327 | Claude vs. ChatGPT Is the Wrong Question - learn how you can use them together with Isar Meitis

0:00GMT20260914-185903_Recording_avo_1280x720:Hello and welcome to the Leveraging AI podcast, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. And I've got a really interesting and really productive and helpful episode for you today. If you have been following this podcast for a while, you know that I share many times that I live mostly in the Claude universe, but that I also work in the other platforms regularly.

0:25GMT20260914-185903_Recording_avo_1280x720:And I said multiple times on this show that I can jump back and forth between the different platforms seamlessly without more or less doing anything, and I can continue to work or work in parallel on all the platforms at once while working on the similar projects. And several of you reached out to me and said,"Hey, how exactly are you doing this?" So I thought it would be helpful to share exactly how you can work across multiple different AI models on the same projects all at the same time without really doing anything in order to switch So what I'm going to do, I'm going to share my screen in order to walk you through the steps.

1:01GMT20260914-185903_Recording_avo_1280x720:Those of you who are just listening, don't worry about this. These are just PowerPoint presentations that help me keep, in the same order that I wanted to share the things. you don't have to be watching the screen. If you wanna watch the screen, there's a YouTube channel, and there's a link for that in the show notes. and also you can go and watch it on Spotify over there. their video is available as well. But that being said, let's get started, and let me share with you exactly how I'm doing this.

1:48GMT20260914-185903_Recording_avo_1280x720:So the goal, as I mentioned, is to have one workspace, basically one place where you can have your entire work, multiple projects. And when I say projects, I don't mean like ChatGPT or Claude projects inside the product itself, inside the large language model. I mean a project you're actually working on. This could be a specific client, a specific demo, a specific product, a specific anything that you're working on, could be a quote-unquote project.

2:14GMT20260914-185903_Recording_avo_1280x720:And I just build them each and every one into its own folder, and I can, like I said, work with multiple tools on the same thing at the same time. So what are we going to cover today? We're going to cover about why this matters, how does it help me, what are the different components that make it possible, how do I switch back and forth between all the different tools, and how to connect them properly, and exactly what you need to do step by step in order to set it up so you can do the same thing.

2:40GMT20260914-185903_Recording_avo_1280x720:So first of all, let's talk about why would you even want to do this? Why would you want to work in Claude, in ChatGPT, in Gemini, or whatever you want to work with, on the same project? And there are many, many different reasons. The first reason is that these different tools have different pros and cons. You may want to pick different tools to do different things. Some of them have features the others don't. Great examples, ChatGPT can generate images, Claude cannot generate images.

3:10GMT20260914-185903_Recording_avo_1280x720:Claude has a dedicated design tool called Claude Design, ChatGPT doesn't, and so on and so forth. But even on the things they both have, they each have pros and cons and things that one of them does better than the other. So this is number one. Number two, these tools are down every now and then, and when they're down, if you're just in one universe, you're completely stuck. There is nothing you can do. But if you can switch to the other tool, then you can c- you can continue working without more or less any hiccups on the other platform.

3:39GMT20260914-185903_Recording_avo_1280x720:Another is obviously token limit. I am on the max plan on Claude, but I still run out of tokens almost every single week because I run multiple things in parallel and automating and so on and so forth. But if I offset some of the work to ChatGPT, I can do more with Claude on the things that actually matter. And I also have a full history of everything. So regardless of where I worked and what processes I did, it's not divided into two universes. Like for most of you who are working with more than one tool, you have different things working in one tool than the other, and you can't switch because the other tool doesn't know what's going on.

4:18GMT20260914-185903_Recording_avo_1280x720:And if you need to switch, you don't have the option to do that, and I do, which is a huge, benefit. So there are, again, many reasons to do that Now the problem is there are several different problems. Problem number one is how these tools actually read files and what they read and how they work. So as an example, if you're running in Claude Cowork reads a file called claude.md that is in the folder you're working on as the instructions on how to work with you on that project.

4:51GMT20260914-185903_Recording_avo_1280x720:ChatGPT, on the other hand, reads a file called agents.md, and so does other different tools. So two separate things are actually reading in order to decide and know how to work with you on that particular project. Another issue is memory. Each of these tools, ChatGPT, Claude, Gemini, Groq, et cetera, have their own internal memory. They remember things. If you're working within projects, now I'm talking about projects inside the tools, each project has its own memory.

5:20GMT20260914-185903_Recording_avo_1280x720:These are not shared between the two different universes. So whatever you did in ChatGPT, Claude doesn't know, and vice versa. Now, there is-- if you are backing up the information somehow, then you need to coordinate how exactly the information is backed up, so it's not only living in one side or the other, and it's not living just on your computer. So if your computer dies tomorrow or gets stolen or your kids spill orange juice on it or whatever the case may be, then you can still continue working, and this backup will work, needs to work for both worlds or as many worlds as you wanna connect.

6:00GMT20260914-185903_Recording_avo_1280x720:and then the last problem is that skills that built on one side do not necessarily work well on the other side. So while skills are an open source model that is supposed to work seamlessly from one platform to the other, every platform adds small little things to it that do not work well on the other side. So now let's talk about what makes this thing possible and how does it actually work. Now, to do that, I need to give you a little bit of a background, and that is how my entire infrastructure work.

6:29GMT20260914-185903_Recording_avo_1280x720:So everything that I do in either ChatGPT Work or Claude Cowork or any other platform that can read and write files from your local hard drive. In every project that I work on, I have three different files. One of them is a claude.md, which actually explains what is the project and what are we working on and how to work on that particular project from a rules perspective, connections perspective, and so on. I have an evergreen document. That evergreen document explains the project itself.

6:59GMT20260914-185903_Recording_avo_1280x720:So not how we're going to work, but what are we working on, what is the goal, what I'm trying to achieve, who's the target audience, what did we agree on, what things it connects to, and so on and so forth, including every aspect and every description of that particular project. the third one is a tasks registry. That tasks registry is a CSV file that captures everything that we need to work on, everything we worked on, and what's the status of everything.

7:25GMT20260914-185903_Recording_avo_1280x720:So this is a very granular level of everything that I'm working on. And then there's a higher level Claude-wide lessons learned document that gets updated with everything that is applicable for more than one project. So how to connect to N8N, how to browse for different things, what not to repeat that were mistakes that we did before, how to package, presentations, how to copy things from files to different kinds of files, et cetera, et cetera. All of that is in the lessons learned file because it carries across multiple projects.

7:55GMT20260914-185903_Recording_avo_1280x720:Now, I'm not going to dive into all of those. If you wanna learn more on exactly my Claude universe and how I built all of those, you can go and check out episode 288 that was called My Entire Claude Setup for Creating AI Teams That Run My Business. So more on that if you wanna learn about that. But these four file-- files exist always in the three first files, so the claude.md, the evergreen, and the tasks exists in every project that I start. Now, this is really the baseline for the entire thing.

8:25GMT20260914-185903_Recording_avo_1280x720:So the four pieces that allow me to jump back and forth between the different platforms is exactly this thing. One is the problem of the claude.md versus the agents.md. So I told you in the beginning, ChatGPT reads agents.md, so does many other, tools like Hermes and others, but Claude reads claude.md. Most of my stuff is in Claude, so that's how I got started. So I have claude.md files more or less everywhere. By the way, it's not a necessity, but you can have it if you wanna add specific guidelines that you want Claude to read every single time they work with you on something specific The simple solution for that is to just add an agents.md file to each of those folders that have a claude.md, and basically put in just one line that says,"Before you do anything, go and read the instructions in claude.md and follow those instructions." So this is a one-time effort that you do for every project, and in my projects that get created with that already in it, claude.md is the actual source of truth and instructions.

9:32GMT20260914-185903_Recording_avo_1280x720:Agents.md just points to claude.md, and that solves that problem. So now if ChatGPT or Hermes or anything else goes into a folder, they automatically, because that's what they know how to do, they open agents.md. Agents.md tells them to read claude.md. They go and read that, and they follow the instructions seamlessly and perfectly fine. So this is number one. Number two is the actual project memory. So as I told you, ChatGPT has its own memory, Claude has its own memory, Groq has its own memory.

10:04GMT20260914-185903_Recording_avo_1280x720:But because now the memory lives in the files that are inside each and every one of these projects, so the evergreen file that carries every decision and the status and what are we working on and so on, and the task registry that has the granular step-by-step everything we did and everything we're doing, that stays there. I have a product requirements document, a PRD, for every more complex project that I'm working on. That is there as well. and all of these gets updated as the work goes on.

10:35GMT20260914-185903_Recording_avo_1280x720:How does it know to gets updated? Because the claude.md tells them to read the files every single time and to update the files with every new thing that happens. So I can switch back and forth, and the, quote-unquote, memory of one and the memory of the other doesn't matter because the real important memory is in the actual files they both read and update with every step of the way. When it comes to backing up and securing your information, and again, it resides on your local computer, so it needs to be backed up somewhere.

11:02GMT20260914-185903_Recording_avo_1280x720:This could be synchronized to OneDrive or SharePoint or GitHub or wherever you wanna back it up to, but it needs to be backed up somewhere so it doesn't live just on the computer. That lives in just one of the models because the backup just needs to happen. It doesn't matter. There is no ChatGPT backup versus a Claude backup versus a Groq backup. They're all the same backup. So the way it works in my world, because most of my universe is in Claude, Claude is in charge of that, and Claude backs up my entire data set for everything to GitHub at the end of each day.

11:35GMT20260914-185903_Recording_avo_1280x720:So in the worst-case scenario, I lose a few hours of work. If my computer dies or gets stolen, everything else can be backed up. Once I bring the backup back down to a new computer, then both or as many models can connect to it and continue working from that point. So ChatGPT can do the work during the day. At the end of the day, the GitHub backup gets triggered from Claude, and then the next time any of the models go in, they see the most updated information.

12:04GMT20260914-185903_Recording_avo_1280x720:By the way, doing this kind of upload and download of information can also allow you to work from multiple computers, not just from the same computer with multiple agents. So you can actually connect from anywhere in the world if you have access to where your backup is. Again, this could be OneDrive, GitHub, whatever. You can now download that information to whatever new computer and continue working seamlessly from that point with any model you want.

12:28GMT20260914-185903_Recording_avo_1280x720:So it enables additional flexibility. It also enables multiple people to work on the information at the same time, and I'm not gonna dive into how that works and how GitHub has branches and the way to merge the branches safely, but it does. I'm just putting that out there. If you wanna figure it out, you can easily figure it out. And by the way, even if you don't, Claude and ChatGPT know how to figure it out. So even if you don't wanna dive into the details of how that works, you can ask your model to help you set it up, and it will.

12:54GMT20260914-185903_Recording_avo_1280x720:And from that moment on, multiple people can work on the same content and really complex projects with multiple tools at the same time Now, the cool thing is beyond the fact that now I can do stuff in Claude and ChatGPT and Grok and whatever, I can also write code at the same time. Why? Because Claude Code and ChatGPT Codex are actually looking at the same folders, which means I can do the research with ChatGPT, the design with Claude, and then write code simultaneously for different components of the project with Codex and Claude Code all in parallel because they're all reading and writing in the same folders, understanding the rules of the game, and referencing the same documents and updating the same documents, so everybody's on sync.

13:45GMT20260914-185903_Recording_avo_1280x720:So in my world, as an example, I have in many cases three or four windows opened at the same time working at the same project on multiple components of it. I will have several different Claude Code windows working on some stuff. I may have a Codex window working on something else in the same project, and I will have regular chats to continue developing and getting feedback and designing the next steps and so on, either in ChatGPT or in Groq or in Claude or in whatever tool that I want So what do you need to do in order to set it up?

14:18GMT20260914-185903_Recording_avo_1280x720:The first thing you need is this kind of structure from a documents perspective. As I mentioned, you can call your documents whatever you want to call them, but you need some set of documents that will be the memory and the ongoing update platform for everything that you're doing, for anything you're working on. In my case, it's in Claude, and I told you the documents that I'm using, and then claude.md is the one that actually governs the way I work.

14:44GMT20260914-185903_Recording_avo_1280x720:Then you need to create an agents.md that just points to the claude.md file. So by the way, if you're working the other way around, if most of your uni-universe is in ChatGPT, you can reverse the You can keep agents.md with your rules of engagement and just use claude.md to point into the agents.md. It doesn't matter. It works exactly the same way. You also want to add a one paragraph to ChatGPT's custom instructions and the same paragraph to Claude global instructions that will tell them exactly how to work with all these files, when to update, when to read, and so on.

15:19GMT20260914-185903_Recording_avo_1280x720:once you do that, then both of these platforms, and again, the same thing exists in any other platform you pick, will understand how the overall system works and will be able to work with it systemly. and will be able to work with it seamlessly. Now You need to make sure that one platform and one platform only is in charge of the backup. So if you are using whatever backup platform, make sure you're not doing it twice. That's just gonna create confusion and just unnecessary backing up.

15:51GMT20260914-185903_Recording_avo_1280x720:So you need just one backup, and then once you set it up once, you can test it and see it run and see how it works. Test it a few times, make sure it is working properly, and that's it. You're done Now, what I wanna show you next is a proof on how exactly that works. So I worked on a project that organizes all my travel and creates a mini app for everywhere I'm traveling. It's showing me the status and the flights and what I still need to book and all these kind of things.

16:16GMT20260914-185903_Recording_avo_1280x720:It's really cool, and it helps me on the day-to-day. I built all of it inside of Claude, and it's been working inside of Claude, but there were some loose ends I wanted to finish. So all I need to do is open ChatGPT, and in ChatGPT, there's a little bit of a difference compared to Claude when it comes to connecting folders. if you wanna connect a folder to ChatGPT Work, you first have to create a project on the left side, like an actual ChatGPT project, and then connect the folder to that, versus if you are connecting a folder inside of Claude, you can just connect the folder to any chat.

16:48GMT20260914-185903_Recording_avo_1280x720:You don't have to build a project first. But what I did here is I connected the folder to the travel planner project that has no other information other than the fact it's connected to this folder, and I asked it,"What is the current status of this project, and what are the next steps?" Again, a very bad prompt from a prompting perspective because it gives it no context. And yet, because it is connected properly and it has the agents.md that tells it to read the claude.md, and then it has the Evergreen and the tasks registry, it knows everything it needs to know.

17:18GMT20260914-185903_Recording_avo_1280x720:So it tells me,"The travel organizer functionality deployed as an AppSheet app with blah, blah, blah, blah, blah." It tells me everything that it does. It says,"The latest document indication was March 9th of 2026, so the system live health, should be reached before relying on it today." However, it says,"Current records show 39 tasks completed, six ready, one pending, one on hold, and five no longer applicable. The local React app is only partial prototype and is not deployed product." So it tells me what's the exact status down to the task level, even though it's a brand-new project inside of ChatGPT that I've done nothing for, and yet it knows everything about it, and it says,"The next step in priority of order are," and it gives me six different things that I have to do, and it gives me the reference of the exact, sheet with the different tasks as it found in the folder.

18:12GMT20260914-185903_Recording_avo_1280x720:So this is showing you that I can develop something in Claude or in any other platform, and with a few clicks, because I have all the infrastructure set up, continue and keep going in ChatGPT or Hermes or Groq or whatever other tool I decide to connect to it as long as I set it up to know how to work with this set of files that I've created as an infrastructure. So a quick final summary for this. This capability enables you to be highly flexible with how you work.

18:44GMT20260914-185903_Recording_avo_1280x720:It allows you to overcome when one of the platforms is down. It allows you to enjoy the best capabilities of each of the models seamlessly switching back and forth. It allows you to maximize your tokens beyond what you could if you did not use this kind of process. And overall, it just makes it fun because I get to learn and see how Claude and ChatGPT do things, and in many cases, I will do the same thing in both sides just to see which one comes out with a better outcome, which allows me to understand how the latest model works in specific things, all of that without building new infrastructure once I set it up once.

19:19GMT20260914-185903_Recording_avo_1280x720:I hope you found this valuable. If you have, please drop me a line on LinkedIn. And if you wanna learn more how to take this to a completely different step, don't forget to sign up for a multi-agent orchestration course that is starting in November. It is a course that have been taken by hundreds of people already this year, and it's literally transforming people's lives and businesses. So if you wanna learn how to take this kind of infrastructure to a completely different level, don't hesitate.

19:45GMT20260914-185903_Recording_avo_1280x720:Come and join the course. There's a link in the show notes. It will take you two minutes to sign up, and I would love to see you there. By the way, if you're enjoying this podcast, please share it with other people and like it and rate it on your favorite podcasting platform because that will help other people find it. It will help your friends be aware of it and learn about how to implement AI effectively as well. That's it for today. I'll be back on Saturday with another news episode.

20:13GMT20260914-185903_Recording_avo_1280x720:Have an amazing rest of your week.