starts with a simple markdown file. So like, I've spent 90 whole minutes explaining what a skill file is and how to make them and how to think about them to companies.

(09:28) That has been the breakthrough moment. But the more fun one is that I've connected Fathom, my note taker to HubSpot, and every two hours, it updates every single record that of people I talked to. And that that's like my happy place. I'm so happy. What might be helpful is I think folks maybe get a little confused about the difference between like a skill versus a markdown file.

(09:48) You know, what are these different components? Like, as I understand it, and I'm, you know, playing dumb for a reason, they're just like a block of text, right? Like, what exactly are these things? And why would you not just use Claude Chat directly? And what is the purpose of even using them? Okay, so a skill file for someone listening is it's essentially like an SOP, right? It's a standard operating procedure, but for AI.

(10:12) And so there's two main parts to it. There's the description, which is formatted in YAML, which you can just ask Claude to be like, write this in YAML. And then the real meat and potatoes of it is markdown. And markdown is the coding language, the formatting language that Notion runs on, ClickUp runs on.

(10:30) Anytime you do like a pound symbol, and then the text gets bigger, that's markdown. Okay. And so what Claude does, it takes your process or your SOP, the way that you do things, and turns it into this file that it's going to follow step by step when you request that output or that skill. Okay. And so most people in their role, they have a collection of tasks and deliverables that they're responsible for.

(10:58) And you can make a skill file for each of those deliverables. Skill files can be updated in any chat. You can ask a skill to run. It can make an Excel file, it can make a document, it can make a PowerPoint, an HTML file. And you can say, okay, I didn't love how you did this part, just update that part of the skill. And then next time it'll get better.

(11:18) And so why do people care about this is that when you think about Claude and co-work and Claude code, I talked about leverage and repeatability. So if you have a skill file that accurately and consistently gets you to a high quality output in a 10th of the time, then you can use that programmatically through a scheduled task or just in your chat, and you will get the same high quality outputs more consistently using less time.

(11:45) And so the skill file is kind of the key that unlocks AI leverage in each individual role and ultimately in a couple. And for me, Grant, talking to you in Lisbon a couple of months or like a month ago, one of my biggest takeaways is that one thing is the skills files and using it for yourself to, you know, make yourself more productive.

(12:06) But you talked about sharing skills with the team. And for me, this was like, I haven't really thought about it that way before, but I've now started doing it and it's like a superpower unlock. But could you talk a little bit about how that works and what that is and maybe some really practical tips on how to do that? Yeah, absolutely.

(12:27) So why do we share skill files? Well, for one, if I'm making, let's say I make a lot of branded documents. So branded documents could mean SOWs, scopes of work. It could mean a prompt that I want to share with one of our clients. So I make a nice document explaining what is the prompt, how to install the prompt, how to use it and what you should get from it.

(12:48) I want those to look nice. And so this branded document skill is something that I had been using personally for months. And then I realized that my engineers are also making deliverables and things that memos, briefs, etc, for our clients to read. And so I was like, okay, well, we need to get you guys the same level of output that I get.

(13:06) But I don't need to teach you what our brand colors are, you don't need to necessarily know what our hex codes are, because all that stuff is inside the skill file. So just by downloading that skill file to my computer, even though I had Claude write it, I've iterated on it probably five to 10 times, I downloaded it.

(13:23) And then I gave it to someone through Slack, I can give it to them over email and say, Hey, go into Claude, go to the customize tab, click on skills doing this all like I know exactly where all the things are. Go to skills, press add new skill, upload skill, bag the file into that little box that appears.

(13:43) And it'll upload to your, your Claude database warehouse, whatever you want to call it. Now, if you're on Claude teams, you can also go into that customize tab, click on skills, and then share a skill with your team. If you are an admin, or if your admin has enabled the team members to share those skills. And the last thing is that if you are an admin, and you have a team account, you can also create org level skills that you would then upload to the organization skills that exists in your organizational settings. That was a lot of ticky tacky

(14:13) stuff. But you asked for the practical. So I'm giving you all the all the clicks. It's good. I mean, I think what the this is exactly what we want to get into and give people the idea of how they do this. One, I have a question about Claude code. I don't think a lot of people have done a lot with Claude code.

(14:30) Or, you know, it's like people have built skills potentially, right? Or they're working on markdown files. But it's like, what are what are some of the like, top couple things that you've seen you've done, or you've seen other people do in Claude code that you feel like are massive unlocks for organizations? So I'll say two things on this one, I have a cool Claude code use case.

(14:49) So I'll answer that part first, which is, I think the the ability to, for product people, engineers, marketers, the ability for you to visualize in an app format, what's in your brain, like, you don't need to storyboard anymore, you don't need to go and like, sit on the phone with a designer or product person for an hour and say, No, the button should be like here, and it should look like an arrow.

(15:11) You don't need to do that anymore. You can go to Claude code and describe what you want. And it'll write the app. And it's very easy to now deploy those apps to a platform like Vercel. And I say that, even though I'm an AI consultant, like I took three coding classes through my university time, hated all of them, barely passed, I didn't even pass some of them, still graduated point being me and code are like oil and water, we don't get along.

(15:35) But I understand how the things connect. And so I'm actually more proficient in a tool like co work that has the same schedule task features that has the same skill features. So for someone that never really got code, co work still does a lot of good for me. But when I want to get a thought out, that's much more interface driven, like I want to click around, I want to see the thing lovable became really popular.

(15:59) But like, I canceled my lovable, because I know that I could accomplish the same thing in cloud code. So for marketers, salespeople, product people, if you felt like code was too far away from you to actually be able to do anything useful, I can tell you that if I was able to just describe what I wanted, and it made something I think that was relatively aesthetic, let's say, you can take a screenshot of an aesthetic website that you like and give that to cloud code, and it'll make something look pretty, right. So just knowing that

(16:28) that's possible, you can describe it and create an app. I think that that is that is a use case that anyone should know that they have in their back. Will, I'm sure you have other notes on cloud code, too. Yeah, I mean, I think it's also worth calling out that like, creating a markdown file, or creating a skill is as easy as going into your cloud chat and saying, you know, maybe you go into a cloud chat where you just had a very long conversation about a client, and you were going back and forth, you know, maybe cloud chat was

(16:53) helping you brief out an asset. And you're like, Oh, God, I just put so much good information in this chat. Like, what do I do now? The solution is as easy as saying, Hey, Claude, can you write a markdown file for everything we just discussed, and make it generic, so I can use it for any client, right? So like, maybe I was using a cloud chat for going about creating emails.

(17:17) And like, I thought like, Oh, this is such a great way to unlock, you know, open rates or whatever it was. Now I want to apply that to any client. So I can create this generic markdown file based on the conversation we have, we had. So Claude is always sort of coming in with that prior knowledge, right? And if you're too scared to use Claude code, you know, one, get over it, it's not that scary.

(17:37) But two, you know, your alternative is you can use cloud chat, you can take all these markdown files, put them in your projects in cloud chat. And like that works just as well. Because I know cloud code can be prohibitively expensive, because it's not offered on the pro plan anymore, you have to be on max. So if you want to just do everything that Grant sort of outlining in cloud chat, you know, that, that works, that works really well, you know, as well.

(18:00) So I guess, Grant, you know, I guess my question for you is, you know, you're chatting with a lot of companies that probably are going from zero to one, like, what is the first markdown file you tell them to create for their company? Branded docs, branded PowerPoint, about me, and voice and tone, those and about my company.

(18:22) So I'll break up. I love decks, PowerPoints. It's like, in business, for some reason, we've been obsessed with like, what the 16.9 aspect ratio, like, God bless it. Everyone loves 16.9. I don't know why. But you want to have PDF. Love it. Everybody does. One time I was in a meeting with a massive ad agency, I was working and I was a consultant with the brand.

(18:44) And they had a consultant, like a massive agency that came in and did this, I think it was 300 slide deck. And it was an all day like presentation thing. And I found out later that they were charging them $75,000 a month. And all they were doing was just like putting this strategy in this deck. I'll never forget that.

(19:02) I haven't forgotten it. Obviously, this was like six years ago. I mean, 75 grand for a deck. So there you go. There's a whole nother business line. Anyway, I'm going to reevaluate what I'm doing for a profession. Anyway, the like people really love decks. So understanding how to make a really good branding skill, I think is really helpful.

(19:23) I talked about branded documents, I make more document files and I do decks. But I got on a call with someone literally yesterday, who's in the sales team. And she's like, our sales team is consistently pulling in different features that we need to sell to different enterprises. But we need to have them all look the same while they sub out different slides, like section one versus section three.

(19:45) And I was like, okay, well, we need to see what your template looks like. And she showed me that she had uploaded the template deck as a PowerPoint, or sorry, as a PDF, instead of as a skill. And so AI, when you give it a PDF, it has to use like optical vision, instead of understanding what's in the code, because PowerPoint, Excel, Doc, those are all code based things, which is why Claude can make them with high accuracy.

(20:08) So if you give it a PDF, you're going to lose a lot of quality between what your what your reference is, and what's actually going to come out. So branded decks, really popular. And then just general branding. So where your hex colors, your fonts, etc, good to have just across the board, separate to the deck.

(20:28) And then I talked about a few personal ones. So your voice and your tone, how do you write? How do you actually talk to build out that skill or that markdown file, highly encourage you to get a tool called whisper flow. Because when you're talking, not typing, you actually maintain your authenticity and your voice. And then you have your about me file.

(20:49) Great, just a you can, I've got a prompt if you want to email me, but like, you can go and just say, Hey, Claude, interview me to understand who I am, what I do, how I like to work. And Claude will interview you and you can respond. And that'll create an about me file to give context about who Claude is working with. And the last one is about my company.

(21:07) And this is one that should be shared across the org. So if you want to have consistency, share your about my company file. And that way, everyone will have the same reference for Claude. I think that's great examples. So let's get super practical into this. I now understand I need to create these different documents.

(21:28) I understand I can chat to Claude and he might do something for me, but could you like, break it down super simple? What's the process of creating these documents in as simple way as possible? So the about me, the voice and the my company will be more straightforward, because you're not using it.

(21:47) Like your reference is your is what's in your head, right? So if you start just to jump in there, Grant. So when it's in your head, then you would recommend whisper flow versus writing, because then it's more you use more words, it's more colorful, it's more in depth, right? Yep, exactly. So in those situations, you can start a new chat or a new task and co-work.

(22:08) And just explain what you're trying to achieve. Like I want to create a markdown file that you Claude will reference when I start a new chat or task. And this is specifically like, this is really much more useful for co-work than it is for chat, because the chat is limited in the knowledge base of a project, or it's limited to your global instructions.

(22:31) Co-work can actually reference files like Claude Code. So this is when we talk about file creation, we're really thinking, okay, well, how can we get leverage on co-work and Claude Code? So stating your goal, I want an about me file that you can reference when you're talking to me, interview me, right? I also really like to use a tool that co-work will often recommend, but I asked for it specifically, which is called Ask User Questions.

(22:56) And so what this does is it just pops up a little multiple choice select. And I like that for those kind of quick, okay, well, how long do you want the file to be? Or in what circumstances will I be referencing this file? Those types of things, like it'll suggest answers that you can just select as multiple, multiple or single select and move on.

(23:14) But then you want to start getting into the whisper flow and really speaking what's on your mind. And you can go for two minutes of just yapping. And that's great context for Claude to have. So don't be shy, really let it all out. And then it'll make sense of that unstructured text.

(23:30) And so you can do that for your voice, your about me, and your my company. Those those three are going to be really good for the self interview. I've done, I think the my about me, I spent literally like an hour back and forth until it felt good. I had a really good starting prompt to work from that was like probably too comprehensive. But, you know, let it take 10 minutes, let it take 15 minutes or 30.

(23:54) Like it's a it's time well spent, because that's really strong leverage for when you're working with Claude. So just to follow ups from that. So one thing is you created the file, you start using it. But as I understand it, you can also kind of ask it to update itself. And and it's it should be like a living document. Am I right? Yeah, definitely.

(24:13) All I think all of your files, all of your markdown files can be considered living documents. Because at the end of the day, this is what I learned from automation, like edge cases exist in life, just generally, right? So if I'm making an SOW, and we have these core offers, and then someone comes to me and says, all right, well, I'll pay you triple your rates if you do it completely differently.

(24:33) And I'm like, Okay, well, I want to start from this template that I have. But let's work in this new option, right? I might want to use this new service again. So you can then say, go to my SOW skill and update it with this new this new path, potentially. And so even when you think about your about company, like if we another great example, we're growing, we're hiring, new person joins the team, part of the about company tells you what the team roster is and what each person's role title is and what they're responsible for. So if you hire

(25:01) a new person, great idea to go and update that file, simply by starting a new chat thread and saying, update my, my company file, because I've just hired Bennett, Bennett runs media buying, senior media buyer, he's going to run these five accounts. Great, that's gonna be super helpful for Claude to know that. That's great.

(25:20) And then my second follow up would be, so that's kind of the the markdown documents. So what about like the designs and getting it to make the correct presentations and all that stuff? How do you go about that? Yeah, so skill files, for the most part, are going to be just single markdown files, single text files.

(25:38) However, when we want to have reference material, like a PowerPoint template or an Excel file as a template, we might want to have our skill reference that material. This is a great example for our decks, right? And so if we are in the process of creating a new skill for branded decks, then we can provide that PowerPoint file to Claude.

(25:59) Claude will read the file, understand what's there, give it as a PPTX, not a PDF. And then when we make a skill file, we can tell Claude, include this PowerPoint file as reference material. And the big difference here is that at the end of its process of asking questions and getting answers and writing that skill file, there's going to be multiple files that it spits out.

(26:20) And you're going to want to click download all because then it's going to zip those files together and you can move the entire zip folder into Claude. And when you upload that, all the material is there. It's referenceable and it's a larger body of work for that skill to work off of. And that's where you get really specific and granular skill files.

(26:44) Great for branding, great for visual things. One thing I've always wondered that I haven't had the guts to ask someone is why and do you recommend setting up like a dedicated computer for all of this? Like, I mean, because I got a thing, you know, dispatch or whatever it is on Claude that was like, hey, you can do this from your phone and like, it's still running on another computer.

(27:09) And I don't, you know, apparently people just have computers sitting around. Like, I don't have that kind of money. Just like having a computer sitting around. But what do you recommend doing that as you continue to sort of build on this? I guess I'm just curious of your opinion about the infrastructure side of it. I'm getting into that much more right now.

(27:26) I kind of went down the rabbit hole of trying to push co-work as far as I could with this idea of like a helpful employee companion. And so what I did was for the longest time, I've had scheduled tasks working in co-work that read my transcripts, Slack, email, look up and identify open loops, things that I said I would do, but I'm not doing in the call. And it logs those.

(27:48) And I found out that I had six different areas in my computer that that information was being logged. So that's a no, no, don't do that. But then I was like, okay, well, we got to start cleaning this stuff up. And so I thought, okay, well, if I had co-work go and index that list, try to fully execute a task. And if it couldn't give it up to someone on my team and say, I need information from you.

(28:09) But an actual employee would first go try to see if the task had already been done, because maybe I do it on a call, but the transcript doesn't say that. Or maybe I sent the email, but I never confirmed with Claude. So you need to go look for proof of completion. So where I'm going with all this is that co-work has limits.

(28:24) Scheduled tasks only work on time. They don't work on tool. They're not tool-based triggers. So for example, difference is when Fathom completes a transcript, it is not going to start a scheduled task. A scheduled task can run every two hours and look for calls that finished in the last two hours, but Fathom can't start the workflow.

(28:43) Okay. So from an infrastructure perspective, I am now taking a playbook, a note out of a play out of Thomas's playbook. I've got a Mac mini that I just turned on today. And I am currently moving all of the edge pushing, the boundary pushing that I've done in co-work. I'm moving that over to my Mac mini. I'm setting up OpenClaw and I'm setting up the Gary Tan infrastructure for file storage.

(29:09) And so I'm combining all these things now into this external thing. Main reason why my computer is on fire all day. And I cycle through the battery. It is a less than one year old MacBook Pro. And I cycle through the battery five times a day. That cannot be good for me. Like I just doesn't stand to reason that that's a good idea.

(29:30) So I moved it to a new computer. I definitely, I definitely identify with this. Uh, this is why I asked because like I asked, it was doing some really complicated shit the other day. And like at the same time was also building a, um, monkey riding a toilet cart game for Nora. By the way, it's pretty dope game.

(29:47) If you want to play, let me know. You got a lot. What's the context though? Uh, she, I asked her to come up with a, um, a funny game and she came up like an idea. So I'm trying to teach her like game design and design thinking and stuff. And she designed basically this Western themed game where a monkey rides a toilet around space. It's very fun.

(30:06) And it's, uh, you have to visit different like Western themed bathroom stops. The monkey does. So she's sick. That's amazing. I'll send it to you guys. It's pretty dope game. But anyway, so, so we're like, I've got this and it's like running this other, like two other things that are like quite complicated.

(30:26) And I mean, it was, it took like the, all of it was taking like 20, 30 minutes and it was comical because my, it was, my computer was flaming hot. I'm like, I think I'm reaching this point. And so I didn't like, I, that's, that's why I asked the question. But anyway, yeah. Just a comment, Grant. So first of all, welcome.

(30:42) And it's, it's great to have you in the, in the cloud community. Uh, what I think going back to the start of your point of like building, building your skill bases and your MD files, right? So what's amazing with, um, AI going forward, for example, with the open clause where you can install it on your Mac is that I use a shared library via Dropbox.

(31:03) So everything that co-work or cloud code or cloud chat is learning and updating these MD files of the company about me. I'm also putting that into my open clause agents who are also reading and writing from it. So all the infrastructure that you have created with cloud, you can just directly put into kind of, uh, the infrastructure can keep working in, in another platform.

(31:28) And I think that's the magical part about MD files and skill files. Yeah. I'm setting up that sync tonight, late, late in the evening hours. And then just to, and just to add onto that, like, so you're new to this ecosystem. One thing I'm curious about you guys thoughts on is just like co-work. Like I don't use co-work at all.

(31:45) I just use cloud code for everything. Is there like any advantage at all in using co-work, uh, versus using code? Uh, that's my first question. Second is just like, sort of like a piece of advice for our listeners here is like your first mark dial markdown file or your first skill or whatever the fuck you want to call it will probably suck.

(32:03) Like it may not be perfect. And so what you need to be doing is iterating. Right. And so like internally we have marked on files for every different type of asset. So if I wanted to create like a before and after ad and I say, Hey, use this markdown file to create a before and after ad for me, for this brand.

(32:23) And it does, it generates a prompt for me. And I need to put that prompt into nano banana, GPT two, whatever it is. And I get an output. What I need to do is I need to take that output and I need to put it back into code. And I need to tell code, these are all the issues with this ad. You need to go back or maybe more politely, please go back and update all of the respective markdown files that you use to get to this point.

(32:45) So we don't get this mistake again. So the idea is like you're creating an iterative loop. And I will say, I may be, you know, I may be on an Island in this group here. I will say having a human in the loop doing that is far more beneficial than having a QC agent do this work. Like I've tried having a QC agent just like iterate and iterate and iterate.

(33:05) It's fine, but it doesn't necessarily get there. The one thing that humans have as an advantage, especially in the marketing space is taste. Like, you know, and we know on this call, you know, what maybe like a good ad looks like or what a good landing page looks like, right? Like we can optimize as much as possible, but like there's, there's, there is an innately human intrinsic thing about taste.

(33:25) And I think that's why, you know, having a human in the loop do this thing. But so anyway, that was just a bit of a tangent and an aside. I want to go back to the cowork versus code thing, because maybe I'm missing something here. And maybe our listeners are missing something here because I think there's a code seems to solve any problem that you could ever do.

(33:42) Can I, can I go first? I have, I have feelings about this. So I've been experimenting with chat, org, cloud, and open cloud. And it's, it's a, it's a strange harmony and there's a lot of feelings around it. But for me, what I landed on is that a cowork is my product manager that connect. It's easier to connect to all the different connectors.

(34:06) It's easier than to just make sure it has all the files connected. Cloud code for me is too complicated. It looks cool when my wife looks behind my back on what I'm doing, but in terms of just making it do consistent, good quality work for me, cowork is better. Especially in terms of making sure all the connectors and everything is working because on cloud code, I have a hard time making sure it remembers that it has to do this connector or the liner connector or whatever.

(34:36) So for me, it is my project manager that I talk about when I'm going to develop different things. And then I get that to give me the prompt I need to do in cloud code to develop. So that's my take on it. Got it. That makes sense. Maybe it's gotten better since the last time I used it, but I've just found that using APIs and code is just like so much more efficient than the connectors and cowork, because I feel like the connectors, and again, it's been a few weeks since I've done it last, but I felt like the connectors just kept

(35:05) breaking down. And then I have to go reconnect them, reconnect them, reconnect them. I was like, this is not a repeatable system. There's a very known bug about this connectors dropping down with the authentication thing going, which I think will be resolved soon. But you're correct. Some of them are breaking down and it is much better to connect directly to the APIs.

(35:28) But I think for the normals, the normies who are just using this every day, I think it's easier to just understand you have cowork, you put in all the connectors and it will just work for you. And of course, they will prompt you if you need to reconnect or re-authenticate things. But it depends on the nerdiness level, I think, of the whole delivery.

(35:47) But for me, showing cowork to different clients, which is like small e-commerce businesses in Norway, they get their mind get blown by just understanding how cowork works. And, you know, applying all the things Grant have been talking about today and just giving them MD files and creating schedules and stuff like that.

(36:09) You are like the coolest person in the office. You're like a superhero. You're like the the peak AI person just applying these simple tactics, I think. And I'll throw in on that completely agree. The percentage of people that are going to enjoy code, cloud code is disproportionately smaller than people that are going to jump in and immediately get value from cowork.

(36:29) So if you're listening to this and you're like, I don't know what to do here. Go for cowork, have a ball. Right. And to your point, Will, is this a video podcast or is this audio? We have video component as well on our YouTube channel. Yeah. Fantastic. Do you mind if I screen share something? No, let's do it. Okay.

(36:46) So this is what I was looking off to the side and in eager anticipation. So this is something that we teach a lot in our workshops. Super colorful. Effectively, to get a good skill, I tell people it takes three iterations. But when we say iterate, it's like, okay, well, what does that mean? Right? Like really tactically, brass tacks, how do I do that? So what I tell people is take, you need to have three inputs that are very similar.

(37:12) So let's say we're trying to make a Excel file of financial report, right? Or we're trying to do some sort of manipulation. You can have input one and you're going to prompt and you're going to say, all right, I really want this to actually be a cleaned up Excel file with my brand colors. And I want the tabs to be really nice.

(37:26) The first tab needs to say, start here. And then everything else needs to be logical from there. Give it your best shot. Cloud's going to make an output. You're going to say, you missed the mark about 80% of the time. And then when you revise it, you say, okay, great. I've revised what you worked on. Make me a skill to do this, like my revision next time.

(37:45) You take that and you go to input two and you say, great. Hey, here's input two. Use the skill I just created. That's going to make another cloud output. You're going to revise that because it's going to miss something, but it's going to be much closer, much better. And then you get the two of that skill. And then lastly, you take input three, which is unique from the first two, apply skill v2 to it.

(38:06) You get an output. Maybe you need to revise it, but after you revise it and you say, great, update the skill to my revision, that gives you v3. And v3 is something that I'm proud of because I've taken the time to actually iterate with it, tell it what it got right, tell it what it got wrong. And it knows I'm crystallizing my information and my process.

(38:26) That way I don't have to keep re-explaining myself. And I think this is the aha moment for people when they realize that they can go from zero to their finished product, especially in the reporting world where reporting takes freaking hours. Like some people's jobs is just reporting, right? And fine, but I want to make your life easier so that you can actually go build more relationships, focus on strategy, be creative, or do quality assurance.

(38:50) Those are the four main categories that people should be spending their time on. This is how you get that. This is how you remove copy and paste from your life. So hopefully that's a little bit more of a, like, take these steps to make a really high quality skill. This is what people pay us to do. Love it.

(39:08) Well, Grant, thank you very much for being on here, sharing with us. If anybody wants to connect with Grant, he is on all the platforms, but I'm happy to connect you with him as well. You know where to find me, Andrew, at foxwell.com. Until next time, though, folks, keep on building. Bye. All right. Bye.

Andrew Foxwell | Co-Founder of Foxwell Digital

Co-Founder of Foxwell Digital, a social media advisory firm focused on honesty and transparency across paid social. Through its membership offerings, online courses, account management, and consulting services, Foxwell Digital helps brands and agencies make better decisions and scale sustainably.

https://foxwellfounders.com/
Previous

How Torii Rowe Built an AI Operating System for His Entire Agency