Agentic AI (Part 2): The Emerging AI Tech Stack

I think what essentially is happening is, yes, we are raising the bar on the ability for everybody to do a certain level of information but i think what's actually also happening is we are raising the bar on the ability for a company like canary to do even more outstanding things and so it's it's like the the level of what software can do now is just far greater.

Well, hello, and thank you for joining us on the Travel Tech Insider. I'm Kara Whitehill. And I'm Gilad Beringstein. We are thrilled to have you back with us. In today's conversation, we're continuing our deep dive into artificial intelligence and the coming agentic era. Last week, we started at the macro level by exploring the future of distribution, and there's a lot more for us to say on that front. And next week, we'll focus on personalization.

But today, we want to talk about the future of building and what a lot of people refer to as the new AI development stack. But before we jump in, Kara, what's new in your world? I am so excited for this episode. I've been playing around with all of the different LLMs and testing out these different agentic applications. And, you know, it's still obviously very new, but...

it's easy to go down a rabbit hole. So I'm very excited for this conversation this afternoon. What about you? Yeah, similarly, I think this is one of the broadest topics we'll cover this season that's really applicable to every company. So I'm super excited about that. And then, of course, over the last... couple of weeks we've seen a lot of agentic ai browsers being launched

Comet was the first in perplexity. We saw Strawberry just the other day. OpenAI released their new browser. So a lot happening in the world of agentic AI. Great. Well, generative AI is not only impacting how travel is bought and sold, as we heard in part one of our AI effect. series. It's also impacting how companies are building and delivering the travel services that are bought and sold.

Companies used to stockpile servers, engineers, and access to bandwidth, all of which required considerable capital. But in this new AI era, anyone can use Replit or Lovable or choose your own adventure there to spin up their own travel app in a matter of hours.

Companies no longer need teams of expensive engineers to code up a new application. They just need a cursor subscription and a few developers with some free time to Vibe code. Legacy infrastructure is getting deprecated in favor of AI-based solutions that didn't exist a few months ago.

So what does the enterprise tech stack look like in this era of AI and how is it changing the build versus buy decision? Is it proving to be more cost effective than the legacy platforms our industry was built on? Is it unlocking opportunity for startups with fresh ideas?

is to take share from incumbents? Or is it enabling the incumbents to finally catch up the startups in terms of speed of innovation? Or is it some combination or collaboration of both? Well, to help us dive into this new future, we have two perfect guests joining us today. We have Josh Dow, the VP of Hotel Technology at Wyndham.

and S.J. Sani, the co-founder and president of Canera Technologies. So let's get started. Great. Well, thanks to both of you for joining us today. Before we dive into all the nitty-gritty...

details here. We want to understand a little bit about the lay of the land in each of your companies and what your tech and engineering organizations look like in this. kind of tumultuous era of AI building. So can you start us off, Josh, talk a little bit about your organization and your company, how that's shifting perhaps in terms of your engineering capacity and focus on AI?

Sure. Thanks, Gary. Look, we are primarily a buy shop. Great partners, actually, with SJ. I'm excited to be here because we do a lot of work together. So we do engineering, but it's primarily in the kind of core organization. registration, you know, web services and data layer. We believe others out there in the industry are moving quick and building great products. So we like to integrate with them.

We've got a common kind of cloud platform that we started a migration on many years ago. All of our data is kind of out of the hotels now. So PMS, CRS, CRM, all those sorts of things. We moved early, I guess, on that. pretty much entirely cloud at this point. So that's given us a bit of an advantage to move more quickly and innovate more quickly.

Look, I think AI now is showing up in a lot of point solutions for us in platforms, in individual products, but we're also using it to bridge between services now. So I think it's an interesting time to be...

evolving the tech stack, I guess, with all these new AI tools. For sure. And SJ, how about you? Yeah. One thing, though, I think the 100-plus engineers on our team would take offense to... Vibe coding taking away their jobs. We'll come back to that. It's an interesting conversation, actually. I think we have some strong opinions around where that actually has a place and where, candidly, it's just...

overhyped right now and leading to false conclusions. But coming back to the question, so, you know, quick overview, Canary. We are a vertical SaaS business for hotels. Think about us as, you know, in between every interaction between the guest and the hotel. So the guest management system is our flagship solution through luck, hard work, right place, right time.

We have become the leader in vertical AI for hospitality and digitizing every touchpoint between the guests and the staff. And so for us, you know, the role of AI is not just the inputs of how we built our company. but in the outputs of how we help our partners across the industry serve their customers. We work with the largest brands in the world, Windham being one of them, obviously. And I think it's been exciting to be at the forefront of delivering that value to our end customers.

Absolutely. And we always like to start these conversations at the basics. I want to come back to exactly what SJ was just talking about in a few minutes here. But to start us off, SJ, help us define what is agentic AI? Yeah. So, you know, I think there's an era of the internet which is about information retrieval. So there's a whole bunch of information out there.

You type in a few words, maybe in a box called Google, and you hit search. And then the job that to be done in that era was about like, please retrieve that information that correlates to the words I put there. Right. And that's kind of like where the world that we have been living in for a long time. And then there's this world of do something like, hey. Book me the flight, not show me the prices of the flight from New York.

to Paris, but book it for me. So do the work for me. So I think that's the biggest difference in my head of just retrieving information to doing something, taking an action. Do not show.

yeah look we think about it kind of two ways right now right the the I've heard people I would not expect to be talking about MCP servers, talking about MCP servers around my office lately. You know, so look, I think we all are, at least I have my eyes on the prize of agentic booking. I know lots of work and hard dollars are going into it, both at the big OTAs and third parties out there and with us from the brand level. So I think we will get there. I think we know the industry is...

is headed in that direction. But I think it's going to be, I think that the adoption curve is steeper than I would like, right? And probably than SJ would like. You know, I think it's going to take time for folks to get comfortable with agents.

booking on their behalf, right? It's going to take time for folks to get comfortable putting their wallets in the hands of, you know, some LLM service. But I do think we'll get there. You know, in the short term, I think the big opportunity for... the way we're looking at it is

you know, agents acting as sort of a bridge or connective tissue, right? Taking systems that would not normally take actions and bridging those to the, you know, oftentimes the people who can take those actions, right? So don't just look in canary. platform they have a great messaging toolkit that you can use to the guests can interact with the front desk and send messages to them and then the

the person at that desk is going to swivel over to their housekeeping application or their, you know, their property management platform and then push the message to a housekeeper or write down a note and hand it to the housekeeper, right? And we see a lot of opportunity for agents to step up in that.

that place right and act as a bridge between platforms so i think we're we're going to see it evolve on multiple fronts so hopefully hopefully quickly on the booking side i'm real excited about it but you know i also think the reality is we are all far more excited than perhaps most of our consumers are yeah And that's a great point, because I think by default, we tend to think of these things from the consumer's lens. But I think there's a lot of other interesting applications.

afoot within the company itself and within your kind of engineering and building organizations? Do you see areas where you could deploy different agents within your kind of...

you know, internal use cases and how you're building out your products or services or kind of internal uses. Yeah, definitely. I mean, look, we are... we're we've started the groundwork i guess on this already we use agent force we're big salesforce customers and so we're using agent force for franchisee interaction with us we are using using Copilot for internal corporate use cases, right? And I think the next evolution of that is...

You know, don't just answer a question, but schedule a meeting, right? Or, you know, connect me to the tool. We've spent a lot of time building a single authentication platform for pretty much everything that our franchisees do. And so we have the ability to orchestrate these.

user journeys between tools. We've been having a lot of conversations around revenue optimization, right? And look, things like, I don't know if all of you have seen the old hotel tape chart, right? But trying to maximize for the density.

of rooms and maximize the gaps for length of stay. It's a real thing. And a lot of our property management systems don't do a good job of optimizing that. And so it's people who do it, right? But you can quickly train up an agent and tell it what the objective is and have it kind of sort order those rooms. optimize for a maximum length to stay um you know and so i think things like that will will spin up at the edges look inevitably is that a

Is that a product that somebody is going to be able to sell? Probably not for a real long time, right? I wouldn't bet on the long-term health of a company that did that. But I think the property management systems and the other big players in the industry will adopt that functionality pretty quickly, right? And so I think...

you have the ability now to innovate quickly and try these things and get some operational advantage while it becomes part of the platforms that we use every day. That's how I see that taking place. I was going to say, I think there's like... Internally, obviously, I also view it as two phases of adoption. There's the LLM adoption, so taking large corpus of data and text and synthesizing it or taking action items from it.

That's kind of one use case. Our marketing team uses it. Our product managers use it. We look at feedback across our products. Millions and millions of guests use our products daily. There's many touch points of taking that data and synthesizing it, recordings of gone calls and how our reps are doing. So that's kind of like one area of taking large corpus of text, essentially.

using LLMs to synthesize some case action. Then there's the more agentic version, because even that, some may argue, is not agentic. It's still retrieval, but with LLMs. We do use Cloud Code quite heavily. Our engineering team does use Cloud Code heavily. It drives lots of things, automatically triggers tests. It actually writes the tests for our engineers.

and then pulls documentation, right? So it can know context knowledge bases to pull documentation from and then use the code that's written plus the context to actually write the test and then run the test.

So there's that area of agentic AI as well that we use internally as well. I think testing is an amazing use case there. We spend a lot of time. in i think oftentimes more time in testing than we do in development and being able to to kind of stop the like waterfall development approach right and bundling everything together to take to to be the most efficient that you can with all

of your test environments and test case execution. That is one of those things that I think is a big opportunity for all industries, right, who do any kind of testing at all, because it is just so time consuming, right? being able to automate that and move more quickly and be more agile, right? Deploy smaller things more quickly because you're not worried about the end-to-end test cycles. I think that is huge.

I wouldn't say it's low risk, but it's lower risk than right guest facing production workloads, right? Automating your test suite. So it's definitely something I'd recommend. And Cara, this is like, you know, going back to the, at the top when you were mentioning kind of. Viacoding and folks building things. One thing I think about a lot, we hire really, really talented engineers at Canary. We have a pretty high bar. Obviously, I'm biased.

discount what i'm saying but like i truly believe we have a really high bar um and what what i i am not a naysayer or doubter on the ability for agents to write code. We do it. Literally, I practice it. We do it. But what I think the difference is, if you rewind to Yahoo directory, like the directory of links,

And then if I told you like that is the way the web was going to be, then you can be like, well, anybody can do this. Anybody can like build a list of links and websites. I think what essentially is happening is, yes, we are raising the bar on the ability for everybody to do. a certain level of information but i think what's actually also happening is we are raising the bar on the ability for a company like canary to do even more

Outstanding things. And so it's like the level of what software can do now. is just far greater. You can request an early check-in through a chat. We can look it up. We can price it for you. We can approve it. We can mark the room cleaned in the PMS status. Send the guest a text message. Your room is ready with your key.

So we've now can do a lot more because we are using AI to power software, but then so can the analysts at Wyndham. Maybe the analysts at Wyndham used to have to ask a vendor to... take this data and synthesize it, but now they can do that themselves. But that doesn't mean that we don't have a lot to do. We actually have more to do now because the possibilities are far greater. And so I think that's kind of going back to your point earlier.

The engineers at tech companies are not out of jobs. It's only the companies where there is not net new things to do where they might be out of jobs. And fortunately at Canary, we're not one of them. Yeah.

Yeah. Well, and that's, I mean, that's a great point. And I was, I was going to ask and Josh, you anticipated my question because I was going to ask what are some, you know, ideal use cases where you're seeing agentic really deliver. you know, a lot of efficiencies in terms of, you know, cost or time or whatever. And I think SJ, to your point, it's not that it replaces the engineers, it just lets them do things a lot faster and frees them up.

to work on even more stuff, right, and increase throughput. So, yeah, it's really fun to see. I'd be curious if either of you have any kind of early metrics that you can share on, you know, how much more productive or how much more throughput you're seeing. as a result of deploying some of these authentic applications? Not getting too forward-looking. We're still in the early phases on a couple of these things, but we are...

exploring voice in contact center. For us, that is, look, we do millions of phone calls. And a lot of people call for things that are not revenue generating, right? Maybe in the big picture. confirming your reservation or making a modification or or maybe because you could cancel right you'll call back another day and make another reservation but all of those what we call non-revenue calls we field a lot of those and so taking those calls and defer them off to an AI.

To just look at it, it's a weird story. And I was telling somebody about it recently. They couldn't believe it was a thing. But we get thousands of phone calls from people who just want to confirm that we still have their reservation.

Right. So they don't want to change their reservation. They don't want to do anything with it. They're not calling for a special request or an early check in. They just want to make sure they're getting ready to get on the plane. They want to make sure we still have the reservation. And that's just super low hanging fruit. Right. It's not something where.

you need a human personal touch interaction, right? They just want to confirm what's your last name and what's your confirmation number or phone number. Great. We do have your reservation on record here. You're good to go. We're looking forward to your stay. of those calls a year that are confirmations or cancellations or what's the pet fee or what are the pool hours.

And so deferring those calls into an AIQ where the customer gets the answer they're looking for nice and quick, but you're not chewing up time on a dedicated... right, seller who is answering those bookable calls. And so what we're seeing is a lower average handle time, right? The phones are getting answered more quickly.

The questions are getting answered more quickly once the call is answered. And our people who are great at helping folks book their vacations are able to spend the time actually on booking calls. So it kind of superpowers them while deferring all of that other volume.

We are seeing early wins on it. You know, no numbers that I'd want to share publicly today, but we are seeing both ASA and average handle time improve. And the booking conversion percentage is going up, right? Because we're deflecting all of those calls. from the bookable agents. Yeah, I mean, I'll piggyback on that. Obviously, I'm not in a place where I can share specific numbers that our customers are getting, like Wyndham and others are getting from these solutions.

But what I will say is that the similar analogy there is a lot of the low-hanging fruit questions and answers via text or voice now do not need to go to the well-trained call center agent. um who's you know selling you your vacation um and and so i think there's a um level efficiency that's happening that is it is um more quantified in the

their conversion on those calls is actually greater. It used to be the case, for example, like there's a group we work with where two or 3% of their calls used to get converted to bookings. Now 7% of their calls get converted to bookings. And so there's, think about that as a business owner.

Now, like, is that person just more effective? Like, and, you know, yes, they are. Cause they're not like getting off the phone with this one to go to the next one or God forbid, like telling you to hold, you're about to give them money. They're telling you to hold.

Because someone's in front of them and they're like trying to check them in. Like it's absurd. It's like someone's calling the hotel front desk. They're like, I want to give you money. And you're like, sorry, can you hold? Right. And then like click. And then what happens? The worst of it. They go to booking.com. And then the hotel owner is like, oh my God, there goes 20%. Oh my goodness. So that's like a lot of external value. On the internal front, you know, I'll be honest here.

We looked at a couple of input metrics. We looked at like pull requests per engineer. We looked at lines of code produced. I think those are still vanity, if I'm honest with you, right? Like, you know. Is the number of pull requests actually a good metric to look at or not? Lines of code has never been a good metric to look at from engineering productivity. So we don't have a good one right now on individual productivity, but we are noodling on.

you know, different things. I think tests are like the one big thing that we have honed in on is it used to be X percent of code used to get released without test cases. And that is significantly down near zero. Great. Well, that tripling of conversion revenue is pretty impactful too. Yeah, absolutely.

And there was so much gold in that. We're actually going to come back to that same topic in a minute and try to unpack even further where AI and Gen AI has been integrated into your tech stack. And specifically, the two points you guys made at the end there that I think are really important.

is one that we still lack kpis we still don't actually have an agreed upon metric by engineering leaders across any industry of how to measure these things and then of course how to translate that into roi and return for your board and your investors and then the The other thing that you mentioned, Esther, that I love is leveling up your staff. And the fact that if we do our...

play our cards right, this AI will make sure that our staff members have a more enjoyable job working on more important problems. And I think there's a lot of great stuff in that I want to unpack. So before we move on to some more detailed topics here, looking at your crystal balls over the next

I'm curious if either of you think your companies will have fewer engineers in 24 months than they have today, or if you think they'll have more engineers in 24 months than they have today. Maybe we'll put you first in the hot seat, Josh.

I, you know, look, I'm in a tough seat to make a bet on that. I am a betting man. But I think it probably remains about the same. but not because of AI, right? Look, we will continue to execute on as much as we can. We like to do a lot with a little at Wyndham. But I mentioned earlier, we're really a partner-focused organization, right? And so my...

My engineers are SJ's engineers, right? And all across the spectrum of partners that we work with. So I'm going to rely on him, and then I will tell you whether or not I'll have more engineers based on his answer. Yeah, I mean, I think a lot of our... So for us, it's definitely more. And that's because a lot of our customers in North America and Europe and APAC.

realize the level of service and the level of automation, perhaps, that software can provide to their business and their end customers. And so, like I said, you know, a few minutes ago, the... Opportunity set of things hotels can do with software now is growing far faster than the productivity gains that AI is giving engineers. And so said another way. There's just a lot more to do, a lot more things software can do. So I 100% think we are going to have more engineers.

And one other thing that I want to touch on, and, you know, Josh, I know this came up at the Destination AI event a few weeks ago, but, you know, we talk a lot about AI and generative AI, and people go immediately to chat GPT and the LLMs, but it's really broader than that, right? And there's some conversation around robotics and visual spatial AI.

What are you seeing in that front? Is that something that you're actively exploring and playing around with and kind of leveraging different generative AI tools in that realm? Yeah, look, I think robotics is... I'll call it a hobby at this point. There are a lot of great companies out there. It's funny, usually I skip one high tech, so I do every other year. Not that I'm recommending that. Everybody should go to high tech.

every year uh but i i usually go every other year and the last two i went back to back and the number of robotics companies uh i think was six or seven x uh in in between the last two so there were just you know and look some of them look we're not putting and i'm not throwing shade here right but we're not putting a robotic cocktail mixer in our lobbies anytime soon um but the look i've i've got a robotic vacuum

and a little duster at my house, right? And there's no reason most of our hotels shouldn't. I just took a trip around China not too long ago. And I will tell you, they are winning the race. on robotics in hotels um i've seen a lot of really cool applications there the you know whether it's

Are they running the right race? You know, that's a good question. Look, we all talk about the labor crunch, right? It is tough to hire people. It's tough to keep staff. It's compensation is always a challenge. Look, I think it...

will come here. If I had to dust the crystal ball off and make another bet, I do think robotics are coming for the hotel market and they're going to make their way to North America. We're seeing big numbers of adoption in China. And I think it's the... modular stuff that is most interesting to me right buying one single service towel delivery robot isn't going to make any sense for anybody right but but i recently saw a company that that is doing a modular unit where you're kind of

buying the navigation smarts uh in in the robotic base and then it changes which which um you know tool for lack of a better way to put it is it's carrying around right so it can dust and mop and vacuum and deliver towels and deliver room

service and it also can interface with the vending machine which i thought was really cool so it will actually pull up to the vending machine and get the thing dispensed that you want in the room and then right deliver it to the room and i just think look the the cost structure on those now they're not going to work

for most of our economy in mid-scale hotels, where the price is at right now. But where the price on that device would have been three years ago, right, wouldn't have made sense for any hotel, except those who were looking to put the headline in the news, right, that they now have a robot in their building.

So I think that's changing really, really quickly. And I do think it's going to have a place here. But I think it is probably in that 12 plus month horizon, right? It's not something we're actively pursuing right now.

yeah i asked that because like we're not really doing anything in robotics um we're not sure yet like besides a vacuum like a better roomba like what else what else makes sense but you know um definitely keep our eyes open on the generative side there's a lot um you know one one thing around generative which is really important is like it's really hard to do general generative ai like you saw from

OpenAI's first foray into generative images and how certain things were like random, you gave it a prompt and the image was like, wait, why did this image come through, right? And so like, especially for the enterprise customers we work with, when we think about generative AI, it's really important, like the way we've architectured ourselves too. So we, you know, we work across all the foundation models. We have our own.

homegrown hospitality intelligence layer on top of the foundation models. And this is like the orchestration context, data prompts, evals, et cetera. And then we, cause that's necessary in our world.

to do generative because for example when our customers come to our platform and are you know adding in an upsell and they're the upsell is like the king room with the balcony and they don't have the perfect image but they have a image of the king room that shows the balcony our generative ai could touch up that image to show the perfect image for the king room and the balcony but like that is not

You can't take that generic hotel image and put it in chat GPT and say, I want that because it'll give you something that maybe is overlooking a beach or the French Riviera. And you'll be like, oh, you know. My hotel is not there. And so I think generative AI, we're using it all over the place to help our customers augment incomplete data they have.

It could be as simple as contact information, could be as simple as profile information, could be as interesting as the 360 of the room if they only have a few images. And so, yeah, we're using it. externally all over the place. I love it. All of our hotels are on the beach now. It's great. Yeah, yeah. Well, if you tell ChatGPT to give you a king view of the room for all your hotels, they might be...

It might be on the French Riviera or on a beach. Yeah, well, this is a really interesting balancing act that reminds me back of the U-trip days in the before generative AI era and the question of what's actually better for consumer. is a photo that's generically held like in the right direction, more helpful than no photo at all. And I certainly don't know what the answer to that question is.

But I want to bring us back to these kind of balancing acts and trade-offs that engineering leaders and startup founders are making all over the place. And I'm very lucky to be close with many of the leaders from across our industry. And these topics come up.

all the time like how should they be thinking about this how should they be balancing these kind of alternatives that are coming up due to ai and i'm hoping that you guys can help all of us in our audience and in the industry to think about these difficult balancing acts and the first one i want to talk about is the build versus buy. And we'll start with you, Josh, and then we'll ask SJ how you think about it from the mature startup perspective.

But historically, Josh, you shared already that you guys were a buy shop or a licensed shop where you brought a lot of technology from the outside due to AI-generated code, due to AI engineering assistance, all those kinds of things. Is your perspective changing? Do you see yourself maybe trying to build more technology in the future? And then SJ, we're going to ask you the same question from a mature startup perspective. Yeah, look, I think...

The generative tools allow you to incubate new ideas quickly. I've spent more weekends than I'd like to admit messing around trying to build my own apps and trying out all the things that are out there. So I think you can... you can iterate more quickly on ideas, right? You can A, B test some things. You could spin up a little proof of concept, but I don't yet feel like, and look, we're not a big engineering shop, right? The way that Canary is. So I think we...

We'll pilot some things faster, right? We'll try out some new ideas more quickly. But I still think, look, we're going to buy, right? We're going to partner with the best providers out there. We are a hotel franchising company, right? We are not a... software shop. And so I would much rather rely on the discipline of those partners that we work with to kind of move things along quickly. You know, we are big and we help to steer the roadmap as SJ has dealt with, right? Some ideas.

he likes and some he doesn't um and so i you know i think we'll continue to lean in there more so than than on the build side you know i mentioned us our focus being more on, you know, web services and integration layers. I think that'll continue, right? We want the ability to plug in new kind of endpoint systems into that overall ecosystem. And so we think that's important and it's something that...

we want control over, right? We want to build redundancy there. And candidly, right, I love what Canary is doing today and the partners that we've chosen. But if we need to replace one at the edge, right, I don't want to have to depend on the partner that we're leaving to go work with some. new to hand over that that ecosystem right and so making sure that we control that

That service bus or that integration layer, I think, is really important. So that's kind of where we're focusing our build today. Yeah, and I think it's really interesting that you brought up there, Josh, as well, being a big company and having the ability to impact the roadmap of your client. And I'd be curious if we had a smaller hotel chain, if they would have a different perspective on testing no code.

Now, SJ, I want to bring you into this, and I want to add one more element to the question for you, kind of the build, license, but then the actually buy, like the actually acquire. So as a mature startup with some funding and some revenue, with the ability to make some acquisitions,

balance all three the build it in-house the license from a third party versus the acquire yeah so on that question let me and then i'll unpack your other questions too on on acquisitions you know i would love to tell you like hey here's the canary philosophy put in the equation pops out the answer we don't we don't try to think about that way at least at this stage um you know we have hundreds of millions of dollars we have firepower but every decision is quite precise every

decision around whether we're building this or buying this is a first principle decision um you may see us build something and then buy something and be like well that doesn't make any sense and i think it's about just unpacking exactly like that opportunity um and so you know we try to think about it that way um in terms of you know canary we we look we're not um the things that are commoditized we buy um the things that are

innovative, we build. We are not building Verizon cables and AT&T cables to do text messaging from guest to guests, but we are building the communication layer. and the translation layer and the AI layer to make that happen. So I think generally it's like whatever is really good off the shelf, like don't reinvent the wheel. But the new opportunities that are created, we lean into heavily and build for our partners.

that's great and you know i want to go back to this conversation you know you mentioned that you anticipate having more engineers sj um in time and josh you said probably keeping the same amount and i I was thinking through that in the sense that we kind of use the term engineers very generically here. And I'm actually more interested to understand what is the skill set evolution that you're anticipating or that you're already seeing and kind of the type.

of quote unquote engineers or the type of developer technical folks that you have on your team or that you anticipate needing to bring onto your team? Is that changing? And then do you have to change any of the way? that you're managing these teams and these projects as well. Yeah, look, from my perspective, I think...

We need, and we have today, right, a great product organization that SJ and his team work real closely with. And I think that is the most important from our perspective, right, more so. than traditional engineers or developers. I think having that product organization, look, we have a lot of franchisees and they all have their own unique take on how they want to run their business and what they need to be successful.

And so our product organization acts as that bridge between our franchisees. And look, we have a lot of brand franchise advisory councils who spend their time with us giving their feedback. And so that. that piece of the machine that distills that feedback and and understands look these folks many of them come from a hotel background right more so than technology and so they can sort of bridge that gap or act as a translation layer between what the

franchisees need and what our tech providers have on offer, right? And help to build that bridge and put, you know, whether it's a new product or it's just an enhancement to what we're doing, right? We spend a lot of time together with Canary's teams helping to shape.

what it is that we think we need right on behalf of the voices that we get from our franchisees. And so that to me is, is really important. And it's a great asset of our organization that we have that, that good product organization. Because I think that is probably more important and more likely to grow in need rather than maybe traditional dev engineering and test resources. Yeah, I think Josh's product actually is like...

now going to be more important. You know, like we, you know, we get feedback from Josh at Windham. We get feedback from our partners at Best Western.

marriott hilton and one of the beautiful things about being a vertical sas business kind of makes our job easy you know like yeah you know uh um i think everyone knows this but i'll say it out loud like a brand brand a tells us this is important and then you know three months from now we end up doing it the reality is like brand a looks a lot like brand b looks like brand c looks like brand d like we're you know

We're not a feature factory per se, but when our partners bring good ideas, it actually... is quite often highly applicable to everybody else. And so it's like a, it's a really, really good virtuous cycle where our partners feel like a lot of the requests they give us and throw over the board, like, oh yeah, canary leans in, canary leans in.

But this is because we have a really good product manager and customer success team, too, that connects the dots that, hey, Josh at Wyndham said this, XYZ it. this brand said this and our European customers said this and APAC customers said this, like, oh my God, there's like a large cone centric circle here. And this is like right up the middle. And so.

Product managers have actually become more valuable because there's just so much more engineering can do. And then you have to have really strong product managers to synthesize all the things that are coming from users and customers. to be able to prioritize. Because engineering is really good at building what you organize. But you've got to know what to build.

And, you know, how to sequence it, which I think has become more important. Absolutely. Absolutely. I have a thing I've been writing about a lot recently where I say I think the technical product manager. is going to become the single most important cog in the product engineering organization because they are that product brain you talked about, but they have the technical ability to bring together the best tools and services out there.

So I love that answer. Now, as we begin to wrap up, we're hoping to share some advice and perspective with our community and our industry. And with all the caveats we all need to do that, we're still in chapter one of the great commoditization. It's all been rolling out. We're all learning together. I'm giving you all the cover here that no one's expecting. The answer is going to last for years and years here.

But we're hoping to share advice and perspectives with our industry. So we're going to ask you guys a few questions, share your perspective, keep it a little bit light, and we'll see what we can learn together. So the first topic I want to talk about is what you're most excited about.

as you think about genitive AI and agentic AI coming into development and coming into technology. So what should founders in our audience and CEOs of companies where technology is a big piece of their offering, what should they be most excited about over the next couple of years? Josh, you want to kick us off? Yeah, sure. It's acceleration.

I think that is the name of the game. You can move faster. You can test faster. You can prototype new ideas. You can adapt to feedback, I think, more quickly than you could in the past. I always undersell the effort maybe, at least when I'm trying to negotiate with Harmon SJ and the guys. But I think it really is just that pace, right?

Look, we have gone from talking about AI having an impact to like all of the different ways it will and how it's now built into every product stack, right? And it's, you know, look, some of that is hype cycle.

There's no question, right? There are lots of things out there that don't need AI and we'll probably get them anyway. But I do think the ability to just... iteratively add new functionality and try new things is is um it's happening at a pace i've not seen in the 25 or so years i've been at this thing um and so i'm really excited about that and i think that is the thing you're aging yourself

I'm starting to measure in decades now, SJ, not years. But look, I just think that pace is really exciting to me. And we're really excited each year at each new trade show or each new event to see what cool new... platforms and products are there. I'm excited about like, there's this part of me that is excited about proving people wrong. I'll give you an example. Like the biggest issue our AI messaging or AI web chat product has.

is people think it's just a chat bot. And like, there's a decade of crap chat bots out there where like 90% of the time you talk to it and it doesn't do anything. And so, you know, there's, there's like this, like, I feel like.

in this moment um we're trying to fight the good fight that like hey ai generative ai communication is not a chat bot even though like you open up a website and there's a thing on the bottom right that you can like message this is different this is different and so like um it's really annoying but when we do get over the hump with some of some of the most

forward-looking customers on it it's really rewarding because then you see like you see the organization then be like oh my god like yeah this is not a chap i'm like yeah we've been telling you it's not a chatbot but so i think there's this like there's this um And I think the root of that is like being at the cutting edge of it, Tara and Gail, like we know the value that this thing can provide. It's like, it's like a new form of electricity and we know the light bulbs it can turn on.

And we just want to show people that, yeah, I can turn these light bulbs on. And then they turn it on and finish the analogy. They see the light. And it's really rewarding. It's really rewarding to get to that stop. yeah i think that shifts right we we talked about it in in the agentic conversation right i think that shift from providing information to taking action

is going to be that inflection point for people, right? When it's not just, can I get a late checkout? Yes, you can get a late checkout by contacting the front desk and that might cost you $20, right? To thank you, Mr. Dow. We're glad you're a platinum member. We've automatically extended your stay for you, right? In my housekeeping app, I've told the housekeeper, I've sent a message to the housekeeper that, hey, this is going to be a late checkout, and so clean this room at 4 p.m. All that.

Totally. Agreed. That orchestration is going to be really cool. Yeah, yeah. And let's touch on briefly before we wrap up the cost structure. So how are you thinking about the changes?

in the cost side of all of this in terms of, you know, whether you're building it internally or, you know, the folks that you're buying or licensing from and kind of how they're constructing this. Are you... Are you seeing the cost come down quickly? Or is that something that you think is going to take a little more time? Not quickly enough. No, look, I think it's...

We recognize and are trying to structure agreements in a way that gives us that flexibility, right? Because nobody wants to bet on it, right? We all have this hypothesis that compute cost is going to come down and it's going to get cheaper and easier. But nobody wants to negotiate lower. prices in the in the subsequent years of their agreement right um so i i think that uh you know for now we're focusing on on shorter terms and more windows um

Look, the piece that we're seeing the most pull on that end is really the consumption-based calls, right? And look, Canary sits in the middle of it for us, right? We are pulling data from... from both our PMSs, from our CRS, from our loyalty provider, from all of those different orchestration components, and they all now meter their API calls.

right and so it's no longer just a oh you're a customer and you have data here you can come get it whenever you want right everybody wants to to put a little api tax on that and so that is getting it's you know it's not something i would say is a barrier right but it's something you need to be cautious of because the cost of the service may not

fully cover right the cost of all the inputs that that service is going to need and so that's something that we are seeing change in the in the cost structure right as we now need to be considerate of okay great we have this new toy but it's going to need these data sources and it's going to

I'll put the data here. And what's all that going to cost, right, in data and API metering? So I think that's something that folks need to be aware of. It can be kind of tricky to forecast, right, because you feature creep. And right now I want to tell Josh that, right, thanks for being a...

platinum member, and so then I have to make a call to the loyalty system that wasn't part of the original business case, right? And so you have to be careful in balancing the how much personalization versus what the cost of that personalization is when you're trying to calculate your ROI on things. Got it. SJ, quickly, any follow-ups to that? Top of mind? I think this is our responsibility as technology providers.

Our customers rely on us to deal with the ebbs and flows of the input costs, like the cost of a token. I don't want our brand customers thinking about that. I think if some things are expensive right now and we want to be at the forefront, then we pay that. And then we pay at the front. Maybe we benefit at the latter years.

That's capitalism. And so I think, I think to move the, to move the future forward, I think that's, that's kind of our approach is like very rarely a canary will you hear our team say, The input costs of solving this problem are too high. Like that doesn't actually happen. Because I think it's about, is this problem worth solving for our global hotel customers and their guests?

And that's like the first question. And then it's like, can we solve it? Because, you know, like I said, we don't build robots that clean rooms yet.

And so I think that's most of our conversations. I don't think I've had a conversation in the last couple of years about input costs. Outstanding. Well, guys, this was an amazing conversation. We're so grateful you're with us because... So much is changing in the way we build technology, deploy technology, maintain technology. So we're so grateful for you sharing your wisdom with us and with our audience.

But before we let you go, we do have a quick lightning round. You'll have 15 seconds apiece, and Kara is going to kick us off. All right, SJ, starting with you, what's your favorite recent AI hack or quick win? Automatically responding to candidates that email me about job applications. Very good. Nice. Josh, how about you?

You know, I think I've seen a lot of growth now in the content segment, right? So when we think about SEO and GEO, there are a lot of great companies popping up now who will... analyze the data from the llms and tell you where you have gaps in your content platform i think that's really valuable and it's something that that's going to create a lot of opportunity that is awesome question number two and you're going to go first josh

Will agentic AI make our industry better at offering personalization or is it going to stay about the same? This is the biggest softball question you have ever given me, Galeed. Yes, it will make it much, much easier to personalize. You know, I think finding the balance in what's worth personalizing and what isn't, right, and where that creepy factor sits.

is going to be the more pressing problem for us, but I have no doubt that AI will make personalization much easier for us. Amazing. And SJ, what's your take? I have a 30-second answer. Personalization has been... the false promise of hospitality for decades and decades and i think what vendors forgot to tell their customers is to personalize you need two things good and structured data

And then hotels were like, well, I don't have good or structured data. And so personalization fell and it's literally on its face. But now, now with LLMs and large language models, you do not necessarily need structured data. And some may argue like not all of it needs to be good because we may be able to parse through the good and the bad. And so I actually, you know, I don't want to be that vendor now in 2025 being like personalizations finally here, but.

i will be that vendor in 2025 because like we have been personalizing millions of guest days for our hotel customers and their guests yeah that's great final question uh sj back to you what was the last thing that you built with a no code application. Or is that blasphemous to ask an engineer that question? No. So a lot of our prototyping happens through software like Replit or V0, where we can use Replit.

and B0 to prototype things. I think this email parser that tells me looks at people's candidates' resumes. and see who's a fit for what, and then I can automatically reply and introduce them to the hiring manager if they're a good candidate, I think has been the cool recent thing. Great. Josh? I'm going to take you down a personal road. So I just traded in my car and bought a new Toyota. And their connected services are awful. But...

there is an API endpoint that you can consume. And so I am currently neck deep. My weekend project is connecting my car to my home automation platform so that when I pull in, the thing it doesn't do is it doesn't automatically lock the car when you... walk away. And so I'm building it so I can just automatically lock it when I, when my distance isn't where the car is, it'll lock the doors for me. So I can't give you any hospitality applicability, but this is my little passion project.

But in all seriousness, this is something he would have never paid an engineer to do. But now that he can do it, he can do it. And now engineers don't have to, they can do... Other things. Yeah, exactly. My car is probably going to get stolen because I'm not going to have a secure endpoint. But listen, I'll worry about that another day. Very good.

That is a perfect note to end this conversation on. Thank you both so much for joining us. Wonderful. Well, thank you all. We really appreciate it. Great conversation. Thanks for having us. Yeah, thanks very much. It's been great being here.

Oh, that was so interesting. And I another one of those where I felt like we could easily go down a rabbit hole and end up with like an eight hour Lex Friedman style conversation if we aren't careful. But yeah, what was what was your. your main takeaway? Well, a lot. I mean, first and foremost, like we said at the beginning, we're still at chapter one of the great commoditization and we're all trying to figure it out. And we heard that.

from our guests when they couldn't even tell us what the right KPIs were to measure the impact of no code on their engineering team. And they're not unique, right? We talked to companies all across the economy and no one knows how to measure these things. So I thought that was a really good encapsulation of the fact that we're-

just still at the very beginning of this whole process. But a few specifics that I thought were really interesting. One is I really liked SJ's answer to the build versus license versus acquire. question and really talking about going back to first principles. And this is something that Warren Buffett always talks about, how at Berkshire Hathaway, they have no rules or processes.

Every time there's an opportunity, they make the best decision they can. And I just thought that was a really, not just honest, but a really useful answer at a time where all the rule books are kind of thrown out the window, we're creating them anew. So I really loved his answers from a first principles perspective. And then I found it really interesting that both of our guests...

said that in two years, there are companies who either have more engineers or the same number of engineers. And that's, of course, very contrary to what Eric Schmidt and some of the big voices in Silicon Valley are saying. And, of course, I'm really excited to see how this continues to play. out and i guess the last thing is i love josh's answer about acceleration and when i asked them what should ceos be most excited about

from the coming of agentic AI into coding and into technology creation. And he talked about this acceleration in prototyping and testing and learning. And for me, as an entrepreneur turned investor, that is obviously an incredibly... exciting opportunity for us to see that acceleration. So those are a few of the things that I loved. What about you? Yeah. You know, a couple of things. I liked SJ's comment about raising the bar and how, you know.

all of these emerging technologies and applications and different areas of AI and generative AI are really bringing everybody's level of competency up, right? And I think that's important. And I think that speaks to the fact. probably ties into the fact of why they're not going to lose engineers, right? It's just making everybody that much more productive so they can...

Increased throughput, right? So I thought that was important. I think, too, I personally liked hearing the comments about product management and that skill set becoming, you know, important, more important. as, you know, this kind of AI wave really takes hold within companies. You know, I spent part of my career as a product manager early on. And, you know, I think it's...

It's one of those roles that's valued differently in different organizations. But, you know, from the people who live and breathe it, it really is such a unique skill set required because it is that translation layer. And you have to translate, you know, business needs. You have to translate customers.

customer request. And, you know, you're not just, you're not just like feeding information back and forth. You're actually having to think about and be thoughtful and intentional about, you know, how you prioritize these things in this world where, you know, to SJ and Josh's points. at different parts of the conversation.

You know, when you have so much more capacity and so much more ability to build things faster and to deploy things faster, you have to be even more deliberate about what things you focus on, right? Yeah. And I really liked... hearing that in particular, that they both anticipate the need for a true product management skill set to become even more important. So I thought that was nice. And I like both their comments at the end on personalization.

You know, a little tag at the end there, but yeah, I'm pleased that they, you know, I think we all agree like there's something there and getting personalization right and agentic applications will have a hand in that, but it really does go back to.

you know, kind of knowing the person. And to SJ's point, it no longer has to be good and or structured data because you can kind of, you know, tap into so many more signals and kind of that blob of information you have about users. So I thought that was...

interesting as well. Totally. And I would say, too, that, you know, we've heard so much about personalization from companies who are under the perspective they can just use an LLM to offer personalization without collecting data about the individual.

And that just doesn't work. And it's something that no matter how good your AI is, if you don't have data about me, you actually can't ever personalize for me specifically. And I thought that was a really important point that we heard. The other thing that keeps coming back to my mind as I think about the conversation.

is that both of our guests touched on the concept of hype. And I was asked this question at a conference, I can't remember where, earlier this fall, about whether AI was overhyped. And I just thought they did a really nice job of making the same point that I like to make, which is that I think there's a lot of AI startups that are overhyped and lots of AI tools that are overhyped. But I think SJ said this. I think AI in general is completely underhyped.

And the amount of transformational impact it's going to have on our economy and our businesses and our lives, I actually think is underhyped, even while at the same time, we probably have valuation bubbles and other things happening. And I thought they did a nice job of kind of showing their personal excitement for the. Yeah, that reminds me of, I think it was Michael Coletta's comment in one of our episodes last year where he said he doesn't think we're talking about AI enough. Absolutely.

You know, so yeah, it's, it's fair point. I think it's, uh, it's still early days. So more conversation to come. Thank you for tuning in to the travel tech insider podcast. We hope you enjoyed today's episode. Thanks to our most excellent producer, Zach Vanoss of Studio 42 for working his podcast magic. To keep up with all our latest episodes, subscribe to the Travel Tech Insider on your favorite podcast platform.

And if you'd like to join the conversation, follow our page on LinkedIn, as well as our YouTube and WhatsApp channels, and be sure to like, comment, and share your hot takes on the latest in travel technology. Questions, comments, or suggestions for future episodes? We'd love to hear from you. So reach out on the socials or email us at traveltechinsider at gmail.com. We'll be back soon with another exciting episode. Until then, enjoy the trip.