The App-less Enterprise:

Awesome. Good morning. Good afternoon. Good evening, everyone. Thank you so much for joining today's session.

The Atlas Enterprise, How AI Agents and Unified Data Will change the way we use software. Before we begin, I'd like to cover a few housekeeping items. Closed captioning is available. Simply hover over the virtual stage and click the CC button located at the bottom of your screen.

We encourage you to submit your questions and engage in the chat, leveraging the q and a and chat tabs in the engagement panel on your right. And we'll do our best to answer questions at the end of the presentation. And if we are unable to get to your question, we'll follow-up with you directly.

We've also added some additional resources which are available in the docs tab and the engagement panel on your right. There, you can find some additional related content. And lastly, today's webinar will be available on demand after we wrap up, it will be sent to you via email directly.

Our esteemed speakers for today's session are Anthony Dayton, Tamr's CEO, and Rich Miner, cofounder of Android and founding member of Google Ventures. Welcome to you both. Now over to you, Anthony.

Awesome. Thanks, Kara. And let me, sort of underscore and reiterate the point that Kara made about the q and a. Please the hope is we can make this interactive and folks can, you know, share thoughts and questions, in that panel, and and we will do our best to sort of, go through those.

And the idea behind today's webinar is is to make it a bit more interactive, and, really, this comes out of a series of conversations Rich and I have had, over the last few months even, longer than that. And I thought it'd be really interesting and fun for the community to kinda hear from Rich about the way he's thinking about the future of enterprise software. So but before we get into that, some quick introductions. So, Rich, is a, full disclosure, tamer board member, and but also has a long history in this space. He while completing his PhD at UMass Lowell, he actually started two companies.

One was Avid, which I'm sure many people know as the world's first digital video editing platform that's been really the foundation of a lot of almost all commercial movies over the last twenty five years, and Wildfire, which was a commercial voice assistant, an early precursor to, tools like, Siri and whatnot. And and but, you know, but thirty years ago. So, Wildfire was acquired by Orange, and Rich, joined Orange as the VP of innovation where he helped start Orange Ventures. And is at his time at Orange Ventures that he recognized the need for an open mobile phone platform, which he helped found in Android, which was quickly acquired by Google, which probably saw the same thing. And at Google, Rich helped build and launch Google Ventures, as a, you know, really world class venture capital fund. And having or since leaving Google, he's, started three more companies.

He's an active angel investor, in his fund XVC, and as I mentioned, obviously, also a a Tamr board member.

As it relates to me, I began my career in enterprise software, with Siebel Systems in Silicon Valley, learned the trade, with Tom and Pat and Dave Schmier and the founding team there, and then joined a small Swedish software company called Click and, later, a German software company called Celonis. And I it's, like, a common theme between Qlik and Celonis was giving people new ways to visualize and work with information.

And the common theme I would hear from customers of both solutions was they love the visualization.

They hated the data. And so for me, the excitement of joining Tamr is to maybe, for the first time ever, make a crack in the challenge, which is messy, unorganized, enterprise data. So with that as a a backdrop in terms of who we are, let's talk a little bit about what we wanna talk about today.

So we really wanna talk about this idea of the Atlas enterprise, And but instead of jumping right to the punch line, I think it would be helpful, to give a little context for folks in terms of how we come to some of these ideas. And we're gonna sort of take a step back and do a little bit of a a history lesson.

Both Rich and I have obviously a lot of experience in the enterprise software space and the software space more generally, and so share some of our perspective in terms of how we think about the space. I think it's also important to ground ourselves in the actual capabilities of AI systems, and there's obviously some debate in the industry, as we speak on this point.

And then I think we, we will get into the meat of what is this Atlas enterprise, what do we mean by that, and talk about what the implication of the Atlas enterprise is on enterprise data. So that's sort of the the flow of the conversation today. And so I thought we might start, Rich, and talk about this connection between consumer software, and enterprise software. And in some sense, you and I represent, maybe, barbell opposite ends of a barbell here, although you've obviously life experience in both domains, of consumer software and enterprise software. And I I've had this point that I've made, over the years, which I've playfully called Anthony's axiom, which is that enterprise software is just reflects in a way what we see in consumer software five years ago.

And I threw a couple of examples up here on the screen. You know, obviously, Google search as a consumer experience launches, and then not exactly five years, but roughly five years later, you start to see enterprise search tools like Solr, but, also fast for SharePoint, etcetera.

And maybe even Google Workspace itself really is a enterprise version of Google.

Also, if you think about a tool like Facebook, which introduced the idea of a news feed, and, obviously, Facebook does many things, but this idea of sort of tracking the events, in your personal life then translates into the enterprise and experiences like Slack where we're sort of tracking or Yammer, the one acquired by Microsoft, tracking the experiences we have in our work lives. And then the other example, obviously, is is Wikipedia with consumer knowledge and information mirrored by what we see in the enterprise with things like Confluence.

So but, anyway, I'm curious from your perspective.

There's one very obvious missing example here, which I'll leave to you, I guess. But, your view on Anthony's Axiom.

Yeah. I think it's, from my perspective, it's it's partly right.

And so I do think I think the key thing here is the user experience is led by consumer. Consumers, kind of by definition, have a lower tolerance for pain, on the user interface. They, because the user base can be so broad across consumers and this idea in the enterprise, you might have domain experts who are getting paid to do a job. They they are willing to learn how to to to how to use hard to use software.

You can't do that with consumers. So it has to be simple. And I think that so the the simplicity, the elegance of user interfaces, the search box interface, what we're now seeing with voice interfaces, but even the graphical user interface, I think these were, you know, probably, you know, probably made, you know, the bigger mark and a bigger shout was from consumer adoption of those experience. I think if you trace them back, it's a little more subtle, and we don't need to get into that.

But, you know, the graphical user interface popularized by the Mac and then Windows for consumer was first developed for the Xerox Star, which was, you know, with an enterprise office productivity in mind. Right? So I do think there are a lot of inventions in software that came out of an enterprise demand because that's where there was potentially gonna be profit. The word processor, again, popularized as the Wang word processor as these devices used in the office.

But then but then they really the the interface gets simplified through that adoption, you know, early invention, then kind of mass adoption for consumer makes things a lot more, consumable and easier to use and and benefits the implementation and cycle of those things in the enterprise. So I think we're effectively saying the same thing. Certainly, from a consumer experience, those trends happen first with consumers, and then they figure out how to simplify enterprise software with the same the same kind of experiences.

Yeah. I would also add that I think, that there's a the personal experience of these technologies in our personal lives then translate into work. So we we have these in the example of just to pick the first one, Google, many people have been googling things for many years and then would come to work and not be able to Google things. And they would say, oh, this is so frustrating. I can do this in my personal life. Why can't I do this at work? And and then the market steps in to address that gap.

Yep. Nope. I would I would, I would agree with that. Right? And, again, that's that's the that's the elegance of the demand.

You know, boy, if I can do this in my consumer or if it's this easy for me to do something with a consumer app, why can't I have that same simplicity in the enterprise? And you saw that even, you know, with the adoption of mobile devices. Right? The preferred mobile device was, you know, the rim, you know, aft you know, and and kind of more sophisticated or complicated, I guess, I'd argue, feature phones.

And then, you know, most enterprises didn't want to adopt iPhones because they weren't easy to, you know, to when they first came out, they weren't very enterprise friendly. But when your CEO and all your c levels start, you know, walking around with iPhones and and kind of demanding that they have access to systems, then you have to you have to bring that level of interface back into the enterprise and figure out how to make it a supported tool in in the enterprise even though IT initially might have resisted it.

Yes.

Couldn't agree more. And in that with that same spirit and sort of stepping back a little bit and thinking about where we stand today with the the rise of AI, again, I think it's useful to put this in some historical context.

And this is a framework you've developed, and I think it's very useful to as a way to think about this.

So maybe you just share your thoughts and thinking behind, where we stand today in in terms of this broad megatrend of AI?

Sure. And I and I think the AI example probably is is a wave that we that I think unlike the other ones, which felt relatively cyclical, and I'll explain that, I think AI is one that feels fundamentally different. And I've lived through effectively all of these, transitions. I learned how to program on punch cards, and, had some of the first PCs, back in high school.

So, you know, one but one perspective is we I think people tend to make a big deal out of how things change. Right? How the mainframe computer to the computer on your desk to then the, you know, to then the web, and then to the mobile phone, like, how each of these felt like huge impacts, and they were. They were huge impacts in terms of the markets created, the opportunities created, the companies that disappeared and appeared with each of those transitions.

You know, and so, I think in fact, there might even be a data general computer on the left, which doesn't exist anymore or a Wang word processor, which Wang doesn't exist anymore. So those but the interesting thing, if you really look at those cycles as a computer scientist or what was going on with the compute models or even the UI, there really wasn't a lot of difference in each one. It was the packaging, but we started with mainframes, these big centralized, effectively personal computers because they were dedicated to whoever was who had time on the mainframe at that time. It was their computer.

Right? And then, and then we introduced time sharing, and the mainframe was shared between a lot of people that had terminals that would share between different people and an operating system that supported the sharing of that big centralized mainframe computer between lots of people. And so it felt like you had your own mainframe, but you didn't. Now and then and then we went to the PC, and that PC was on your desk, and that felt very different.

It was now yours, and you could open it up and play with it. And then and then all of a sudden and you ran all your programs locally on your PC.

You know, so again, getting back to, like, the mainframe when it all ran on there. But then you went to the web, and all of a sudden you didn't run things on the PC. It was back in a centralized cloud based system that was connected over a network somewhere that the compute was happening, and it was just the interface in that browser on your PC.

But then mobile phones were back to PC were back to personal computers again because this thing in your pocket is just again, I've got my apps on this device. I'm running them here. And so the interesting thing is those were cycles.

I would argue there's not an awful lot different between an app running on this mobile phone and when I had it on the PC. I would argue there wasn't a huge difference between when I was, you know, time sharing on a screen, a bit of a computer, and had the data on the computer and just the interface on the screen to a lot of our web experiences today.

But but each of those steps enabled lots of different apps and lots of different capabilities. But the funny thing was the compute model really didn't change that much through all of those iterations. We used very similar types of programming languages called Algol style programming languages, c, Fortran.

Right?

You know, Pascal, you know, then Java and Perl and other languages, but they were either interpreted or compiled languages that were still kinda step by step languages. And it really just was, is the computer you know, is that program running here on my device, or is it running like this webcast on some centralized server somewhere?

The the big thing about AI that's different is that in each of those transitions, the man machine user interface didn't change a lot, and we'll talk about that in a minute.

But I think with with AI, it is a trend. It is definitely gonna be as impactful as any one of these preceding it. And I think I think, you know, for reasons we'll continue to discuss, I believe that this one, it feels different than past trends. It feels like it's gonna definitely have at least as much impact as each of those previous trends, but I think potentially more and for some, you know, some fun reasons that we'll, you know, hopefully chat more about.

Yeah. No.

Agreed. And I I it's funny you mentioned that you did your first programming on on punch cards. My first experience of punch cards is having my dad bring them home from the Dartmouth, and I would make paper airplanes out of them. So less programming, more functional use. But I think you're the the meta point is really right, which is really we've been the debate has been where we locate the compute.

And as we'll talk about, the question and opportunity that I think we we are presented with is really a different kind of compute model. But but let's get to that. Before we get to that, let's talk a little bit about, I think, the underlying idea here, which is disruptive technology.

And, we won't go through this in excruciating detail, but I think it's good to locate the our time today in the context of this isn't the first time we've seen a disruption in a market.

And, one criticism of at least from and I had the benefit of taking Clay Christensen's class, when he was teaching it.

It's it's a very good description of a situation, but it's a very poor at describing what causes that situation or how you would identify that you're in a a moment of disruption.

It's almost always like, you almost tell executives, you'll know it when it happened, and then you're too late to react, which is feels very, unsatisfying.

But one of the things that, I might sort of add to this idea, like, how do you know you're in a moment of disruption?

There are these kind of telltale signs, and I think we see a lot of these currently in the AI space. So, for example, when you're part of a disruption, it occurs because there's a platform shift. There's something outside of you that happens to you. I call this exogenous, platform shift. Something happens to you, that allows you to change the dimensions of competition.

Often, it's changes the price value, mix. So all of a sudden you go from, you can you can see, like, see not like ten percent price improvements, but ten or a hundred x price improvements. And you also see, what I labeled the incumbents doubling down. So you see this, incumbents will sort of blow it off and say this isn't a real change. It's only a minor thing. And you also see the market getting bigger.

I think this is a really important idea that that Clay points out, which is that in these disruptions, you see a whole new set of customers. And, you know, like, if you think about something like AI, you have new people who can be programmers because now they they can they don't have to there isn't the high barrier to entry there. And last but not least, this idea of futurization, which is incumbents will just tack on AI. And so, oh, no. We now we have AI because we added a chatbot. But I'm curious, Rich, your thoughts here. And and, I think it would be fair to characterize you and Android in particular as a true disruptor.

And, you know, you you really, you know, led the the charge in disrupting an industry and a market. But your view on on this.

Yeah. I mean, I so I, you know, I think that I I quote this a a lot for people because, say, I I see a lot of companies that, you know, kind of bury their hands heads in the sand when there's evolutions, and and I've I've, for decades, say you have to disrupt or be disrupted. Like, and you you you know, whenever there's a new trend coming, you wanna be the one that disrupts you. And it's really hard for companies to do that because, you know, disrupting it, you know, and this is why they double down, you know, puts at risk all of their current revenue streams and how they do business. And it's really hard for companies to do that because companies are really set up to largely do, ten percent incremental improvement. Like, they focus on very incremental, let's you know, this thing's gonna make this thing ten percent more efficient. Wanna make our sales team ten percent more efficient or maybe twenty percent.

One of the things I liked about the early days at Google was Larry Page fundamentally believed that something wasn't worth doing unless it gave you a ten x improvement, not a ten percent, but a ten x.

And and that drove a lot of projects, and that drove a lot of people's thinking. People like Jeff Dean would often quote that to their teams.

And that's how I've always thought. I always like working on, you know, big disruptive things that take advantage of evolutions and coming evolutions of technology to do things, and and I and you see it in each wave. Some people either can get ahead of that and and realize it's gonna be disruptive, or they, you know, suffer and and, you know and again, if we go back to that PC example, right, the WANs, the Apollo computer, the, you know, digital equipment corp, data general, like, all of these, you know, icons of the computer age thirty years ago just disappeared like dinosaurs because they couldn't understand the PC and the impact it was gonna have. In fact, were, you know, the incumbents, they were all denying it.

And then it was the same thing in the mobile phone space. I sat down with Lazardis at at RIM, you know, at at Sergei's request very early on in the days of Android and tried to get him to get onboard and adopt Android. And for him, the OS just wasn't important. Apps weren't important.

They had the core communication apps. They had the core productivity apps, and they had a keyboard. That was all they need. And and so, you know, for them, they just thought the form factor won, not the OS.

And they again, you know, they don't exist anymore. So I think you see these cycles. I think the interesting thing here is I don't know many incumbents that actually and this is pretty different for an evolution like this.

They might not know what to do, but they're all panicked and trying to do something relative to AI. And and when they and when they for their first attempt fails, like with Facebook, they get freaked out and and are spending billions to try and correct that. Right? And if you look at Salesforce so so it is interesting that I I think somehow most corporate tech leaders, at least, have gotten on board that this is really different.

I don't think a lot of enterprises enterprises have really figured out how much and how different this is gonna be in part because the apps and the solutions aren't there yet. We're still catching up on what it actually means to have, you know, the the the integrations aren't there, etcetera. But I think we're on the cusp of that those pieces getting figured out, and it will certainly be interesting when it when it does. And it will be very disruptive, and I think I think the tech company realize I think enterprises are gonna need to realize we need to get on board thinking about this because cost of doing business, how they do speed of doing business, all of those things are gonna change pretty markedly with the adoption of AI in the enterprise, by those who do it well.

And I think you'll see a gap, you know, in a moat created by those who are doing it well and quickly iterating in those who don't.

So alright. So we agree we're in a disruptive moment. We agree that the consumer experience leads the enterprise experience. The last sort of leg on the stool, so to speak, on the history lesson was to be clear about the nature of this change.

And you alluded to this before, but now let's let's really dig into it. It's this man machine, interface. And if you don't mind, kinda walk people through your thinking here. And and before you start, I think it's worth noting that, I mean, you literally are a, world expert, and and you've spent much of your career thinking very deeply about how people interact with their computer.

Again, to underscore this point, you started wildfire, which, you know, in its time was just a completely radical way of thinking about how you interact with a computer.

You know, maybe too early. Well, you can just you could tell us. And, certainly, the mobile phone, would would be another example. So, I mean, I cannot think of a better person on planet Earth literally to be thinking about this. So what's your thoughts here?

Yeah. And by the way, Wildfire was was you know, unique in implementation, but anybody who watched our Trek saw that talking to computers was was a thing we were gonna be doing in the future. And a lot of the things I do are kind of led a little bit by what we see in science fiction. So, the you know, but I think that and I alluded to this a little bit when I was talking about those different trends.

Right? There's, like, kind of two key things. One is the interface. The other is apps.

But in in terms of interfaces, even with all of those evolutions from mainframe to, you know, today's mobile phones, web, whatever, you know, and and going back before computers for from the very first machines, and that's why that's a a photo of the controls on a model t there, in the upper left.

The, quote, man machine user interface has never been about the man. It's always been about the machine.

And even when we think you know, when people look at the thing in the upper left and then look at that screen, they're like, oh my god. No. There's nothing like that. One's a machine and one's a beautiful user interface. But if you look at that, quote, machine in the upper left, you'll see levers that you slide back and forth. There's buttons. There's a little pressure gauge, you know, showing you the pressure inside.

I think that was actually steam driven, perhaps. So there's a gauge maybe showing you, something. So all of those, you know, quote, mechanical interactions, buttons, sliders, gauges, are the same things you see on that interface. Right?

You've got a whole row of four sliders down in the kind of lower, lower left. Right? You've got these little kind of dials and gauges, in the upper right. Right?

There's a bunch of buttons in the right in the middle of the screen that you can push on. Those aren't natural things for people to communicate with. We are communicating to a machine, so we're giving you things that translate numbers into a slider. Right?

Instead of just speaking the number or saying a percent, you know, you need to communicate that by sliding something a certain percentage or value up, by indicating something, by pushing something. You don't put I don't poke you when I want your attention. Right? I I might raise my hands or kinda say, you know, at one minute, I got something I wanna say or clear my throat.

So there's nothing about either of those interfaces that are actually, you know, I think evolved from the machine or friendly, and I think that is fundamentally the biggest shift in, in in the user interface today and also the underlying tech that built the user interfaces. Historically, you really couldn't build computer software unless you knew the underlying machine language. So even the programmers, they might not have, you know, been controlling the computer with buttons. They were controlling it with these kind of arcane, you know, forty to sixty year old, Algol style programming languages to tell it instruction by instruction what to do.

And all of those programming instructions are behind these buttons and levers. And the big difference today is that you can you know, the interface to the computer is evolving very rapidly to being the natural user interface that we all, you know, started using literally as soon as we were, you know, shortly after we were born and realized if we're cold and we start screaming, somebody puts a blanket on us. Or if we're hungry and we start screaming, somebody, you know, gives us, some nourishment.

And, and the most natural user interface, again, our voice flailing, our hands, right, our facial expressions, These are how we communicated and got got people to respond to us, and that is the big difference in AI. Or one of the big differences in what AI does is it it actually can respond to gestures and voice and typed word. We communicate to it this exact same way we can communicate with each other. And what that allows you to start to do is, one, you, you know, to to get anything done with a computer today, especially in the enterprise, often is this multistep process.

Right? First, I need to do a data query, produce a table. Once I've got that table, I can then use it to kind of run something to do some analysis. Once I do that analysis, I can generate, you know, some, interesting graphs.

Once I have those graphs, then I can start to write the report and, like, you know, or generate the report with those graphs and the labels and the whatever. And you might need to go to five or six different screens and type on a bunch of different buttons and and enter commands. The big thing about how we naturally communicate with voice or by typing is you can you can give a relatively complex request of take a look at the database. Can you look for some trends?

Right? Graph those for me and produce a report that has those graphs labeled. So I just said in, you know, one sentence something that might have required a whole bunch of different screens and interim steps that I was kind of, you know, dictating and and and pushing through, you know, is represented a little bit by that, you know, this kind of flow through a user interface that you would normally have to do to versus saying that, you know, that one command, that effectively flattens that whole user interface from being you know, and flatten comes from you know, typically, these are menus and submenus and submenus that gets all flattened to a natural language request or query.

Yeah. So this is a a a very fundamental shift in the way we, interact with you. Because I really like this idea that you're putting out there, that the long history here has been, us having to translate our communication into a language that the computer understands. And the and if I could really sharpen it to say that the the big idea here is that now the computers have learned our language, and that's a that is a big and important step forward.

So that raises By the way, and it's not just one directional.

Right? If you do a if you do any sort of, if do a deep search in ChatGPT or some of these other systems, you can ask it a question. It actually comes back and asks you for clarification and for more data and more information. Right? So it's this it becomes this bidirectional conversation that you're having with the computer very different from, you know, from anything that we've seen in the past.

Exactly.

But which of course begs the question, how good are these things? Where are we on this evolution?

And so I'll share a framework that, we've talked about and and may have, may have seen out there.

You know, our my view is that, large language models today are a lot like interns, and by which I mean, the good part about an intern is that they work extraordinarily hard. They're very ambitious, eager to put in the hours, they'll walk work all weekend. The challenge with interns is they don't know anything. They don't they literally know nobody at the company because they're an intern.

They don't know where the the data is. They don't know where the bodies are buried. They don't even you know, they can they're they're just lacking the fundamental knowledge of how to do the job. In contrast, your best employees who've been there forever, they know who to talk to. They know, you know, where the secret files are, etcetera.

And interns, almost by definition, they're smart, but they're not as smart as your best employees. Your best employees on the way in a way, have been road tested. And to make matters worse, a lot of times, these interns are the worst kind of intern. They're an MBA intern, which is to say they've specifically been trained to sound confident and eloquent, like they're always right.

And even when they're wrong, they're gonna really, like, kinda double down on their Oh, you meant Harvard MBA students.

You know, it's, it's like a an albatross.

Or is anybody in the stream that's a Harvard MBA student or an MBA student?

So, you know, it's very funny that we're I can practically see HBS from this, from where I'm There might be a physics PhD as well, intern that also thinks they're always right, perhaps.

But yes.

Well, which actually raises the question, which is, where are we on this curve, and and do we end up with, systems and and intelligent agents that are at, PhD level? And just to support this idea, and then I'd love to get your take on it, Rich, I pulled some data in for this presentation.

I admit that the graph on the left is practically impossible to read, and it doesn't actually matter.

You can go find it yourself if you're interested. The website's called Our World and Data, and they are this is data going back to nineteen forty about the a number of floating point operations needed to train a variety of models. And, you know, these are a variety of different kinds of models from image recognition to even including large language models.

The important point about the graph is that the left axis, the y axis, is exponential. It means that every step change in the y axis is a doubling of the underlying data. And and so what this shows you is that these things are in the similarly, to something like Moore's law, are doubling, very, very frequently. And what I did on the right, not very sophisticated, is I just took that data and I forecasted it forward using that exponential model.

I think I did ten periods, ten years. And what it shows you is that we're it's still very early days in terms of the capabilities of these models.

But let me throw it to you.

I I'm making a a a conjecture here, which is that, you know, we're on the cusp of, relative sophistication. In other words, what we see today is actually pretty basic. Do you agree?

Well, yeah. I mean, I would say I was using that analogy a year and a half to two years ago. I think it's almost obsolete because I think at the rate these are improving, they are starting to be not just interns but experts. And you said the one thing I'll correct you on, I think, is when you said the AI is not as smart as some of your best employees. I would say it's not as experienced, as some of your best employees. I wouldn't but I but I'd I'd I'd challenge yet whether it's it's as smart. It might have not have as much context, and it might not have as much domain experience.

You will note in the PhD version of the slide, I I upped the intelligence from a c to an a. So I Yeah. Exactly. I'm aware of it. Yeah. Agreed.

I think that I think that I think the real conclusion here is whatever limitations you're hearing people talking about LLMs relative to being a smart worker to help you in your enterprise, just assume they're gonna all go away. Like, the only thing that they haven't really shown a trend that they know how to solve yet is, basically, these things have been close to trained on every bit of information in the known world that we have digitally available for them to train them on, and we've run out of information to train the models on to help them get smarter. But even that's being solved because people are figuring out how to generate new data, and generate new, you know, real time testing and real time scenarios for which they can actually continue to train and improve the predictive capabilities or the reasoning capabilities of these models.

So I think it is it should just be assumed within a relatively short period of time, like everyone on this call will still be employed when these things start entering the workforce effectively effectively way smarter than, than any intern is entering the work environment.

And so let's, in a way, let's bring it home. So bringing all these ideas together, I think your, idea and the really the idea behind the webinar is this idea that we're moving to a model where, we're moving we're contrasting this kind of app centric world with this, more, intelligent agent centric world, a a world that moves from, you know, screens and forms, lists and forms where we expect users to do a lot of the work to a model where we instruct an intelligent agent to do work on our behalf.

And nowhere is that more obvious than in the data space. Like, we give people just a ridiculous number of very low level tools to work with data, and, really, the opportunity is to think about collecting that data together, normalizing it, cleansing it, organizing it, and making it available to these AI interfaces.

And so I I thought you know, share with, us, and you had a a beautiful sort of analogy or example of this, what you know, your view on the Atlas enterprise.

Yeah. And it does start with I mean, I think we're you know, we we've talked a little bit about this, the and you and it kind of your thesis about seeing this first to consumer. We're you're already seeing a hint of what an Apple's phone experience could be like in many ways because so many people are starting to make requests like, hey. Get me, you know, get me tickets to the show, or can you get me, you know, order dinner for me.

Right? We we talk to agents, and it's not too hard to see that you're gonna be able to book dinner, book a show, get a ticket, right, get a get a get an Uber or some other ride simply by asking your agent to go off and do that for you. That it's it you're not gonna need to to, you know, to to plan your Friday night, You're not gonna have to use one app for booking the tickets for the show, another app to get you the dinner reservations, and another app to get you the card. You're just gonna ask your agentic experience.

Hey. Help me plan my evening. Do these things for me. And then it's gonna create for you effectively almost a dynamic app.

So this idea of apps going completely away, we're still gonna probably tap on some things on our phone. We're still gonna be swiping through content and looking at stuff. So it's not really truly app less, but the what an app is, who built it is gonna get blurred because I think a lot of things that we traditionally think of as an app are gonna be dynamically created by the agent that you went off and said, do all these things for me. And, and then it's gonna create your Friday night app, which has the button you press when you're ready to leave the house and the car shows up, takes you to the restaurant, and you show them your reservation that you have.

Next on your phone, when you show up at the theater after dinner, it's got your QR code right there to pick up your tickets. Like, all of that was orchestrated for you and assembled for you as a kind of Friday night app, dynamic, ephemeral app by your agent.

And I think that that is gonna rapidly follow and be able to follow into the enterprise. Because if you think of what most enterprise software is that people are using in an enterprise, from the Salesforce system that you buy a bunch of seats for, right, to the ADP or other software for running your payroll, right, to the HR software for managing, you know, people's, you know, benefits and vacations and things, from, the inventory software. Like, if you go back in history, those were all things done by people. Right?

People that kept track of some data. You had a ledger, and that ledger was your double book accounting for handling, balancing your books, and managing, you know, who's you know, who owes you money. Right? Who do you have to pay out to?

Right? Also, track the inventory. Right? All of that was done in ledgers, and and people manage that.

And the payroll was just keeping track of who came in the office and punched in, you know, or out, and then and then adding up the numbers and writing out the paychecks and handing them to people as they left the office. Like, all of these things that we think of as enterprise software were done by people.

So it's not hard to believe, at least for me, that if I have this Uber intelligent AI and I have, and this is an important piece, really, you know, you know, data that I believe in and that is organized and cleaned up and mastered in a way that I have index you know, that I can index it across and have visibility to it across my enterprise, that, that running your business is is simply the AI do helping you do and helping others in the company do these tasks based on the data in the company. And I the example you were talking about, you know, I I was having having solar installed actually turned on today.

Hopefully, nothing has turned off today. It's just being turned up. But, and and talking to the the it's the second time I've worked with this firm. It's, you know, owned you know, started by a father and son.

They've grown considerably, and and especially with tax credits going away. He's getting all of these emails of people that want solar, and then each one of those has to be vetted, and you have to have presales and sales taking a look at that for the opportunity. And then, in addition to that, you need to then cost and quote out the process. And each of these is different teams of experts, typically working with separate systems creating data, but those systems need to talk together.

And this poor guy is trying to figure out how do I connect my HubSpot CRM thing that we were just told to adopt to our, you know, our customer tracking, to our sales system, to our planning systems, to then my my planning needs to know where my employees are, and do I have the resource to do stuff, to then where were the employees and the billing, and then, oh, I need the parts and time and so the inventory.

Those are all a lot of complex systems for a small business or any business of scale, but it's a good example to think of Danny and his dad's business.

So imagine if Danny was to adopt an AI that he first basically, the first day on the job, the AI is like, tell me about your business. Well, we're selling solar. I got a bunch of people that do installs, that quote, and I got all these jobs coming in and it it's just asking a bunch of questions about the business. And it's like, great.

I'm on it. And so what happens? An inbound comes into Danny or one of his sales guys' inboxes. That's grabbed by the AI.

The AI can automatically look up the address of the person that says they want solar installed in their house. It can do lookups on public records to see if there are liens on that address. Right? Does this look like a credible client?

Do they seem like they're in financial trouble? Okay. They look like a good client. Let me take a look at the house.

Oh, here's the location. I can see geo maps. Those are all publicly available, right, enriching the data that I got about the customer with this other outside data. Now I can do a roof analysis and see it looks like they actually have good exposure for sun.

Right? And I can maybe even take a look at some other information and try and size the house based on census or other data or look at it and try and size and understand the house. And then, literally, what was probably three or four people over a week or two to try and get back to the customer, in seconds that, you know, Danny or somebody has something pop up that says, we got a great client. Like, they have this home.

It looks like, you know, based on their electricity usage and needs and what they said in their email, that they could, you know, have a a seven hundred and fifty kilowatt hour you know, ninety five kilowatt hour solar system.

You know, here's, you know, my initial response to them that's proposing that. Do you wanna kind of approve that I send this? And, like, click and there's literally something that would be a multi week process probably. There's already an offer out to that that user.

If that user says, yes. I'd like that, that same AI can see in the company. Well, what resources do we have? When do I think we could get to that house?

What projects are going on? Right? It has all of, like so all of these different systems can be automated. Now there's still people involved.

The experts still need to approve the system. Somebody needs to decide, you know, again, based on the AI going out and maybe checking inventory from the suppliers you work with. Like, well, of the three panels we work with, it looks like we could do this system with either of these two. Does somebody wanna make a decision of which one?

Okay. Great. Forty eight panels, that type of panel, I'm gonna put in an order for that. Right?

So inventory management, tracking, purchasing, all of these things, right, eventually slotting the resources to go out and then time tracking, generating the payroll, generating the invoice to that customer. You could imagine all of those tasks being done without those thirty different software or not thirty, but at least half a dozen or more different software systems that these poor companies need to adopt, integrate. And if you're larger, it's just worse. Right?

If you're a really large company, you have two choices. You either talk to a management consulting firm about building bespoke software for you to automate your processes, which, by the way, it's the same process. Team comes in, asks you how your business operates, you explain it to them. Right?

They charge you millions of dollars to set up the workflows and whatever, and then they charge you licensing for the software that they buy or build or license, and then glue together every piece of software that's glued together is an opportunity for fragility, for things to break, for security threats. Like, most a lot of security threats come from those APIs that we put in systems to integrate those systems. And at the end of the day, most of these processes in the future, I believe and by the way, the the the spec that you gave, that AI through that dialogue, the AI generates a written document that says here's how your business runs.

Right? I will only accept jobs of this size and up to this size. Right? Because, you know, you only do residential.

Right? But then at some point, you go edit that document and say, actually, we're now able to do, you know, industrial projects up to this scale. Anything that's up to so many kilowatt hours, your workflow just changed by editing that document that is the the script and the job description between you and that AI system that's implementing these things. So in you know now, again, you can only do this if your data is organized clean, can be accessed by that AI.

But if you do that and you have that AI and that process, which, again, nobody can deliver that today, But if you look at that timeline, it's five years or less when this vision can be realized.

Yeah. So we're very much on the cusp of this, and I I I wanna sort of highlight this point and really to connect the dots of what we've talked about.

Historically, businesses would compete by having human beings who understood the systems and interfaces and and data that sits behind the work that they do. Like, I'm a, you know, I'm a invoice, and, you know, data entry person, or I'm a sales I mean, or it's extreme. I'm a sales rep, or, I'm a software user interface designer. It's like my job is just defined by the data I work with and the interface I engage with. And I think your point is, that increasingly as we organize the data of the enterprise, connect these different silos, and put an intelligent agent on top of that, it can really orchestrate much of that work directed by an intelligent human being.

Yep. And and and the processes. Again, it's the data and it's the processes of the company and the decision makers and key players in the company, you know, again, still being a participant in in all the decision making. Yep.

So with that as context, let us, open it up to you, the audience, q and a. And Cara kindly joined us to, guide that process. So over to you.

Yeah. Thank you both. If you do have questions, feel free to enter them into the q and a box on your right hand side. But we actually already have a couple for you. So first one from Chris, information curation requires lots of time typically from domain experts. How can we keep humans in the loop in the age of a gentric age when innovation cycle times are getting so short?

Yeah. So, one, assumption embedded in the question is, or or or one way to think about this is, look, when do we escalate for a human expert?

And if in the current model without an AI interface, the answer is everything goes through a human interface.

To use Rich's solar example, every order requires a human to sort of do data entry and do a bunch of work with it, and even if there's no even if it's very basic.

So I think the the way to think about this is to center human energy on the cases in which the AI is struggling, where it's less obvious and where it needs to escalate for decision to use the prior example, Rich, you made.

Do you agree, Rich, or did you Yeah.

No. I do. I think you I think you said it quite well. I don't know that I need to I mean, you you are the experts on cleaning up data and and and, and so, and yeah.

But yeah. And but the the general idea here is, the reason, probabilistic interfaces are so helpful here is because they can give a degree of certainty, whereas atomic transactions either happen or don't, or a rules based model, which are atomic, there's no degrees of uncertainty. And so, we've we've essentially tried to make the world binary when we know it's that continues.

But it's a great question, and I appreciate you. So, Kara, what's next?

Yeah. A long one. So while LLMs have largely solved the challenge of making unstructured data accessible and useful for AI applications, many organizations are still struggling with effectively leveraging their structured data for AI agents. What are the key challenges you're seeing organizations face when trying to make their structured data whether it's in databases, ERP systems, or data warehouses truly useful for AI agents in an app less enterprise model? And how do we bridge the gap between having clean, organized data and actually making it actionable for autonomous AI systems?

Yeah. Yeah. I think there's two points here. One is, I think enterprises need to lead first by making sure that their data and data that they control is unified, organized, and accessible so that they can interact with it using agentic, platforms.

And then I think you also need to be pushing your your system providers that provide vertical software that holds some of that data to also make that data accessible, not just to there, but to any agentic experience that you as an enterprise wanna have looking at that data, and that you have the right to get access to that data, which is yours, and manipulate it. And I think, you know, if you look at what HubSpot does and look up Dharmesh Shah and some of the postings that he's done, they've done a fantastic job of making anybody that uses HubSpot and has a HubSpot data repository of, you know, contact management and inbound, sales activities, etcetera, accessible through ChatGPT so that you can have these incredibly powerful dialogues with ChatGPT about your enterprise data as it relates to the HubSpot data.

And that sets a bar, I think, of what everyone should try and do with their own data, and you have the ability to do that with your own enterprise data, and you should push your other software providers to just do that and point to that as a use case so that data is accessible. And then once it is, then that agentic experience can start taking that context of the data that you have, the data that's maybe in that HubSpot repository, the data that that's in SAP, and start having visibility and be able to answer questions, some of which when we're making decisions as an enterprise, it's important not just to look at a silo, but to understand the context of what's the salesperson saying against what we've seen from forecast, against what we're planning for a new product.

Those are three different repositories maybe that you need to understand and look at and and and have an integrated understanding of them.

And so let me add a very nerdy, postscript to that, which I totally agree with.

I think one way to think about this is IDs, foreign keys. And and I hate to sound like a database person, but fundamentally the question is how do we link data across silos and how do we create these confident connections between data silos and then also then the enterprise knowledge graph which links not a data view of the world, not a database view of the world, but rather an entity view of your business. So rather than thinking about data in HubSpot versus data in Salesforce versus data or I should say, rather than thinking about customer data in HubSpot and Salesforce and Oracle, what's the view of the customer independent of the dataset that it's sitting in, or in, or aggregated across the the database that it's sitting in?

And then the last thing I'll add and or underscore actually is a point that you made about accessibility to data. I think one thing you're gonna see in the next few years is data or or, legacy enterprise applications locking data down and not providing access to it. And I think that will be a really important flag to watch for as a customer, and, and be very clear. And and the analogy in in real life is when you have groups within your company, they treat their business process or their data as proprietary.

And that no. We're one company. We serve the customer. And if you have data providers that do the same, you should have a similar reaction.

I know we're running low on time, Cara.

Is there, are there any more or any Good question on the PhDs.

We could quickly answer, I think.

Alright.

Yeah. So how do you justify an a plus in everything with a PhD? Yeah. So I I think if you look at that question, the the way I I think about this is, yes, if someone is a PhD in economics, they are very deep in that one subject and can answer questions about that one subject, and they don't necessarily help you in the enterprise broadly because they might not understand a lot of the other domains because they've been so deep here. The fact is these LLMs not only have a PhD in economics, they have an MBA.

They have a PhD in computer science, a PhD in electrical engineering, a PhD in mechanical engineering. They understand design. They've read every, annual report of every company since the beginning of having them published and online. Like, it is not just a single vertical PhD. These things will be expert in a lot of domains, and then we'll learn the context of your business.

And that's where it's it's it's probably not a straight analogy to say they're like having a PhD, I think.

So I know there are a couple of other questions, Carrie. I leave it to your better judgment as to, where we go.

We just we'll have time for one more.

How about that? Sure. So from Chris, are we entering a world of stateless data or is it just the location locations of where managed data changing? Did I say that right?

I think that was.

I'll copy and paste it in the backstage chat so you can read it.

So I don't know. Yeah. Go ahead, Ant.

No. No. You're sorry.

Yeah. I mean, to me, I don't know that it's so much stateless data because the data has context and the data has relevance with that context. I think it's it's not that the data is necessarily stateless. It's that the data is available for interpretation with other data and with knowledge and expertise of both the corporate employees, but in concert with this very powerful AI, which has visibility across all of that data as well. And so, it it's able to, I think, answer questions and do things in a way that, you know, typically would require multiple apps and multiple smart people looking at generating reports or information from all of those different apps and then interpreting it, presenting it, and discussing it. Here, you you have a much faster ability to do that because you don't have the I you won't need to have those app silos to get access to and to get some distilling of all of that information and its relationship to each other with the help of AI.

And the essence of the question is also asking whether we should expect this to change with time. And I would argue that you should expect a lot of change with time. You're gonna be decommissioning systems, adding systems, changing data structures, and so if and I hope nobody interpreted my comment to mean that what we need to do is organize the data once and then we're done. Quite the opposite.

We should think of a dynamic system, which is in a way always watching your data, organizing it, but organizing it through the lens of the, to use my language, the entities that matter to your business. You know, who are our customers? Who are our suppliers? Where do we do business?

What locations are we in? You know, like, you can run down for your business what the entities are. Organize that way, and leave the systems, to morph and change as in fact, create that level of indirection. That'd be very, very powerful for for everyone.

Alright. Well, we have a ton more questions, but unfortunately, we're out of time. So we have a record of them, and we'll follow-up with you, directly.

But thank you so much for joining us today. We hope you found the session valuable. And just a quick quick reminder, today's webinar will be available on demand. You'll see this in your inbox shortly. So you can rewatch it or you can share it with your colleagues. And then before you go, I am gonna launch a survey in a moment. We'd love your feedback.

Should only take a minute or two and will really help us continue to improve and bring you the content you wanna see. So thanks again for spending part of your day with us, and we hope to see you at a future event soon. And, thank you, Anthony and Rich.

Thank you all.

Cheers.

‍