Building Portable AI Workflows That You Can Take Anywhere

Hey, before we get to today's show, I want to tell you something we just released. It's our 18th annual Social Media Marketing Industry Report, and it's out now. Here's something that jumped out at me as I was preparing this report. Two-thirds of marketers say the pace of change is overwhelming. If that's you, you are not alone, and I want to help. Download this free report. It covers everything from platform shifts to AI adoption, to the growing divide between B2B and B2C marketers. And you can get it for free right now by visiting socialmediaexaminer.com/report26. socialmediaexaminer.com/report26.

Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner. Hello. Hello, hello, hello.

Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators, and business owners who want to know how to put AI to work. There are really big advantages to building AI workflows that you can take to any platform, and today we're going to explore how to build portable AI workflows. My special guest is an AI expert who helps B2B marketers improve their results with AI. Her course is Foundations in Generative AI for B2B Marketing. Nicole Leffler, welcome back to the show. How you doing today?

I'm great. Thank you so much for having me. I always love getting to talk to you.

Super awesome to have you back. So let's start with this. We're here to talk about AI workflows and you and a lot of our audience is very much excited about this, but there might be some misconceptions that people have out there about AI and workflows. And I'm just curious from your perspective, What do you see as maybe one of the biggest one or two things that a lot of people seem to get wrong when it comes to workflows and AI?

There's a lot, so narrowing it down to one or two, but I would say right now what I'm seeing is a couple of the biggest are one is that agentic is always better than the chat interfaces. Like everybody's gotten very excited about the autonomous, agentic, all of that, and it's cool, but there are still times and places to be using ChatGPT or Claude in chat or Gemini in chat. And so I think I think that that's probably one of the biggest misconceptions is that it's always better to build an agent. The other one I would say is like really people getting like pigeonholed into a single platform and building their entire structures and tools and everything for one specific platform. And this misconception that it's like you have to do that and that's the only way to do it is to really be building specific to Claude or Codex or ChatGPT or whatever, or Copilot, whatever you're on, versus thinking about it in a way that allows you to move from platform to platform easily.

So I am curious about that. I think I can understand this because a lot of people, once they get into a tool that they really like, they kind of stick with it and they just say, this is where I'm going to build my house. What's the downside to that? Like, help everybody understand why that might not be super wise.

So there's a few different reasons. One of the biggest ones is stability. Like, if you're actually building things that your business depends on and your operation depends on, there are days where either a tool is completely down, like this happens, right? Like just it goes down or it's just not responding, not working. And like, what, are you just not going to do any work? Like, what, you know, is nobody in your company going to be doing anything because Claude is down or ChatGPT is down or whatever it is? Could be a matter of minutes, but sometimes it does last a lot longer. And the more ingrained we get with AI being a part of our workflows, that becomes a bigger risk that you're going to lose being able to do anything if they have an off day. The other thing is sometimes the tools change performance. I mean, I've been using these tools since basically 2 days after ChatGPT came out for the kind that we have now. I was using it even prior to that in other forms. And we have seen that, you know, when they're making upgrades and updates and sometimes like things get tweaked and it just does not perform in the way you thought. And sometimes it's just like, oh my gosh, this tool has been off for a few days and it's just not giving me what I want. And to be able to flip the switch and go, okay, so I'm going to use a different model, I'm going to use a different platform. And not lose all of that infrastructure you have is huge. And then the other big reason, and I think this is something I have not really heard anybody talking about, is you give yourself a lot more flexibility on the financials of this because we are right now operating at a price point with all of these tools that is not gonna be forever, right? We don't know exactly what is gonna be happening, but we do know that realistically you have really had early adopters using this technology, right? Like it's been pretty early adopters, more and more and more people, companies, infrastructure, like all of it is getting moved on to this tech. They only have so many advanced AI chips and there's constraints in the supply chain. This has always been a problem. It's been an issue. It's going to become a compounding problem and issue, not just the chips, but the compute and everything from the data centers. And so decisions will be made like we are highly likely to be in situations where decisions are made either of throttling and you get less access to that, or you have, you know, the price goes up because supply and demand and it's economics. Like, we don't actually know what's going to happen when more and more and more people pile onto the platforms and the chips and the compute and all of the things that run this does not increase at the scale that, at the pace that the people are piling on. If you can move platform to platform easily, you're going to have a lot more resilience, whether it's that they throttle how much each person or each company company can have, or that they end up going the path of raising prices but another company doesn't, you're going to have a lot more resilience if you are not stuck into a single platform when all of this happens.

Very well stated. I also want to state that, um, really two of the biggest models out there are privately held companies, uh, Anthropic and OpenAI, and they are going into massive debt. And right now they're effectively giving away the technology for like $20 a month. And some of them are getting smart about it. Like Claude is putting caps on how much you can use. And I think that's fiscally smart for them. But eventually the gravy train is going to run out, you know, and we're going to have to pay the real prices because it is very, very expensive for these things to do what they do behind the scenes.

Exactly. And so I think just putting yourself in whatever position you have to be able to be resilient to however this all plays out You can choose to build in a way that doesn't force you into one ecosystem and one ecosystem only. And to me, that's just a wise way to operate, knowing as much as I know about how much we don't know about where this is going.

I like that. Now, one of the other advantages to portability might be function.

Yeah.

Right? Because you wanna talk about that a little bit? Because these models are all slightly different, are they not?

They are all slightly different. Like, I will say though, like the mass majority of things that like the majority of marketers are doing, you can easily switch between them and like you're not going to have dramatically different functionality that you would need in marketing. I mean, there may be some minor things, but like overall you can easily switch between the models. That said, it does give you a lot more functionality. Like if most of the time you're using Claude and then, but like you need to do that same exact thing, but now you need access to be able to bring in like the image models inside of ChatGPT or OpenAI AI ecosystem in, you can easily just flip over which one you're using. Obviously, things are probably going to be different by the time even people are listening to this, but like, you know, Codex came out with some features and functionality this week or in the last few days that are not available inside Cloud Code, if that's your ecosystem. So being able to go, "Oh, I wanna take advantage of that. I wanna be able to use this in that ecosystem instead of this ecosystem." It's just really powerful to be able to switch between them.

Yeah, and it's really very interesting because Gemini in particular is one of the few models that was built multimodal from the ground up, meaning it has the ability to see.

Yeah.

And understand the world. We're getting there with all these other models, but we're not quite there yet. You know what I mean? Like the ability to understand and process video isn't necessarily there yet with all the platforms. I'm sure we're all racing towards the same end destination, but it is true that some models are good at certain tasks that, that maybe are superior than the other models and being able to flop back and forth is highly advantageous.

Yeah.

So let's just say that we want to actually begin with the concept of building something that allows us to kind of go wherever we want to go. Because like you said, we don't know what innovations are about to come out, right? Like right now it's pretty much 3 major players. It's Google, it's Anthropic, and it's OpenAI. And then there are other players, right? Overseas players, you know, and there's X in there and all these other things. We have no idea what's going to come next, but Wouldn't it be interesting if all of a sudden there is a new player that comes on the scene?

Yeah.

As there always is with all the other platforms, right? Like with the social platforms, you had TikTok come in, you know what I mean? And all of a sudden, like, came out of— came out of nowhere, right? So having built something that can, that can allow you to go wherever the greatest and best tech stack is, is pretty advantageous. So like, where, where do we actually start with all of this? Because so many people are probably like me, and they're just using one model most of the time.

Yeah.

So, where do we begin?

So, I think that it's really starting to think through, like, how are you storing all of the pieces that you're putting into that model, right? Like, how are you doing this? And a lot of it is just a matter of having, like, copies of your instructions that you can use in a format that is, like, really well organized outside the model. It seems super simple and silly, but it's like, you know, if you're building, let's say, skills in Claude, for example, you know, skills are actually a portable format that can work. Like, I think a lot of people who are in that Claude ecosystem know that skills, which I'm gonna stop and just tell you what a skill is before I go too far in this. Yes.

Yeah, yeah, yeah, please do.

Skills are essentially, without getting overly complicated, it is a zip file that contains a markdown file with instructions of how to perform a specific task. So it's just written in markdown of like, this is what you do to perform this task I want you to do. And then that markdown file can also direct, it doesn't have to, but it can also direct like, here are assets you should be using. So it could have like templates or logos or even code and resources and scripts of things you want, it wants to run. So assets or resources, like, scripts, things like that. So it could have calculators built into it to do the calculations or code it's supposed to run. And it can also have resources. So it could have files and documents and things like that all zipped together. All a skill has to have is that markdown file in the zip file, but it can have a bunch more attached to it.

Real quick, do you think it's kind of like Neo in The Matrix when he learns kung fu?

Yeah.

I mean, do you understand everybody who's like, he just downloads the skill of kung fu and before you know it, he knows how to do kung fu? Kung fu, right?

Oh my gosh, I had never thought about that. That was the perfect explanation of what a skill is. I do want to pause before, like, and I don't want to go down a rabbit hole with this, but I do want to tell anybody listening, a million people are on the internet giving away skills, selling skills, all of this. Please do not take other people's skills. Do not download them. Do not buy them. They are one of your biggest security vulnerabilities that you have with AI. And I This is something nobody is talking about as much as they should be, and they need to be, because those skills, like I said, it's not necessarily just written directions. It can run code, it can tell it to connect to your external data sources, it can contain like images that could have viruses in it. And so, you run the risk, if you download somebody else's skill to download kung fu abilities, you could also be downloading something that you don't even realize is telling Claude to connect to your data source, you know, to your CRM, extract customer data and send it here, and you are not even the wiser. So do not be taking all of these skills libraries. It's not the same as when people were like, take my prompt, right? Like, this is not the same. They're sharing it in the same way, like, take my prompt. It is not the same thing. So unless you are really sure this is a reputable source, and I'm talking like it's actually Anthropic, it's actually OpenAI, it's Google, I would be very very, very careful. And if you have a company, do not risk downloading somebody else's skill into your company account and risking what could happen with that.

Let's talk about why in the world we even need skills, because I've had a number of people on the show and I feel like I can't talk about this too little. I want to keep talking about it, right? So here's what we know so far. We know that a skill is a set of instructions to do very specific tasks or to take on a role of some sort, right? Like, uh, marketer or kung fu teacher or whatever. So why would we use a skill? Just help everybody wrap their brain around why that's different than, for example, creating a project. Do you understand what I'm asking?

Yeah. So one of the coolest things about skills is that one, they're composable.

Which means what?

Glad you asked. It means that you could be in a conversation and you could use multiple skills. Ah. So you can give it like, you know, you could use 5 or 10 different skills in the same conversation. Starts to make it a little more agentic, even just like in a chat. With a project, you're using one project at a time, right? So it's different. There are ways of like ChatGPT, you can compose, like use multiple GPTs, but in general it's designed you're using one GPT at a time. Where with a skill, so like let's say for a marketer, you have a skill that is like all of your brand guidelines and it teaches how to review to make sure it's on brand. Mm-hmm. And that you're doing everything in alignment with your guidelines. You could use that anywhere in anything you're doing, and you could use that in like as in a project that's using other skills that do other things because that's just the last step is to review.

So it's something you can call on from anywhere within the interface is effectively what you're saying, right?

Yes. And then the other piece of it is, and where I started and then we kind of were coming back around to, is it like maybe you created this skill for Claude Chat. You can actually download that skill save it with all of the components. This is how people are sharing them online, right? You save that skill and now if you use Claude Cowork, you could upload it to Claude Cowork and have it saved in Claude Cowork. You could do it if you're in Claude Code, but it's not just an Anthropic thing. You can use it if you have ChatGPT Business or ChatGPT Enterprise, you can use skills in chat, which nobody even seems to have noticed that ChatGPT added skills. They're like, I use Claude 'cause it has skills. So does ChatGPT, right? So does Codex, which is the OpenAI equivalent of—

I think Gemini just recently started announcing that they're working with skills.

Yeah, I don't know if it's actually live or not, but they definitely have said something about it. It's so hard to keep up day to day of like who's actually implemented what if you're not like in that tool for a project. And Copilot is adding skills as well. Again, I don't think it's live. Well, I think it's like in the beta program, although by the time people watch this, it might be live. In People's Copilot. So all of this is so, so powerful. And then like the Claude skills, like I have Claude skills that I built inside Claude and I can use them in Claude for Excel or I can use them in Claude for PowerPoint. So you can start using these skills in other locations and other harnesses and that becomes a very powerful portable feature. Now they may operate slightly differently, But like, I have skills I built for Claude that I also have in my ChatGPT that I use in Claude, in Claude. I use it in Claude and Cowork. I use it in ChatGPT. I use it in Codex, and I use it in Claude for Excel and ChatGPT for Excel. Built it once, uploaded it to all of them, can do exactly the same functionality anywhere I wanna do it. And it starts getting really powerful for portability when you do that.

How do we actually make a skill? Let's talk about that a little bit because obviously we haven't addressed that yet.

We could do an entire course about skills. I might actually do a course on skills. I've thought about that. I think I might because that could be an entire course. We're not gonna like do the whole thing in this conversation, but the shortest version of it is the easiest way to do it is either in whatever tool you're using. So like at least I know ChatGPT and Claude and Codex. They all have like a Skill Creator skill. And so you can literally just chat and it can help you build the skill. This is what I want to do. This is the information I have. This is the context. A lot of times I'll just give it a prompt that exists that I use as like a recurring prompt or GPT directions or something like that. Be like, I need to turn this into a skill and then it will turn it, it will write the information. It'll, you know, ask, do you, you know, if I say I need to add this template to the skill, you just explain it, and they have a skill that creates skills. It is actually, um, I believe it's the only pre-installed skill on all of these platforms. So like anything else, like it's just automatic. I mean, that's not true because there's a few that are built in, but like, so for Claude, yeah, for Claude users, here's what I remember.

You have to go under your settings and somewhere in there is this skill creator. I think you have to turn it on, but it might be on by default now. But basically what it does, if you have a really good conversation about something that you think is a repeatable thing that you're gonna do, you can just simply say, make this into a skill. Is that effectively what we're talking about?

Yeah. I mean, if you do it this way, you're definitely gonna have to like test, validate, like all of that. But yeah, so that's like the simplest way to make a skill. You are not gonna sit out, sit down and type out a Markdown file, right? Like you are not gonna write a Markdown file. You're not gonna make a skill as good as the creator is. Now you could give it a prompt you already have and have it turn that into a skill and improve it because it knows what goes into a really good skill. But yeah, that is like the easiest way is to let the skill creator help you create the skill. Now, obviously everyone's a different level. If you have like, you write code and you want your skill to run your scripts and your code and that, you may want to write that code or you want to, may want to build code with, you know, with Claude or Claude Code or any of this to do specific things. Like you could get real complex and powerful with your skill. You don't have to though. So like the beginner version of this is just chat with the skill creator and have it make the skill, test it, iterate, fix it. You know, you can go in and edit it and adjust it. And you can do that in really any of the tools that have skills, have a skill creator skill.

An example that I can explain to everyone is I've created a really good copywriter for Social Media Marketing World, our conference that probably already happened by the time you hear this. And, I had a spreadsheet of all these testimonials and stuff, and we don't have like Teams accounts because, because we haven't put the money out there to do it. But I have another person, my director of marketing. So I created a skill from this for him and I ported it over to— I gave it over to him and this skill had some reference files in it. And I asked basically Claude to just go ahead and create the whole thing and zipped it up and it had all the reference files in it, you know, the Excel CSV file and all the stuff. And now technically I could send that to anybody who wants effectively the version of my Claude project that I had spent a long time refining. But now it's like a standalone skill that they can use without having a project at all and they can just call on it, which is kind of cool.

Yeah.

So we could talk about skills a lot more, but I know we have more to talk about, but keep going. Go ahead.

Yeah, but if you are in like team and enterprise type accounts, right? Like anybody who's got like multiple people on the ChatGPT account or Claude account. You can share the skills with your team too. So, like, you can either build for yourself or build for the team.

Oh, that's cool.

And so, it's just good to know, like, you can actually just, like, add it as something that gets added to their accounts as well, or give them the option to add it to their accounts as well. It's not a problem to share within your own team. It's just a problem to get it off from, like, some random person's GitHub on the internet, right? Like, that's— Right. That's where you run into the risk.

Okay. So, the big thing towards portability is to have skills and presumably to download those skills and put them somewhere safe, like maybe in your Google Drive or on your hard drive or something like that. So you have access to those skills if something crazy happens, right?

And you know what? Like that brings up a good point. If you have projects, you have GPTs, you have GEMS, like you have Copilot agents, you should have those outside your AI too. Like having it someplace safe, downloaded someplace safe, that should be a best practice no matter what you're doing. Because that is key to portability also. So, using skills is one way to have that portability, but even just having GPT project, Gem agent instructions, and when I say agent in this context, I mean Copilot. Having those instructions, it lets you, you know what, a ChatGPT, GPT, a Gemini Gem, a Copilot agent, you put those same instructions in any other, Claude project, put the same instructions in any of those, they work. They're gonna give you almost the same. Obviously each model is a little different, but it's really easy to make that portable as well. It takes 30 seconds to go up, you know, copy and paste those instructions and upload the files. So that also is a portability thing. If you don't wanna get into skills, if you're like, everything I do is GPTs, fine. Copy your instructions. You know, have a file in your Google Drive, your Dropbox that has, here's the instructions, the assets. Sets, the connectors that I have. It's just documented. So something happens, you need to do it in Claude, you can go recreate it in Claude. Something happened, you know, in 5 minutes, not in 5 years. Like, it's not that hard. It's just a lot of people aren't doing it.

Okay, so what's the next part of this process to kind of design workflows that are portable?

So one of the big ones is how you're doing like data connections and context, right? So you could be uploading, like the skills give you an opportunity to potentially have portable context if you're using the skills format. But you could also be using MCP connectors, which are essentially like the simple version is an API for your AI to talk to other tools. And MCPs are very portable. You could connect them in multiple different tools. So that is one way to have portability between data connections. You could also, you know, just having like really solid context engineering. So like knowing exactly what context. So you just keep those files so you can bring them tool to tool that you're not trying to give it your entire asset base, right?

Explain what you mean by this context thing.

This kind of goes back to a misconception with the AI. A lot of people seem to be under the impression the AI is actually going to perform better if you give it access to like every document you've ever created in your business, and it can search through it, and it can identify what does it need to do this project and run with it. The reality is that is not accurate. The reality is the AI will do far better if you just handed it the correct document and did not make it spend its thinking and energy and like all of the GPUs and all of the tokens and cost and all of that, looking through all of your stuff and making its own determination and thinking about what is or is not relevant. So if you actually take the time on the front end to know what's relevant for different types of things you do on a regular basis and actually just having that set together in a specific spot that is portable, like a lot of times that is not something that's going to be updated frequently. It's not— it could just be a link to a specific page on a website. It could be you know, you could do a specific file that you have a connector set up to, or an MCP set up to a specific file or folder or something like that. That's easier to be portable because it's always gonna send the AI the same context no matter where you are. You don't have to get like each different AI to be able to search through all of your stuff to find the right thing.

I love it. Let me ask a couple clarifying questions.

Yeah.

With a lot of these things, we know we can upload PDFs, we know we can upload text files to all these different AI models. But it gets a little wonky because all of a sudden you update one and you forget to update the other because maybe you're using Gem and Claude and ChatGPT. So does it make sense to have these central areas? And if it does, does it also slow things down a little bit because it's got to go out and retrieve that information? Like, let's talk about that a little bit.

I think, like, to me, it makes sense from a portability standpoint to have them in a central area. And to have it— that, but then again, it's not your entire Google Drive you're sending it to. You're sending it to a specific place, a specific Dropbox folder, a specific Google Drive, a specific SharePoint. So it's not— it's not going to slow it down if it doesn't have— or not much if it's— I mean, it has to open and read a document if it's part of your, you know, your project too. If it just knows, I connect, I go get this, I bring it back, The amount of time it's going to take is very minimal, and then you're not going to have to constantly be updating it lots of different places.

Well, and I like the idea that you can update it in one place and then all of a sudden it's just like updated kind of in all places because it's referencing that one source, right?

Right, exactly.

Hey, just a quick interruption. I want to share something with you. If I told you that 62% of your peers are using AI tools every single Would that surprise you? Well, that's the data that we found in the recent study that we just did. In fact, nearly half of marketers now say that video is their most important content format, more than written, more than visual, and more than audio combined. These are just two statistics, two stats that are in our brand new industry report. It's our 18th annual study. It's 44 pages long. And over the next couple of days, you can download it for free. Free if you visit socialmediaexaminer.com/report26. Grab your copy now. Okay, let's get back to the show. All right, so, um, let's kind of connect the dots a little bit. You mentioned earlier that you can actually call AI inside of Excel, and, um, you and I, when we were prepping, had a really cool example of how this is working. And I want to spend some time actually explaining what the heck you did here because this is pretty innovative. And by the way, this probably wouldn't be possible if you weren't able, if you weren't doing all the things that we're talking about here. So kind of set the stage of how you tie all this stuff together in this example.

Yeah. So I'm going to tell you guys my absolute favorite workflow right now that I am doing and explain all of this in the process so you understand. So what I will do is I will use Excel and Excel has a ability to have like, you can with Claude, you can download the Claude for Excel plugin. It actually, it's not like a Chrome plugin or anything like that. It's like it actually gets activated inside of your Excel. You're using your Claude account inside the Excel workbook. It can manipulate the Excel workbook like an agent, right? So we can actually go in and edit and adjust and go tab to tab and create formulas and put them in the spreadsheet and do all the things a human could do in a spreadsheet for the most part.

Just for, uh, the Google users out there, is this in the cloud or is this on an Excel app? Because it's been a while since I've used Excel.

This is actually Excel specific.

But does Excel have a cloud version and an app or does it work with either one? Do you understand what I'm asking?

I gotcha.

Um, like, are you actually like on Excel's website doing this or is it actually the app? App?

So I'm using the desktop app.

Okay.

On the desktop app. So you can download the plugin for Claude that lets Claude do all this. That plugin can use skills from your Claude account. You can also have custom instructions specific to the plugin. So like anytime you open Claude in Excel, it reads those instructions. ChatGPT also has one of those. So it's a little bit di— I don't think you can put Custom instructions.

Oh, ChatGPT, one of those means also works with Excel.

Yes. Oh, ChatGPT for Excel.

Okay.

Uh, Microsoft Copilot has their own version of this that's like built into Excel. Again, I've only done any of these on the desktop. Okay, so those three inside of Excel, and you can have all of them in your Excel, all three of them that you can open and chat with. It will understand anything in that workbook. It can skills and it is like full-blown abilities that you can do in that chat. Like it's the same model. So it's like really smart, just like if you're doing anywhere else. And with Gemini, now Gemini, you can use Gemini inside Google Sheets. So like that would probably be the closest equivalent. I haven't done a ton with it, but like it's not quite as sophisticated as of the time I'm like you know, last checked, which was a couple of days ago. By the time you're seeing this, I'm going to guess they're going to be like catching it up because it's Google. They can. So what I do, I turn my entire Excel workbook into an AI agent that can work with any of— it can work with Claude, it can work with ChatGPT, and it can work with Copilot. So if one is out, I can just switch to another. There is no tie of which one I'm using. And the way I do this is I start out with— I, you know, what my favorite tool— I like to use this in Codex, but you could do it in ChatGPT, you could do it in Claude. I tell it what I want to create an Excel workbook to do. Like, this is, you know, we're creating, um, a, a tracker that I can chat with when I do consulting work with a client, and it's going to understand the contract, it's going to understand exactly how much time I've used, how to record it, add it to my dashboard, write a briefing for my client, all of this, and it lives in Excel. So I'm like, what is our plan to do this, to build this AI workbook? We'll flush out like, what should this workbook look like? How does this work? And then I say, okay, great, now hand me the files, like make me the files to brief an AI agent that lives inside Excel Excel so that it can make this workbook we just came up with the concept of, all of the formulas and everything.

Okay, real quick, are we now inside of one of these AI models building out this skill that you plan to deploy inside of Excel, or are we inside of Excel right now? I'm a little lost. I just want to make sure.

Yeah, sorry, sorry. So right now I'm in— not in Excel, I'm just in like either ChatGPT or Claude or Codex, not inside of Excel, and I'm just flushing out the entire concept of what I want to build and like thinking it through, talking it through, going, what should the structure and strategy be around how we build this workbook? What is all the tabs? How do they tie to each other? What are the formulas? All of that. We're just chatting through it, right?

Okay, perfect. So let me just bring everybody up to speed because I have seen this with my eyes. I'm going to kind of describe this. So what we know so far with Nicole is Nicole had this idea. That she wanted to use Excel and spreadsheets have tabs, you know what I mean? And you can do all sorts of stuff. It's like a big database, for lack of better words. You can, you, she can use this to do all sorts of creative stuff. So she's interacting with her AI platform of choice and she's saying, hey, this is what I want to do. And she's building, for lack of better words, the set of instructions. Is that what I'm hearing you say, Nicole?

First off, we're starting with this, like just figuring out what the plan is. Like, what, here's what I want to do. How do we do this? Like, how do we How do we actually do this inside of Excel? What does this look like in an Excel document for it to have this functionality that I want it to have? So that's what we flesh out. Once we have this figured out of this is the plan of what the workbook should be, then I just say to my AI agent, again, outside of Excel, I say, great, I have an AI agent that lives inside of Excel and I need you to put this into a briefing. That has all of these step-by-step instructions for how the AI agent can build this, give me a Markdown file briefing. And if there's any, like, extra data that I need it to bring in, so it's like, sometimes I have it do research of, like, categorization ahead of time and things like that for different things that are gonna live in this, 'cause we go down, like, the assets that go into this workbook. I'll say, put those into JSON files that I can give, give me everything I need to give to the AI that lives inside of Excel so that it can build exactly what you and I just planned.

And we're creating a skill effectively here, right? Is that what we're doing?

Mm, not yet, no.

Okay, so this briefing, this markdown file, keep going. Where does this come into play?

So it gives me this markdown file. I now take this markdown file, I open Excel, I pick whichever one I use. Like, I usually build with Claude just 'cause I like the aesthetic of it the most. But I open Claude in Excel. It will work perfectly with ChatGPT in Excel. It will work fine with the agent, like the Copilot I file it with the agent that can edit your workbook in Excel, and I say, here is a briefing of a workbook I need you to do and all the assets you need. I upload them to the Excel file, like to the, to the AI that lives in Excel, and I submit it.

You're just dragging this into a little chat interface on the side that shows up to the right. Yeah, okay, good. Keep going.

Just the files the other one gave me. It's literal, like this is prompt. I worked with another AI to write you this briefing. Make it. Here you go. Like, that is literally all I do.

And what happens?

And it runs it, and it like goes and it puts it all together. And I'll get like— you mean it may take 10 or 15 minutes? I'll go drink.

So it starts building out in Excel something for you.

It literally builds the entire thing out in Excel for me as a starting point. And then, okay, I'm gonna step back to that initial conversation because this is where a lot of the magic happens. Part of what I ask for is what I want to be able to do functionally with it from an AI standpoint, not just building it. So like my consulting tracker, I wanted it to know, how do you update this for me from a verbal— like I could just tell you what I did. How do I have the AI add all of the updates and understand? I want to be able to say, okay, I did the update, I want you to add it and I want you to write the email recap to my customer based off of this Excel. When I built that initial plan that I handed over to Excel, to the agent in Excel to build, I ask it to write prompts for the AI that lives in Excel to be able to do all of the things, right?

Where does that live?

Both live in tabs in the Excel file. So each tab has a prompt for different different things that I might do inside this workbook. It just becomes these prompts that live in there. Then that is— that's all, and that's all in that briefing and assets and everything that I handed over to the agent. So when this workbook gets built by the agent that lives in Excel, it also has all the instructions that the AI needs to be able to work with me in the workbook, because that's just part of the structure and strategy we put together. Okay, so now I can go open that workbook once it's built. We may refine it. Sometimes you see it and you're like, oh, it'd be really cool if we add this or that. You can just chat with like Claude in Excel or ChatGPT in Excel and ask it like, could we add a tab for this? Or, oh, now that I'm seeing this, could we add this to the dashboard? And it'll just make it all for you. So you build this whole structure, you have it in Excel, and now you have an AI agent that lives in there. But what's really cool in this case, up when you and I saw each other last time and we were talking about this was that day Claude was down because I wanted to show it to you and Claude was down and I went, oh wait, I can just use it with ChatGPT. ChatGPT had the prompts, ChatGPT had the context, ChatGPT had everything built in. Now, where skills come into this is I have a skill in Claude and in ChatGPT that say, this is what you do anytime I'm using you inside of an Excel. Like, anytime you're accessing a spreadsheet, you need to go through and you need to read every tab that has anything about AI agent instructions, any of that. You have to read that before we even begin, before you do anything, and follow those instructions for anything you're doing in this workbook. So my skill tells it how to read the workbook to turn it into an agent for me that can always do it I want it to do it. That's where I'm using the skill. It's the skill is like to turn it into my AI agent. That skill is in all of them. If you didn't have the skill, you could just say, read the tabs that have your instructions before we start. Like, it's not the end of the world if you don't have the skill, but you could also have skills that you pull in to bring in context or other things that aren't in the workbook. Like, you could get way more sophisticated with this through skills if you wanted to. I'm just using the skill to say, hey, check the tabs for your directions. Also in this skill, it gives it directions, or maybe it's in my— no, it's actually, it's in my custom instructions on each. Uh, one of the tabs is a history log. So anytime the, the AI— and part of its instructions is anytime that it makes any kind of an update to my Excel spreadsheet, it is logging the history. So my agent has memory. My agent knows every single thing we have done, and it is cross-AI memory, because when it starts, it reads all of its instructions, it reads its history, it understands the workbook, it knows everything that's going on. And so when I switch it to ChatGPT, ChatGPT is going to read all of that and it's going to pick up right where Claude left off. I switch it to Copilot, it's going to pick up right where ChatGPT left off. So you now have a completely portable AI system lives in Excel of all places, and it is so powerful. So if you are collecting, like, tracking data, tracking, you know, anything like that, any kind of data-based anything, this is a really powerful way to build an agent that you are not dependent on any platform.

Okay, I love this, and I just want to reiterate a few things and get a very specific question I think people might be asking in their minds right So first of all, we are using our platform of choice and we are creating, in this case, a Markdown file that we use a little bit later once we open up the app that has a plugin that allows that, that like Claude plugin or ChatGPT plugin. Let's just stick with Claude for simplicity. We've opened up Excel, we've activated the Claude plugin, we drag in the Markdown file. The Markdown file as work that's been done on another platform to help train basically Excel to know exactly what the heck it's supposed to do, right?

Yeah.

And then, but the skill side of it, this is the question, is do the skills live— do they also get dragged into the thing, or are they accessible because you've connected your Claude account? Do you understand? Like, like, where is it finding the skills? That's the part I'm—

so with Claude, the skill lives in Claude.

Okay.

And so like, in the context, like when I'm building this, for the most part, the skills I have used so far in this, like I've done several projects this way, the skills I have used are really like, it's the skill that tells it how to use the workbook.

So when you connect this Claude connector, effectively it's smart and it knows the skills you've already created and it somehow calls on those skills when you call on those skills, right?

Yeah. And you can just call on it too. So you could do it either way, but I have in my custom instructions to use that skill in Claude, like inside of Excel. But really the only skill I'm currently using is that, that it just tells it to read the tabs. That said, I actually am working on a project where it's gonna have a lot more in-depth skills because I want it to be able to do certain things that I'm not just gonna be able to put in a tab in Excel so it would pull it in. So in Claude, it's just uploaded in your Claude account. It pulls through automatically. With ChatGPT, you have to actually let it know like add the skill in Excel. And you're only going to be able to do a ChatGPT skill if you have a Business or an Enterprise account. They don't have skills yet, at least— well, by the time we're watching this, maybe they will. But like, you have to have one of those two account types. Um, and then Copilot, you're not using skills. So that's part of why right now I'm like thinking through the skills and which ones I want to be able to do this way with skills. because eventually everything's gonna be skills. But like, right now, I want it to be able to live alone. And part of this is that portability, that it doesn't matter what's down, I will still have an AI that can run my platform, whatever my project is.

We've just scratched the surface of what is possible here with this portability, because if we think about the multiple dimensions we've talked about here, at the very beginning, it's about Hey, at the very least, get some skills and download them and put them somewhere safe so that you can bring them up into any particular platform that works with skills. And it's not like you're starting from scratch, you know, and you get a chance to like kind of like download the Kung Fu skill, you know. And eventually we're going to make lots of skills and we're going to have a lot of skills and those are going to be like our secret sauce, you know, that's going to allow us to do our job in a way others don't know how to do because we put some energy into developing these skills. But it gets next level when you start now connecting existing existing applications like Microsoft Excel with some sort of a plugin, for lack of better words, that allows you to tap into that AI platform of choice. And now all of a sudden, like, you can do things that are beyond your wildest imagination, right? And that's the exciting part here, isn't it?

I mean, this stuff is so cool. Like, you saw the, the tracker that was for my hours. I'm building another one, and I probably shouldn't say this on a podcast. Nobody— this is just for like what I'm doing. Do not copy me on this because I know what I'm doing and you might not. So please do not run with this idea. Um, I am doing an entire, like, retirement management, like, everything in my IRA, like, a full-blown AI-based tracker, and it's all being built this way. So it's, like, gonna keep all of my investments in this, like, wild, crazy organization and build me dashboards for every stock I own and all of this stuff. Um, it probably is completely unnecessary, to be totally honest. It's like an entire decision-making thing so that I know what I've got.

It's going to live inside of Excel.

It totally lives inside of Excel. And so, I mean, it's going to be able to allow me to like understand my portfolio in a way that like no existing like platform does, because it's going to be able to like categorize off of why I bought it, not because of like what somebody else says of what a company is. So I'm like, oh, I like have data center cooling companies. I want to see all the data data center cooling stock I own and how much exposure I have in this. And it can do all of that. It's just like very cool, but it's all built inside of Excel, managed inside of Excel, updated inside of Excel. Um, it's like a living and breathing AI agent that lives in Excel to help me manage my retirement, essentially.

And folks that are really creative are like, oh, I could probably do this with Airtable, and I could probably eventually do this with Google Jump— uh, I mean Google Sheets, you know?

Yeah.

What's really fascinating here, folks, is so many of us that use AI, we have to use a special platform that is not what we use to do our everyday work. And then we have to take that information and get it back into another platform that is what we use in our everyday work. And what I love about what you've done here is you've figured out a way to kind of, um, allow these things to interconnect in a powerful way and still work collaboratively together. And I believe that's the future. Like eventually we're not gonna care what app we're using.

Yeah.

It's just gonna be able to do the job. And who knows if we'll even need Excel in the future. We might just have the ability to do all this stuff natively with all these platforms. Nicole, I know we've just scratched the surface of what's in that brain of yours. First of all, thank you for coming on and sharing your insights with us. If people want to work with you or want to connect with you on the socials, where do you want to send them?

So my LinkedIn is my primary social, so it's Nicole Leffert. It is a verified account, so make sure you're going to me. And my website is nicoleleffert.com. And I have, right now while we're talking, I have my Foundations of Generative AI for B2B Marketing course. I do have a couple of other courses kind of spinning around in my head that I'm working on. Um, by the time this comes out, I might have a little mini course on how to do this Excel thing I'm talking about, cuz I've had several people ask me.

Well, and you already hinted you might have one coming up on skills too as we were working.

Yeah, so, uh, I don't have exact release dates, but if you're looking for that, definitely take a look. So, um, yeah, but I would love to connect on Excel— I'm sorry, on Excel— on LinkedIn. I've clearly been in Excel way too much lately. Uh, connect on LinkedIn, and also my email is just nicole@nicholeffer.com if anybody wants to reach out.

Nicole, thank you so much for sharing your wisdom with us today.

Thank you so much for having me. This was fun.

Hey, if you missed anything, we took all the notes for you over at socialmediaexaminer.com/a106. Be sure to follow the show on your favorite podcasting app, and if you've been a listener for a little while, we would love a review. Do let your friends know about this show, and also check out my other show, The Social Media Marketing Podcast. This brings us to the end of the AI Explored Podcast. I am your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day, and may AI help you become more successful.

The AI Explored podcast is a production of Social Media Examiner.

Hey, thanks for listening to this week's episode. Before you head out, I wanted to personally recommend that you download our brand new industry report. I've been publishing this study for 18 years. It's a long time, and this year's data points to a really clear picture of where the industry is headed. Whether you're focused on organic, paid, video, or artificial intelligence, it's all covered in this 44-page report with more than 50 charts. And right now it's totally free. Visit socialmediaexaminer.com/report26 to download it right now. Socialmediaexaminer.com/report26. See you next time.