AI for Hotels: MIT Lessons, Claude, Vibe Coding and Real Use Cases - Sloan Dean & Josiah Mackenzie

Josiah: Last time we recorded Sloan, you were in the early days of thinking about the launch of Not Done. It has been an incredible success since then. Every episode has CEOs of some of the world's largest hospitality companies, incredible conversations behind the scenes, what's going on, what's really going on, and I very much enjoy every episode that comes up. Just listened this morning to your most recent one. I wonder as a starting point for this conversation, if we can catch our listeners up a little bit if they haven't had a chance to follow your journey. Last time we recorded, you were going to study AI Strategy and Leadership with MIT, and I believe you have wrapped up that study.

Sloan: I have.

Josiah: What stood out to you from that experience? It sounds like an incredible opportunity.

Sloan: Just the speed at which things are changing, and I don't think most people really understand how dramatic AI is and how it's going to change everything and it's already changing. Just the speed at which things are changing. Hotels have historically been pretty insulated, but this is the one thing that they're not going to be insulated from. It may come slower to hotels than other industries because we're more reliant on physical places and physical work. But everything that I learned, you fast forward to five and for sure 10 years from now, you are going to have a dramatic change in how hotels are operated and how consumers experience them.

Josiah: I think we're getting into a number of elements of that in this conversation, but just to stay on this for a moment, I am very curious about learning about new things. I know you're such a lifelong learner. You learn so many different ways. What was it about the MIT experience that stood out to you as especially productive? Was there any approach they took you feel like you took away a lot from, was really helpful to learn about this?

Sloan: They definitely scare you because the course I took is designed for CEOs and executives that are not like the CTO, and so there's a whole couple of weeks that you spend on security and the implications, and you have to have a closed box system, particularly around large enterprises. So just thinking about it, it can be beneficial, but it can be incredibly dangerous. That was really eye-opening. They had some security experts come in and it makes you worry about having all your personal network digitized and just the hacking and the capabilities that are happening. So security is something we have to really think about.

Sloan: I tend to be a glass half full kind of person and I think pausing and thinking about the implications. They even get into some of the psychology about what does it mean for a blended workforce. If all of a sudden you're relying on AI so much, what does that do to critical thought? So that was really eye-opening. And then the thing that stood out to me was the guest lectures. They had the most powerful were actually the current PhD students. They're doing detailed research that they had come and present because things are changing so fast that the curriculum was three or four months old. It was already outdated. So they had to actually have live guests and they were doing research and sharing about their research. Actually one guy is now a big Claude hire. He was doing his research PhD at MIT in AI and had done his undergrad in machine learning. And I actually tried to get the guy on my podcast and he had already signed his employment agreement with Claude and couldn't discuss anything. They basically threaten you in the class, like you can't take any of this. This is this guy's PhD. But that I found to be the most helpful because it was real time learning just to see how fast everything was changing.

Josiah: Well, speaking of real time learning, you just posted a clip of this. I'll include a link in the show notes with you and Roman Faeen talking about vibe coding. You and I have talked about vibe coding a little bit, but to this notion, if things are moving so quickly, I don't know if you feel this way. I feel like the best way to learn about AI is just doing stuff, building stuff, and I've really spent some time every day for the last couple months vibe coding because I find I learn so much. Do you feel the same? Do you feel like you're learning just by doing, by building stuff?

Sloan: Absolutely. I actually use it for the pod. Anytime I have a guest on, I do all the research through AI. I also have a context document that I feed it, so I do advisory services where I'm chairman of a SaaS company now. I'm then the executive advisor for a pre-series A company, and I've started doing this advisory work using my past CEO skills to help kind of new upcoming companies. And I actually have a context document for each one of those advisory agreements that every time I turn on Claude Code, I feed the context document so it remembers everything and then I go in and start doing analysis. I mean, it's unbelievable. I actually have tinkered with the coding both personal and professional. This past weekend, my son and I tried to create some code for a Roblox game using it. Roblox is kind of a closed system, so anybody listening and you try to do that with your kid, it's actually kind of difficult to do, but it was cool. Literally we're just in the context window telling it a paragraph of what we wanted to do and then watching it build the code.

Josiah: I love that. I want to talk a little bit more about the coding piece, but the use of Claude Code is interesting and just the name itself I think scares a lot of people off. It seems like you gotta be a programmer. We can get into this a little bit more, but that use case you alluded to, you are kind of wearing these different hats, these different roles that you're operating from. And for me this has been the big shift in my own use of AI over the past, since November of 22. I feel like these chatbots I've treated by and large sort of like a glorified Google, asking things. And with Claude Code and other tools, there's an ability to save context, which is sort of a game changer, right? Because the memory persists over time. How did you go about setting this up? Because it sounds like you have a context file for each of these different roles that you have. What did the setup look like?

Sloan: So what I do before I actually, if I'm doing work, I'll actually ask Claude specifically, this is who I am in this scenario. So I'm an executive advisor for this pre-series A company. So I give it that context and I say, this is the task or this problem we're trying to solve. Give me all the things that you need, all the context you need to provide an optimal solution for it, and then it'll feed me back. I actually did this this morning. It gave me 10 questions and then I answered the questions and I gave it bigger context, and then I said to it, create a full sequencing document for this, using a certain type of sequencing, and then see where we made mistakes. Correct it and then question it again if you've made a mistake and correct it and then publish it. And then it actually does an iterative approach. So I think you can't just feed it. It's an iterative context, but if you ask it what it needs for an optimal solution, it'll tell you.

Josiah: Which is amazing because I think this is a fairly recent development, at least as far as I'm aware of. Each LLM is wired a little bit differently, and so I think in the past when I've been trying to give context, sometimes they'll do a voice note or try to upload documents. Sometimes those play a role, but more information isn't always better. So I think being very clear, here's the role, here's the outcome I need. What do you need from me?

Sloan: And you can get token degradation within Claude. So if you keep talking in the same thread, you'll actually start to get diminished results, and then eventually you may limit out if you're doing really big problem solving because the context window still isn't infinite yet, although they're working on that. So I actually will take a document that I'm working on, and then like the sequencing document, I have now 10 versions of it because what I'll do, I'll think of something, I'll feed the sequencing document in a Word document back to the LLM and say, this is the latest version, I want to work on A, B, C, D, and I want you to reframe it. Instead of thinking as a hotel CEO, I want you to think as a hotel owner, and give it that, and then it'll reframe the document and fix the problems.

Sloan: That's actually how I created the document that is a prompting guide for hotel GMs. That was an iterative approach working with AI. I basically took my MIT coursework, created a base document, and then I worked with Claude to build it out over a couple of days, and that's how the end product ended up happening.

Josiah: I think that's incredible. I'll include the link where people can learn more about that, but I am very curious, let's use this as an example for the general managers listening. I think whether it is kind of with a system like this, or maybe this is a bridge into talking about vibe coding. Let's say I'm running a hotel property, how should I be thinking about this on property now, in early 2026?

Sloan: I think the two biggest things as a hotel GM that are big unlocks that you can have, but you're going to have to work with, I don't want somebody going rogue and feeding it proprietary information. So you need to think about security protocols because once you feed it to the LLM, what happens, right? If you have something that's proprietary to the company. So you really gotta think through that. But if it's allowed within your company security parameters, what I do is I actually connect my Gmail. I have a note taker that feeds into Claude, and then I have friends who use Slack a lot in their companies. They have their Slack channel, and it can basically help prompt and organize your communication, your calendar, almost like an executive assistant. And it can also do auto generation of emails, which I have it doing now. I don't have it automatically send them. So if you emailed me today, it'll actually go to Claude and it'll propose a suggested reply. And then I'll go in and I'd say half the time I'll tweak it and then send it. So it just makes me far more efficient. And for a GM you're probably getting, when I was hotel CEO, I was getting 400 emails a day. A GM's probably getting 50. It'll make you more efficient so you're not stuck in the office.

Sloan: The second thing that it can do that is really powerful, but again, you need to be very careful. I use it for financial modeling. And it can do everything from a discounted cash flow to, if you have a profit and loss statement, if you are allowed to feed it, if your company allows it, you can feed it and it'll do analysis for you. It can do basically, most profit and loss statements don't have GL code data, but you have ProfitSword. You could have some of that ProfitSword data fed through an LLM, do some financial analysis. You could even, I've done some benchmarking using Claude and I fed it multiple hotel statements that were something I'm not worried about confidentiality around. And I said, compare these and tell me where in the rooms department we have some costs that are out of control. It will spit back some really insightful analysis. So I think it can be the financial analyst and the executive assistant for communication for a general manager to make them more efficient.

Josiah: Super interesting. I mean, to our earlier point, security's really important. I think you need to talk with your legal counsel about what this looks like in your organization. At a minimum though, it would shock you how many people I talk to that are using these free tools, which by default, I don't think you have an option to turn off the little toggle where it's like, I'm not going to allow this interaction to train the models. So I think spend that 20 bucks a month, even if it's on your own dime, these tools are powerful enough. Spend the 20 bucks, turn off the data sharing, at a minimum you need to be doing that. But it's interesting, I think for connecting email, in your situation, are you seeing middleware for that? Is it direct? Claude right into Gmail?

Sloan: I actually do a direct Gmail into Claude, but you can use like Otter and Fixer AI have a middleware solution where it can even take your calendar, your Gmail, or if you're using Outlook. I use Gmail. And then it can basically funnel that, or you can do a feed into Claude. But I actually have Gmail going directly into Claude.

Josiah: Interesting. I'm not doing this. I was looking at middleware, and middleware is maybe useful in some cases, but it seems like these technologies are evolving so that it's becoming easier and easier.

Sloan: Well, you can do a direct integration with Claude now and Slack too. Because I have a good friend of mine, we do a little business on the side, but he uses Slack for his company and he was showing me, and it's not that hard to set up.

Josiah: Interesting. As you've been setting up and experimenting, what surprised you? Because it's amazing to kind of have the idea of this. Was anything harder than you thought? Easier than you thought?

Sloan: I think there are two things that surprised me negatively that I think people need to be cognizant of, and the models are getting better, is that the prior model of ChatGPT is really creative, but it will hallucinate a lot. The thing I like about Claude with technical writing is that if you specifically tell it not to hallucinate, it won't, meaning it won't make things up. I was drafting some stuff for my newsletter. I wanted very progressive companies outside of hotels that had used AI to automate workflows, and it made up some of them. It gave me a couple of real ones and then blended in a few fake ones, and then it gave me actually false URLs. ChatGPT, because of how they built it, actually has a biasing system that wants to please more often. Claude and a few others are more likely to tell you, I can't do that, or I don't know. But I always, if I'm doing something that is going to be in a public format, I always say, do not hallucinate. Make sure this is factual. Double check it, tell it that. So that is one lesson I've learned the hard way. And I would say ChatGPT, which is the most used by people, actually hallucinates the most, and it's something they're working on, but it's a real problem.

Sloan: The second thing is that on the Claude Code side, what I've found is that it needs very specific details on API integrations. So it can build something that's really cool, but you may not be able to compile it or it may not actually work. You can literally feed it a workflow and say, I want you to create an AI agent that's going to automate this, but then it can't integrate with anything. And so I think there'll be these companies that are integrators, like you said, a middleware company that are very valuable. And if you're trying to create something, the integrations are often more difficult, more technical. They may not, even for certain systems, you think about a lot of PMSs. There are PMSs in hotels today that don't even have an API. So if you're trying to deal with some of the interaction with the PMS, a lot of times it won't work because the integration isn't there.

Josiah: It feels like a whole other conversation, but I feel like that is going to be more and more of a risk if you're not able to integrate, because I feel like this will become table stakes. You're just going to want your LLMs to interact. And so we'll kind of see where this goes. But this is why it's so important, I think, for those listening to us to be experimenting with this stuff because I feel like you might read something or see something, and until you're actually building, you don't know what is actually true. And we're recording this at the end of February, 2026, but if someone's watching this in a month, it might be different. So I think that evergreen takeaway here is you have to be building. And I think on the vibe coding side, you alluded to this earlier, this has been my experience. It's easy. You can tell an LLM to create some dashboard or something like that. It can do it in seconds. The real hard part is the integrations of the data, the interactivity, because creating a front end, all these leading LLMs can do in seconds. It looks beautiful. It looks magical, and then actually making it interactive, workable software often requires a database. And so that was something that's been interesting for me. It's worth it in some cases. But I've been, I'm curious to hear more about what you're building. I've been experimenting with everything from trying to do software builds to just a daily email that's useful for me. For example, as of today, it is decently good at doing web searches. It's not good at the specificity I need to make it an interesting email, and I think I'm testing all the LLMs, all the middleware. It's just like the technology isn't there to do it perfectly how I want it. I think that's probably going to evolve. But have you run into some of that stuff with the building that you're doing?

Sloan: Oh, a hundred percent. I mean, that's where you have to know some of what you're exactly looking for in iteration. I think that's where, if you look at Claude Code, for example, vibe coding can create a pretty decent first iteration, but it's not going to create commercial grade product. You still need an engineering team. Now that may eventually change, they may eventually become so good, but we're not there yet. The same is, if you use Claude to develop a new PowerPoint presentation or you use Gamma, which I've done recently using both, I probably had to go in and edit 30%. So it's a good base starter, but there's still work to do. And I think there are AI that are good at different things. I think Claude is the best for technical writing, the best overall LLM, the best for coding. But then I still think you have other, like ImageFX, which is part of Gemini, is the best for image generation by far. You then have Manus, M-A-N-U-S, if you haven't checked that out, check it out. That is the best at actually automating actual workflows.

Josiah: It's super interesting and I think this is where people have to be using it because I've been experimenting with this the last couple weeks and it's interesting what it can do, but it raises questions that are maybe not immediately obvious. Because to do these workflows that I wanted, it can technically do it. But it's going to be thousands of dollars of tokens and credits. And so that's where you have to be in it, otherwise you just don't know that. I think before testing, I was sort of like, can it do it or not? And then there's a whole range of questions that emerge once you dig into it. It's like, is it worth paying this much? Is it worth waiting this long?

Sloan: Yeah, and I think it'll be interesting to see how that evolves, how the token cost comes down. And also I think, what is energy cost going to look like long term too? Because I've seen projections that on the current trajectory, AI by five to 10 years from now will consume more than 50% of the world's energy. I mean, it's exponentially significant. If you kind of run the numbers now, I think they're solving some of that, making the models more efficient. So to do things, it doesn't take as much power, but that's a whole other consideration. What is the energy needed for this stuff to keep going up and to the right?

Josiah: I don't know if you feel this way Sloan, but I feel simultaneously very excited and energized. I find on flights now I spend the whole flight vibe coding because it's almost addictive. It pulls me in. I want to make things. It's so fun. At the same time, I feel that because our capabilities have increased, I feel so far behind and increasingly so. And so there's sort of this fatigue and this feeling of, how do I stay in a mentally good place even as I am so excited about the capabilities. Do you feel any of that?

Sloan: To an extent. I feel like I'm so much further ahead than most people in hospitality, so that makes me feel better. When I was in the MIT class, there was a ton of people in healthcare that were there. There was a lot of people in the actual tech industry, a lot of SaaS business leaders, so I was way behind them. But we work in an industry, you and I do, that we love and we're not going to leave. And so I guess my anxiety is eased because you and I are probably in the top 10th percentile of adopters in this industry. But yes, if we compare ourselves to others, we may be mediocre. My point there is I feel pretty good when I think about it as a hotelier compared to everyone. Yeah, we're maybe a little behind, but it's just moving so fast. It was really clear in the MIT course, the PhD students knew more than the professors because they were in the weeds on how things were changing. The two professors that were the leads, I think one gentleman was probably 60 now, incredibly smart. He sits on one of Google's boards, but he's not in it. The PhD students are doing all this research. That's how fast things are moving.

Josiah: One of the reasons I like talking with you, Sloan, is I can ask you CEO questions. You have this perspective of what does it take to lead effectively, and I feel like in the past when it came to leading, you didn't necessarily need to know how things were done specifically. But at the same time I see you leaning into the technical details of doing things. Is this wave of innovation a little different? Would you encourage other CEOs to actually be vibe coding now, getting into the details just so they can understand the true capabilities of what's possible, or is that not really needed?

Sloan: So there's a theory, I forget who coined it, that if you look at most business in the future, most value's going to reside in innovative, smaller companies. And you also look at what's happening from a branding perspective. People are branding themselves, right? That's partly what I'm doing. Not Done is my brand. Friends of Sloan dinners is part of my advisory brand. And you're seeing that more. And so I do think in a new world of AI where IQ is faster, you can tokenize it, individual contribution matters more. And smaller companies are going to be leaner, meaner, are going to disrupt legacy industries. And so having capabilities that are deeply individual matters more.

Sloan: So if you want to learn in this new world, whereas I was CEO of a large 10,000 employee hotel company, you get to a point, and I think a lot of leaders do this, you're really good at your job and you maybe aren't learning as much. But I think in this new world, if you really want to evolve and survive, you do need to understand. I'm not saying I'm going to go be the head engineer of any company. I don't have that capacity. But you have to have enough techno know-how to be able to, if you go start a company, you're also the chief product officer, which is what Brian Chesky did at Airbnb. He understands technically enough and he created a ton of value. You look at the companies that are leaning in and growing exponentially, you tend to have some pretty technical expertise or really good understanding of the consumer by the leader at the top, and that is changing.

Sloan: I think the old days of large companies like General Electric and things like that where you have a leader at the top that is kind of a general manager of the business, today's leader has to know more and be more capable. And AI is leveling that out to make you more efficient, to increase your capacity to learn areas. And so I think this is the new leader. If you want to be a CEO in 2035, you better be very astute in AI and understand how that's impacting the business. And I think that's actually where a lot of the leadership in the hotel industry is going to get upended. We're going to have a whole wave that needs to kind of sunset. And I don't mean that in a mean way. That's just part of the evolution of business. And I don't want, I'm 44, I'm going to be 45 in May. I don't want to be part of the old guard and have a big wave crash on top of me. So I just see this coming and I'm trying to be part of it rather than have the wave take me out, if you will.

Josiah: It's awesome to see you leaning in and learning and building and being Not Done, as your brand's all about.