Introduction to Vibe Coding

Vibe coding means building software by describing what you want in plain language—no programming required. Janette Roush demos the full range live: a Minor League Baseball road trip planner, a Global Ambassador application form wired to a Google Apps Script, an RFP tracker, a strategy recap site built from Plaud meeting recordings, and fam-trip itinerary websites. See how Lovable.dev gets you started and how Claude Code and Claude Cowork take you further, plus how to share sites securely with GitHub, Bitbucket, and Vercel.

47 min
27 chapters
JR
Janette Roush
Chief AI Officer, Brand USA

Chapters

Key Takeaways

  • 1Vibe coding means describing what you want in plain language—typed or spoken—to get a usable app or website, without knowing a programming language. Andrej Karpathy coined the term about a year ago.
  • 2Lovable.dev is the friendliest on-ramp: a freemium platform where you build entirely through plain-language prompts. Janette started there and moved her work into Claude Code and Claude Cowork, which let you give the AI more granular direction and edit the underlying code directly.
  • 3The session walks through real DMO use cases, all built with AI: a Global Ambassador application form (a branded front end on a Google Form backend, with a Google Apps Script that auto-emails each application to the right market), an RFP project tracker, a strategy recap site built from Plaud meeting recordings, fam-trip itinerary sites, and interactive keynote recap websites.
  • 4Free tools come with a tradeoff: keep confidential information out of them, because when a tool is free, you are the product. To share sites externally and track versions, host the code on GitHub or Bitbucket and deploy through Vercel with password protection—a step where your technology team should be involved.
  • 5The future of software is bespoke tools built for one person or one team that never need to scale. These tools come together by iterating through back-and-forth with the AI—letting it cook while you work across many chats at once.

What You'll Learn

After watching this video, you will be able to:

  • Define vibe coding and explain how plain-language prompting replaces knowing a programming language.
  • Build a first website or app in Lovable.dev using a free account, and recognize when to graduate to Claude Code or Claude Cowork for more control.
  • Point Claude Cowork at a local folder of brand guidelines and source files so it has the context to build on-brand tools.
  • Build practical DMO tools end to end: an RFP tracker, a strategy recap site, a fam-trip itinerary, and an application intake form.
  • Turn Plaud meeting recordings and keynote transcripts into living websites.
  • Share vibe-coded sites externally and safely using GitHub or Bitbucket plus Vercel with password protection.
  • Manage AI cost and rate limits by choosing the right model (Haiku, Sonnet, or Opus) for the task.

Full Transcript: AI for Tourism Professionals

Thanks everybody for joining today. we are here for the May Agents of Change Webinar Series from Brand USA, and today will be a lot of live demoing, so let's cross our fingers that I don't hit any session limits and that everything goes smoothly. My name is Janette Roush. I am the SVP of Innovation and the Chief AI Officer for Brand USA

I want to introduce us to the idea of vibe coding, which is a term that Andrej Karpathy coined about a year ago now, saying that we no longer need to code by knowing or understanding a coding language.

We can just now code with vibes or really with text, by either talking or typing to our computers and being able to describe very clearly what we want in order to get a usable output. What kinds of things can we actually do with vibe coding? This is an example of something that I have vibe coded, and it's my own personal website.

If you are looking for a starting place, maybe for something outside of work, just to understand better how does AI work and how do these vibe coding platforms, how can you use them? The QR code takes you to this website. You can find it yourself at janetteroush.com. And there's a lot of assets in this website.

I use this as my personal playground to understand what works and doesn't work when you're using AI. but primarily, I use it to host all of the videos, and all the video content that I create as part of the Agents of Change series. So this is a, a learning system like this, you could expect to pay twenty thousand dollars a year, right?

If you have something that's a truly integrated learning system. This lacks some of the bells and whistles of those types of systems, but for our use, it's free. And so to be able to create and host this, other than the hosting fees for janetteroush.com, everything else that's part of this website was done absolutely free of charge.

If you're playing around with it on your second screen right now, you'll see not only does the homepage have links to all of the videos, when you click through, it includes a full transcript of the video. It includes a description of all the resources that I talk about in the video. That's not something I even told the AI to do when I was building these web pages.

I gave it the video link that is hosted, with a free service called Mux. It looked at the videos and read the transcripts and came up with a number of other things for these landing pages that it thought, "Oh, this is what other learning websites have on their landing pages," and really made it this great, robust experience.

Part of the fun of the vibe coding process is that you don't have to be the only one coming up with ideas. AI will help you come up with those ideas.

The next example, this is another website that I vibe coded, and it is a Minor League Baseball road trip planner. I made this for personal use, so feel free to scan the QR code and play with it yourself. The idea behind this was my husband wanted an easy way to plan a road trip, with his oldest son this summer.

They wanted to either pick a destination and start a road trip from the destination or look at a number of weekends and see which weekends and which starting place would give you access to the most number of home games played by a Minor League team. And that's something that doesn't really easily exist today, because there are a number of home games happening at any given time. There's not really one place you can go to for all of this information. I had Claude go out and find schedules for every Minor League team and put it into a database that it set up by itself using a system called Supabase.

And it then created this overlay, this map, to show, all right, these are where we know of every stadium in North America. And it allows you, as you can see on the left, you either know where you want to start or you know the dates that you want to travel. And this will allow you to map out a plan that will pack in as many games as possible.

Again, I don't know anything about coding. I just did a good job of describing what we were looking for, and Claude made this as the output

And then finally, for a work-based example, we'll walk through actually building a replica of this shortly. This is a strategy website for Brand USA. So I-- many DMOs or many other organizations I know when it comes time for strategic planning, a lot of that is going to happen in a PowerPoint deck because I think many of us live our lives inside of PowerPoint decks.

But PowerPoint isn't always the best way to articulate and then track your strategy, or your plan in this case for, fiscal twenty-seven. And so I took a recording device called Plaud, P-L-A-U-D, and recorded meetings that we had with our agencies where we were discussing our strategic plan for the next year and a half.

I could then upload that transcript into Plaud and say: "Would you look at this and help us organize it into what our strategy will look like for the next year and a half?" And it created this, this website that we have password-protected that allows you to scroll through pages to understand what is each pillar of our strategy, what are the ideas and tactics behind it, have a running list of what do we need to do and what has already been achieved.

And it takes the strategy from just being a flat file that nobody looks at again into a living document that we can use for every meeting moving forward. So we'll walk through that as a use case shortly. And then finally, I'll have an upcoming session on this, and so I'll have a registration link for that at, the end of this webinar.

But when we get to the June webinar, we'll be talking about how you can use Claude Code as your personal operating system. And so this is what that could look like So let's get started with the vibe coding piece here. We are going to start with a tool called Lovable.dev. And so now I am coming over for the live portion of this.

So you can find it, obviously, just Google Lovable. You'll see when you do that, that there are a number of, competitors to Lovable. Replit is a popular competitor.

And once you log in, this is a freemium platform. So as a new user you can have a certain number of free prompts every day. But if you want to be able to use it repeatedly, you'll want to have a paid account. The few things that I have done in Lovable, I have entirely done using a free account, and I will show you an example of one of those, which is a DC visitor readiness website.

And I have this open, so you can just see how does this work. This is something clearly I created last summer, and I just said, "Build a visitor readiness scorecard for Washington, DC." clearly this was a prompt I got from somewhere else because I didn't know personally that these things existed.

But I'm like, great, this is the name of an API for pulling in live weather. We know that the transit system in DC has their own API that's publicly accessible, and you can also find wait times for the Smithsonian Museums, and there's a service called Twilio that will have a text to me feature. Lovable thought about this a little bit, and this is very much a standard chatbot interface, and it's great, I'll make a beautiful visitor readiness scorecard with live data integration.

We'll give it a government civic feel. So it built everything out, and then it gave me some options for how I could refine and customize this website. So I added in a couple of other things. I'm like, "Great, why don't we get the status of, cherry blossoms?" Which, oh, we missed it, guys. That was last month.

but we could also feature TSA, wait times and hotel rates near the National Mall. And here we see the hotel rates have real-time simulated data. So if I were going to continue to iterate on this, I would want to find an actual live feed, of hotel rates so that Lovable could not just make up information to put in this particular dashboard.

But now we have the opportunity to build our own app inside of Lovable. And so let me find... I have a sample of what I wanted to make, which is a national parks trip planner

So this is the prompt that, actually Claude wrote for me. And so that's a helpful way to approach how are you going to put this together, right? You don't have to have every idea yourself. You can ask AI to give you ideas. But I will say because there are only a limited number of prompts that you get to use with Lovable, unless you're a paid subscriber.

I tend to do myideation inside of another tool, and then I come over to Lovable, and I use one of my five free prompts a day or whatever the limit is right now to have Lovable do the generation. And here you see that Claude knows some ideas for APIs that can bring live information into this website, without needing an account

And so now we're going to start building, and this is going to be the boring part and why it's scary to do this live on a webinar, right? Because now we have to sit and wait for Lovable to figure out what it is actually going to build for us. Uh-oh, and now it wants us to have, this actual API key, which I don't have.

I might save this for a later time. But again, like don't-- if something needs an API key, that's not necessarily a blocker here. We just need to jump through the hoops in order to get this API key. And if you don't know how to do it, you can ask AI what do you need to do?

I'm going to come back over and just say, use dummy data. Uh, so Billy Kolber is asking, "Why are we building this dashboard in Lovable instead of Claude Code or, ChatGPT's Codex?" We're actually going to move on in the next examples we are going to do in Claude Code. Uh, I'm just showing this really as a more user-friendly way of doing this work.

Uh, oh, here we go. We're making good mock data. So there's multiple ways to build and design these websites. This is just one example of one service that does it. I got started with vibe coding using Lovable because it is so user-friendly, and I very quickly graduated from Lovable and moved all of my work into Claude.

Many developers prefer Codex over Claude or a tool like Lovable. So this is just to illustrate that there are many different tools that do this work, and if you don't have a subscription to Claude, this is a great one to try out. I would not put confidential information into a tool that you are using for free, because remember, when it's free, you are the product.

This is going to be thinking for a few minutes. I'm going to take that opportunity to move back over here and show you a few other things that I have built, using these tools. So one example is this application process for the Global Ambassador Program that Brand USA is currently recruiting for.

This is a global program where members of the travel industry who have some kind of online following can become one of our two hundred and fifty global ambassadors in celebration of America250. Originally, the trade team wanted to make this a Google Form, but the thing about Google Forms is all Google Forms look alike, and you can customize the inputs, but you can't really customize what the experience is of filling out that form. I was able to work with Claude Code and tell it all of the things that we were looking for. And it built this front end for a form.

And then on the back end of this form, it's actually, it's still just a Google form. When you fill it out, it is populating a spreadsheet like this. I'm showing this because this is test and dummy data in our offices testing out how it worked.

You will also see that the code engine built out on its own a separate RSVP page here, or tab inside of the Google sheet to show who were the applicants from each of these markets. The part that I like best about this is that it wrote code for something called an Apps Script.

To access it, you go to Extensions and Apps Script. And if you were like me, you didn't even know this existed, much less feel comfortable writing an Apps Script for this project. But what it has done is written an Apps Script, if we go through here, that when it collects the input from the applicant, it not only adds it to this Google sheet, but it emails the application to the correct person in the correct market so that they have a copy of the application.

They can do a double check by coming here and clicking on the tab for their market and making sure that nobody fell through the cracks, but they're going to get an email with every single application that comes in. And that's something that without AI, there would be somebody, at home office, at Brand USA who is receiving every single one of the applications and then forwarding it to the correct person.

This removes that game of telephone. And now we're going to come over and see what Lovable has made for us. It's added dummy data to these pages.

You can see that it's a little imperfect, right? Like this image at the top is now overlapping some of the real information that we want. But we get a description of the park. We get information about active alerts and closures, the entrance fee, featured activities. Again, it reminds us this is not live information.

This is all dummy information, so that we don't forget and then distribute this website. Now we could come down here, and I could say-- And I'm actually going to take my own advice, and I'm bad about remembering to do this, but anytime you can talk to AI instead of typing, it's going to be better because we speak faster than we type.

So we're going to give more information to the tool when we remember to speak to it. I am using a tool called Monologue to do this. And I want you to look carefully at the header image because it is overlapping and cutting off part of the first text box that has the name of the park and the weather in it

And here you see it pasted in my text, and we're going to ask it to build that

And so now while it is thinking about that, we're going to look at something else that I built using Claude Code, and this is-- I haven't finished it yet, but this is an interactive game to help the Brand USA staff learn how to use Claude CoWork, with a copy of The Oregon Trail for all of my Gen X friends who remember The Oregon Trail.

So here we see, yes, I'm a Brand USA employee. I've been doing things manually like it's 1849, and a rumor reaches you that there is a better land out there, and it is called CoWork.

It has little inside jokes about AI.

And then it takes you to just like you were playing The Oregon Trail where you have to actually answer a question. I need to create a presentation for the Travel Week Europe team. Do I want to start from a blank PowerPoint? No. Do I want to email the design team? I may end up doing that actually, but let's see what Cowork can do for us because we have PowerPoint presentations built as a skill inside of PowerPoint-- inside of Claude Code to help us with this.

Uh, and the build for this, I'm going to answer Cyrus live, is that I just went to Claude and said, "Would you build an interactive app for me that is a a takeoff of The Oregon Trail that would teach my team how to use Claude Cowork?" Like, that was it. So building these things does not have to be very hard.

This was pretty stupidly simple. What I would want to iterate on and why I haven't distributed it to the staff yet is that there's not that many options in it, and so you can play the game twice, and then it gets a little samey. That's something that's on my to-do list. I will also say the reason that it knows about Brand USA examples to put in here is because I have set up my own internal memory inside of Claude, where I have a ton of files saved on my computer, that are markdown files that are part of my personal operating system.

And because Claude is pointed at that folder, it knows that I have things like the Travel Week PowerPoint presentation skill. It already has all of our brand guidelines. It can see the things that we have that would benefit our staff to know about. I didn't have to come up with those ideas.

Claude came up with the ideas of how to populate this game for me. So now let's come back over to Lovable, and it actually made it worse. So that is... What it probably needed to do was decrease hero padding and not increase the hero padding. So that is, that is pretty funny. I'm just going to go back one more time and say

You actually made it worse, and I think by increasing the padding, you increased how much of the top s- screen you are overlapping with the image. Maybe you need to try decreasing the padding or increasing the white space between the end of the image and the start of the text on the screen

And we'll give it that, and we'll see how it goes. I will say this is part of what I don't like about Lovable and why I don't continue to build with it. because it is really the only way that you can change what it's doing is through plain language. And Claude, I find in the setup that I have, you're able to get more granular, and you can actually give it code in order to update its code, which I find helpful.

We are now going to come over to Claude. So if you want to be building in Claude, you're going to want to have the desktop app downloaded on your computer. Here you see we have our traditional chat, but we also have Cowork and Code at the top of the Claude screen.

So Cowork is really just a nicer consumer interface for Claude Code. For these demos, I'm going to be working with Claude Cowork because typically when I'm using these tools, I'm inside of VS Code, and I am using Claude Code in a coding terminal. So here in the desktop app, what I've done is create a fake DMO that we call Visit Beige County. And you want to get started in Claude Cowork by just if you go to your computer, you can click here, and you can pick any file in order to have Claude write to files on your computer.

So I have created, and you see here personal operating system. This is where I do most of my work, so everything is organized inside of that system. For Visit Beige County, I have made now a new operating system.

Screenshots are a great way to troubleshoot. We are now inside of Visit Beige County, and I want to make a project tracker for an RFP that Visit Beige County has out. So if you were at last month's webinar, you might remember that Skyler Clark was with me, and we talked about using Claude and ChatGPT to manage an RFP process.

But let's say we want to create a visual tracker.

This is clearly, it is a prompt that I had AI write for me. I find that to be an easier way to start the ball rolling when you want to create an app or a website.

You'll see here when you are inside of CoWork, again, these things are a little slow, so it's going to take some time to prepare this session. You are also able to see, like here, Visit Beige County. This is a folder on my computer. So we can see the files in this folder by clicking on that folder button.

And here we have Visit Beige County. These are the RFP files that I actually created for last month's webinar. I am doing a separate training on how people who are attending IPW can use AI tools for their follow-up by recording what they're doing, so we have some fake materials for that project. We have Visit Beige County's brand guidelines and a fake fam trip itinerary. So those are all of the files, all of the context that this tool has access to. And another thing that I like about this tool is that you can work in multiple chats at the same time.

So here I have already started a conversation with it earlier today about just like this, the webpage for the Brand USA strategy I was showing you. I want to make a webpage showing Visit Beige County's strategy. So it's "Okay, great." So it's a fictional DMO. We need to understand the purpose.

I'm asking it if you can see the files in my folder. It's telling me, "Oh, I don't see your brand guidelines." So I renamed one of the files And it's "Oh, okay, looks like the file got renamed. Let me peek inside. Okay, good. We have a demo-ready design system like Muddy Fork Brown and River Trail Sage as Visit Beige County's brand colors."

For the webpage, since we have the brand guidelines in hand, we want to understand the audience and the scope. So this is... And once again, I forget to talk to it. The audience will be the staff of Visit Beige County and their agencies. So it's not public facing. It is a working document to help the stakeholders who are in the middle of this strategy process keep track of their conversations and understand what decisions they've made and what decisions they still need to make.

For the scope, I would like a full recap site that has one page for each of the strategy blocks. I'd also like a list of next steps from this meeting and a list of action items.

And we're hitting go on that, and now that's going to cook for a little while. So it's, over here it's going to think about what it's doing. We can go back to the first task

And we have the opportunity to see what they built. So now we're back to the RFP portion of the demo. We see that this is an active process, that scoring is in progress.

We see that there is a six-month evaluation period here. I am working in Claude Cowork, and when it finished and it made this file for me to look at, I was able to click this, and it opened up the file in Google Chrome for me. This is all based on materials that I've already written and that Claude has access to.

If I come back over to this folder, this is what Claude is reading in order to do this work.

So we see here, this is the document that Claude Cowork is reading to create this website. You see I wrote this back in March, so this was for an earlier webinar, and it put out this RFP. These are the things that visit Beige County needs as part of the new agency that they would like to hire

I had AI write every bit of this, and I would not be able to do these demos with fake information if AI couldn't write all the materials for me. It's been a huge help. That's what Claude Cowork is looking at. It created fake names for the members of the Visit Beige County board who are part of this process or their staff.

We see the finalists for the actual RFP, and it's tracking the evaluation criteria and the decision log. Now, I'll say, looking at the names of the finalists, it actually didn't look at the real names in the Drive, so I need to be a little more specific with it. I see, let me go back to talking, that you made this using dummy data.

But if you look at the folder that is titled RFP, all of the actual proposal responses are saved in that folder. Can you integrate the real information here?

Now it's going to take some time to cook on this particular project, and that gives us a chance to come back to the strategy webpage And here

It's running commands, and it says it's still working on it. AI is terrific for people with a little bit of ADD because it allows you to jump around all of the time rather than if we were sitting here waiting on this one project, to finish, the webinar would get boring pretty quickly, right?

So this is what my day looks like now, is that I'm jumping back and forth between 12 different chats at any given time, moving projects forward and then sitting back and letting the AI cook while it determines what the next step is that it needs input from me about So that is cooking

This is cooking. Let's come back over. Oh, it looks like Lovable finally figured out our space in here So this is just a sign that when something isn't exactly what you want initially, it's going to take some back and forth before it understands what you are looking for and is able to reproduce it.

And that back and forth is part of the process. So it's completely okay. You don't need to get impatient with the tool

So now we're still thinking here

Let's see And we're still thinking there. I'll have one more demo to show you. A big way that I have been using these little micro sites recently is that when I do a speaking engagement, so this is from, the Wyoming Governor's Conference, I also did one recently for Maine.

I record myself when I am doing my talk, and I can upload the transcript of my speech along with a copy of the actual presentation that I gave to Claude. And I will say, honestly, nothing fancier than just here is, a speech that I want you to turn into an interactive website. And here you could see that as we go through this information from Maine, like it really did make it a little bit interactive.

So we just saw that as we scrolled down, it was ticking up to the sixty-two percent

We have some quotes from the Q&A in the main website. It's describing some of the ideas that I shared. And you can see here, I wonder if these are actually links. They're not. So it makes it look like they're clickable links. That's something I could go back and say, "Would you link out, in these places where you have this little bit of interactivity?"

It took the questions that were asked in the Q&A, and it organized them by topic. It really delivered on the brief of how do you make an interactive website out of, a 60-minute keynote

Billy is asking what kind of background files and skills needed to be in my operating system before that prompt worked. I have it built out as a skill now, but when I was first starting, I just said, "I want you to turn this into a website," and the only files that it had access to that would be relevant were the brand guidelines for Brand USA.

And I will show you what the brand guidelines look like for Visit Beige County. Brand guidelines, and you'll see I had AI write this, but it Google came out with a schema for how they say brands should list their brand guidelines in order to work best with AI. Because what a number of these organizations want to do moving forward is make this easier for all of us by just having one schema, one set of how does this work, so that every AI system knows this is what you look for when you are looking for this information.

And so when I tell it, "Use Brand USA's brand guidelines," or in the case of the project we're working on now, the Visit Beige County's, brand guidelines, this is what it means. Billy is asking, "The skill is just the accumulation of knowledge and guidance from my back and forth to create this website?"

And the answer is yes, ex-except that at the end of the project, I say, "This is perfect. Would you turn this into a skill for me?" And then the AI will take all of this information, everything that it learned from doing the back and forth about what I like and don't like, and it used that information to create in Claude what is called a skill, which is...

It's essentially you could think of it as an agent. it's a prompt that has access to tools, and so those tools are what allows it to build a website So let's go back over to Claude and see. All right, so we are still working on the strategy meeting

And we now have a real tracker. Okay, so now we have the engagement budget. That is correct

I didn't offer a review committee ever, so that would be made up. And these are the real... The- these are the actual files in the system. So it's gone through and seen the actual files And I think it may have actually evaluated the files because you can see here they're named like AI when it created these fake proposals, it gave them a specific name, to be funny.

So Redline & Associates was meant to be the worst proposal, and you see here that it had the lowest score and was disqualified from proceeding with the RFP

And it's still like the decision log, it made this up, it's placeholder. But, it's crossed off what steps we've done and what our next steps are. And so thank you. That's a pretty good way forward for the Visit Beige County, RFP process. So let's come back over and see

We are still digesting what to do over here.

And you can see here, okay, the fam itinerary website. This is the third thing that I wanted to build. And again, this is an image that for the last webinar that I did, Gemini, Google Gemini created this image for me. So let's talk about fams, right? I'm going to make a new task still in this Visit Beige County page.

And right now I know a lot of folks when they are preparing for a fam trip, we're making a PDF file that has the full itinerary on it, right? that's hard to read on a phone. it makes it more difficult to make updates and changes to an itinerary in real time. So let's see if we can take a static itinerary and turn it into an actual interactive website for the people attending this fam.

I would like to make a interactive website that, shares the itinerary for the upcoming familiarization trip to Visit Beige County. there is an itinerary file saved in your drive. Could you find that and turn it into a website using the Visit Beige County brand guidelines?

It's going to think through that process and come up with a fam trip website. Now, the next step here after these websites get created is that you're going to need a way to share those websites externally. So websites like this one, you can see here because of the way the URL looks, the server is pointed at a folder that is saved on my computer.

But that's not going to be the easiest way to actually share this kind of information externally. And so what you would want to do for the next step here, and this is something where you're going to need to engage with your technology team, is that all of the information, all of this code needs to be hosted online.

And so a popular service for that you may have heard of is called GitHub. GitHub is actually already integrated into Claude Code. So once you have an account set up, it can automatically push your code to be hosted on the server. And then the next step is that there needs to be a hosting service to point that website at.

the one that Claude seems to recommend the most that I've been using is called Vercel, and I think it's V-E-R-C-E-L.app. we also at Brand USA have a paid account with Vercel so that we can make these ty-- these mini websites, and then we can put password protection on them so that they can't be accessed outside of the people who need to see it.

So there are secure ways to set that up. There's also completely free ways to set all of this up if you are just exploring using personal information for your own use, and I think that is a great way to get started. One of the first things that I built, was a fitness tracker because I was tired of logging in on my phone to a Google Sheet and then tracking what I was doing in Google.

And now I just have a cute app on my phone that is only for me, and nobody else has access to it, and it has everything exactly the way I like it. And that's the future of software, is that we are going to be able to easily create customized tools that are only for the use of one person or one team, and they don't have to be scalable.

We don't have to think about is this the SaaS solution for something. It can be very person to person in terms of a solution. Let me come back over. Oh, okay. So-- Oh, nope, that was the agency evaluation. What's going on with the website? All right, so it is still, building here. So I am going to take a moment while these tools are still building, to look at the Q&A quickly.

Uh, oh, great. I actually already answered this question. Cyrus has also asked what are my thoughts on token cost with Claude? what's the best way to scale without running into a paywall? And in terms-- that's, that can be difficult, right? Because Claude has limits on how much you can use it before it tells you need to take a break.

and you can see what those limits are, and that's probably a good thing to check on. If I come down to my settings and usage, you could see here I have used twenty-four percent of the tokens that I am allotted for the time period that we are in right now. This is something that Claude doubled their usage limits last week, so ideally it's easier to do this now than it used to be, in terms of coding without hitting limits.

But if you find that you are working in Claude and you are hitting these limits a lot, you have the option of upping your subscription, so going from the Pro plan, which is twenty dollars a month, to the Max plan, which is a hundred dollars a month. It starts there, and I think it scales up to two hundred.

You could change to API use, which allows you to continue to build, but you are paying at a per token cost. You can get really good at watching when does your current session say, "You know what? you're being too thirsty with the tokens, and we're going to rate limit you," by using a slower model.

So if we come back to the tasks we've been working on, down here that it says Opus 4.7. This is the most compute-heavy model. And honestly, when we are doing thought work, we want to use the smartest model, but you're going to run through your tokens more quickly. So if you were just making a quickie update to this, maybe you wouldn't need the Opus 4.7 model.

And you have the option, in Claude, it's all named after poetry, so the most simple model is named Haiku, the intermediate model is Sonnet, and then Opus is the most advanced model. So you can try out Sonnet and see if it does the work you need it to do at a similar quality because that's going to allow you to continue working with the tool longer.

I'm going to go back and see if we're still, we're still ideating on everything. So these are links I might have to share with everybody when the webinar, in the follow-up materials from the webinar, if they're not done in the next five minutes. I have some other questions here. Jose, as you are working a number of projects at the same time with different versions, how do you handle the final versions?

so what I find is that there is, there-- I'm not saving on my computer anywhere the iterations. So if I were uploading something to GitHub, the way that GitHub works is that it will... You can roll back to earlier versions, and that's the way we should probably think about coding projects. So as it's building this HTML file, it's just upgrading the same file.

It is not saving ten different HTML files where, you know how we're now accustomed to in knowledge work where it's like final, final one, use this one as your file name. It's just going to be one file for the most part. But if I were to connect one of these websites to GitHub, or the tool we actually use is called Bitbucket, it tracks your changes over time, and it actually, it tracks them by collaborator because these tools were built for writing code, which is what we're doing.

And so you are able to roll back one of your commits, and commit is just the coder term for saving your work. So you can go back to any point in time, and you will see a description of this is what changed when this person committed this update. And you can see the person who did it and what files were changed as part of that.

So that allows you to track if something breaks on your website, you can roll back to an earlier version of that website where the functionality wasn't broken. But it is like keeping track of this is a larger conversation we have been having at Brand USA that I will tell you we haven't quite figured out yet.

if we start making itineraries for, fam trips, at what point, who is in charge of going to a web-- going into Vercel and/or going into Bitbucket and saying, "We have 300 fam trip itinerary websites saved in here. Which ones can we get rid of?" And that is something that, Kat, who's actually watching this webinar with me from the technology team, so she's, she's hidden off camera right now.

this is something that she has brought up that I do not have a great answer for yet. So if you have a good answer, please let me know because that would be very helpful. Mark is asking, "It seems like Claude is my preferred platform, and why is that?" And that is because it has just made en- enormous bounds in what it is able to do since about November of last year.

So it has what you would call, both, I think, a stronger language model, but Claude code as what they call a harness for the language model, just meaning the instructions that it uses when it's working. It's just better. It's incredible. oh, now we can view my fam trip website. I hope it's good. Oh, so this is, I'm going to guess, yes, it has responsive design, so if one of the media attending this tour, needs to resize it on their website or on their phone, they can do that.

We have arrivals and departures. cute itinerary that you can click out to learn more contacts for when you are on site I, I didn't even ask it for a packing checklist, right? We have information about Visit Beige County down here, what's included, what's not included. I am very happy with that output.

I want to thank everybody so much, for joining the webinar. you can go to thebrandusa.com, /events, and that will have, registration links for upcoming webinars.