Matt Calkins of Appian, founder and chief executive officer, joins theCUBE Research hosts John Furrier and Dave Vellante to explore how process technology makes artificial intelligence, abbreviated AI, fit for mission-critical work. The discussion examines agentic infrastructure, governance, data integrity and application modernization, with examples from Appian World that show how workflow and reconciliation approaches integrate multiple models and human oversight across industries such as financial services.
Calkins asserts that organizations must pair probabilistic AI with a deterministic process layer to catch errors before they cause harm. They present Appian's DocCenter as a reconciliation framework that combines multiple models and human review to achieve near 99% accuracy. The hosts also discuss AI-driven legacy risks, the modernization trade-offs of convert versus rebuild and why trust and safety rather than development cost drive software value. Topics include Mixture of Experts, agentic infrastructure, process automation and model integration for enterprise AI in regulated sectors.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: Mixture of Experts Series. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: Mixture of Experts Series.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Register For theCUBE + NYSE Wired: Mixture of Experts Series
Please fill out the information below. You will recieve an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for theCUBE + NYSE Wired: Mixture of Experts Series.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: Mixture of Experts Series. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: Mixture of Experts Series.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: Mixture of Experts Series
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: Mixture of Experts Series. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Matt Calkins, Appian
Matt Calkins of Appian, founder and chief executive officer, joins theCUBE Research hosts John Furrier and Dave Vellante to explore how process technology makes artificial intelligence, abbreviated AI, fit for mission-critical work. The discussion examines agentic infrastructure, governance, data integrity and application modernization, with examples from Appian World that show how workflow and reconciliation approaches integrate multiple models and human oversight across industries such as financial services.
Calkins asserts that organizations must pair probabilistic AI with a deterministic process layer to catch errors before they cause harm. They present Appian's DocCenter as a reconciliation framework that combines multiple models and human review to achieve near 99% accuracy. The hosts also discuss AI-driven legacy risks, the modernization trade-offs of convert versus rebuild and why trust and safety rather than development cost drive software value. Topics include Mixture of Experts, agentic infrastructure, process automation and model integration for enterprise AI in regulated sectors.
>> Palo Alto Studio connecting Silicon Valley and Wall Street. I'm John Furrier, host of theCUBE, here with Dave Vellante, my co-host.
Welcome back here to theCUBE. I'm John Furrier, your host here at our NYSE Studios. Market's going crazy. Technology is driving the market. A lot of big trades happening here today, but also this is our mixture of experts here. It's part of our NYSE Wired program, connecting Silicon Valley and Wall Street. Matt Calkins is here, CEO of Appian. Was recently on theCUBE at their event, Appian World. Matt, thanks for coming on theCUBE here remotely into our NYSE studio.
Matt Calkins
>> John, it's a pleasure to join you.
John Furrier
>> So as we uncovered with your conference and your company we've been covering for quite some time, you guys are kind of in the perfect position as the world spins to agents because agents are essentially a proxy for the future workforce, future processes, and the people process technology change management equation of IT is now being applied to business. And guess what? Everyone cares about governance, data, workflows, and process. All those things now are talked about more than ever. This is a great time for you guys as well as the industry as this next level of the stack emerges in generative AI to now agentic infrastructure. Take us through what was the keys from your event and where are you guys positioned today?
Matt Calkins
>> Yeah. Well, the number one thing everybody's got to know about agents is they occasionally make mistakes. And that's the same with AI generally. AI is probabilistic technology, and once in a while it's going to get it wrong. And so organizations that have an intolerance for error and the most important work in the world are generally unable to make use of AI. We're trying to bridge the gap and create a framework in which AI mistakes will be remediated before they take effect. The purpose of Appian today is to create a layer of process technology that ensures that agents output will be correct and successful so that organizations can use agents for the most important work that they do. So we've been focusing on that and that's what our annual conference was all about.
John Furrier
>> And you guys have a lot of industries you guys play in that require accuracy. Financial services, you name it, those are the ones that have good data. They've done analytics. Now they're in the agentic world. They got to get the data right. And the other trend that's happening is people are recognizing that the models are separating from glue layers, semantic layers, harmonization layers, and infrastructure themselves, which is enabling all that speed. So the role of the models now play a big deal. You're starting to see that with Anthropic and some of the cybersecurity areas. But also the coding agents need to be more disciplined. So you have coding advancements, you have pressure on cyber, and also you have the requirements for not being wrong, to your point about hallucinations. Now, that sounds very easy to say that, but scope the problem statement there because there's a lot going on. Make that work. Explain that concept because that's essentially the enterprise requirements.
Matt Calkins
>> Yeah. Let me start by saying that you and I use AI every day and it's easy and there's no problem, and so it's easy to think that everyone must be getting value equally, but they're not. The largest organizations doing the most mission critical things, they don't dare use AI because they can't afford the mistakes. It could be illegal, it could be disastrous, it could hurt their reputation. But if you look at the high end, and this is true in study after study, there's one recently by the HBR. PwC did one earlier this year. It finds that the biggest organizations in the most strategic use cases are getting nothing from AI, which is the problem because that's where the greatest value was to be had. What we need to do is make AI fit to be used in the most important and valuable use cases. And the way you do that, you're just getting to it there, John, the way you do that is to make it safe, make it reliable. Today, AI is going to be wrong a couple times out of 100. Doesn't seem like much of a deal when I'm just doing research on my computer. But if you're allocating loans or adjudicating insurance cases or sending humans into orbit, you just can't afford that. So instead, we need a reliability framework and that's the challenge in front of us. And I think it's actually the number one question in business in this entire year is how do we apply AI to strategic work? It's not just something that Appian's working on. This is a cross economy challenge. How do we apply AI to strategic work? The kind of work where you can't make errors. The kind of work where you've simply got to get it right. Well, there are ways. We just need what's known as a deterministic layer to counteract the probabilistic uncertainty that comes from AI. And provided you couch AI in a framework where everything does what it's supposed to do every time, then we can catch errors before they happen and a process is an ideal medium for doing that. Some people don't realize this, but process or workflow, if you prefer, was invented a couple of decades ago with the express purpose of remediating errors in human behavior. Humans would make mistakes. And when a person makes a mistake in a process, it gets reconciled and checked and escalated and it doesn't end up being a mistake that you make at the end of the chain of actions. You make a mistake in the middle and you fix it before you get to the end. That's exactly what we want to do with AI. And so process is the ideal vehicle for keeping AI accurate.
John Furrier
>> I was joking the other day on a podcast around. If Peter Drucker was alive, we would have a whole nother set of management principles because what you're getting at is the nature of work, process management has been around for a while. A lot of theory, but now you've got this non-determinist or probabilistic environment, we think in a very probabilistic way as humans, but the humans ultimately got to do the work. You got to have determinism in the process to execute. Those are workflows, so it's pretty deterministic if you notice. So explain how you guys do that, because I think this is important because you can do both. You can map non-deterministic things into deterministic workflow, which are workflows. This seems to be the linkage.
Matt Calkins
>> Absolutely. Let me give you an example. Yeah, you can absolutely do it. For example, we've got this product we call DocCenter, which is basically reading all the incoming documents into major organizations that we've deployed around hundreds of companies around the world, including some of the largest. And it uses AI to read the contents of incoming documents. And as you and I both know, AI is not going to be perfect. So we give it not to one AI, but to multiple, and we reconcile the results against each other and we follow back and check when there's a discrepancy and we assign it to a human if we need to. And in the end, the accuracy gets up to 99%. It's incredible what we can get, provided you use the safety of process with the magic of AI. You needed to do all those reconciliations and checks and comparisons and escalations in order to be sure of the accuracy.
John Furrier
>> This is a huge point because that's up and down the stack because you now have things like sovereignty, right? That's going to need to require accuracy of where to put things, where to do AI and where that value is being created, which is basically money, revenue, not just privacy. Then you've got the whole safety piece, because right now guardrails have been kind of the code word for, or dog whistle, or whatever word you want to say, for being ethical. But now AI safety is a bigger picture. Talk about your view on AI safety because the recent executive order around cybersecurity and Mythos and Anthropic, we're seeing a lot of activity with the models coming in adding value. So you have the intelligence layer feeding into this kind of deterministic kind of reconciliation technology. Safety becomes key because if you've got accuracy data and you could tie that to some KPIs, you can get that safety. Explain your vision there. I think this is something that you guys are sitting on right now.
Matt Calkins
>> John, it's incredible how fast the world is moving. And you're right, guardrails meant one thing a couple of years ago and they mean another now. But fundamentally they mean keeping us safe and making sure that we don't delegate authority to an entity that will misuse that authority. And in this day, we want to delegate authority to AI. We want to use AI for as much power as it can give us and we are restricted by the distance of trust, by how much we're able to trust AI is going to get it right. And so we can use guardrails in order to deploy AI into environments that it wouldn't have been qualified to be deployed in otherwise. That's why providing a guardrail these days could be an incredibly important feature that makes or breaks the ability of AI to address a specific problem. If you can give it a guardrail, if you can give it a process, if you can make it safe and reliable, then AI can go where AI has never been able to go before. So that's really exciting. Secondly, you mentioned Mythos. Mythos has changed the game around legacy applications. We've been in this business of converting legacy applications into modern applications for a decade, but it's been a small business. It's been a small business when only the applications that you most urgently need to port are the ones that you spend the money to switch from old platforms to new platforms. But Mythos has changed this. First of all, AI makes it faster and cheaper to convert a legacy application into a modern application. But secondly, it makes it necessary because Mythos can also crack old applications. And if you've got an old application from a vendor that's still doing business and it has a modern mindset, then you're going to be getting an update this month or next month. And I recommend that you do the update, right? Because they're informed by Mythos and they're going to patch their product so that it's not vulnerable. And that'll work for every product you're using that still has an ongoing vendor presence. But for most applications, and you know 70% of the Fortune 500 systems are more than 20 years old, for most of those systems, they are out of date and there will be no patch, and instead they are vulnerable to being cracked by Mythos or the latest from OpenAI or many other AI applications coming soon. Those applications which used to be an inert annoyance in the back closet of the enterprise are now a security limitation and a vulnerability. Something's got to be done. So organizations are looking for inexpensive ways to move their old applications to a new platform, a vital platform that's secure and updated all the time. However, the cost of making that move is more than just the cost of the AI or the code or the services. The hidden cost is the mistakes that you might make. Those old applications have lain inert for all this time because people didn't dare touch them and at least they worked. Now, you want to move it to a new platform and you run a risk. Maybe the new one won't work. Maybe you'll make some kind of mistranslation and it won't work exactly the way the old one did. So you've really got to get it right. And I caution people when they're interested in evacuating from their old applications, I say it's true, it's bad to be on old applications, but it'll be even worse if you invent a new one that doesn't do the job. So be very careful.
John Furrier
>> Yeah. And that really talks to the whole SaaS apocalypse kind of narrative which people are poo-pooing. I was poo-pooing it. I mean, my narrative around the SaaSpocalypse was bad software always dies. So that's kind of like a known thing. But what you're getting at is really more of the modernization wave. Okay, so AI can help with modernization. So take us through that piece of it and what you guys are doing, because this is what people are struggling with. Do I build or buy or retrofit or pave over that? Take us through the options. When is it better to do something new and when is it better to just modernize the existing app, whether you abstract away or harden it, whatever it's called. What's the choices for customers?
Matt Calkins
>> First of all, you mentioned the SaaSpocalypse. Let me address that topic and then move on to whether it's better to convert or rebuild or harden or any of those options. There has been a lot of concern lately that if AI allows code to be very cheap, if AI can make code at an extremely low cost, does that mean that everyone can now be their own SaaS competitor, that you don't need to buy a prominent SaaS product, like say Salesforce, you could just write your own? And there's been a lot of fear based on that and then worries that the whole SaaS industry will collapse. And I find that a little misguided because we've already been through a free software scare. It happened 20 years ago, it was called open source. Software was free. Software is free. Most of the world's software that runs today in big organizations is open source. And despite the fact that open source now makes a great deal of software free, the software industry has grown by five times over the course of those 20 years. So there's been plenty of growth. Software's still a great business. And the fact that code becomes cheap isn't the determining factor as to whether software is valuable. Mistakes are the real cost, not the code. You've got to get it right and the software business is the trust business. So first thing I want to say is I'm a skeptic when it comes to the SaaSpocalypse. Second of all, per your question, if you've got a lot of old applications in your enterprise, should you be updating them or hardening them? Well, ideally you update them but you don't want to go faster than you can ensure correct translation. So while you are ensuring correctness and while you're working with a reliable partner who's going to translate accurately all those old applications, there's going to be a while where you need to take care of those old applications and mind the vulnerabilities. And I suggest that you just not expose them. To the greatest degree, just don't allow them to be exposed to external inspection. Internal, you've cut your risk by a great deal. And so the truth is if you had somebody in your organization who wanted to hack it, they have other ways as well. They could log on legitimately and then do something wrong there. So I primarily would want to protect against external actors. And so the first thing I would do is just reduce access to these applications to the degree that you can.
John Furrier
>> I really appreciate, Matt, your comments on SaaSpocalypse. Totally agree. But the point about cost, the cost isn't about the code, it's about the mistakes and the consequences that can be quantified. A hack, misconfiguration, lack of attention to code.
Matt Calkins
>> Yeah, that's exactly right, and people don't understand that. They think the software industry is about making software the kind of way the hamburger industry is about making hamburger, and what they don't understand is we're in the business of accurate decisions. Software is just a list of instructions, but the output, the true output of our industry is good decisions, good actions, accurate responses to stimuli. That's what software is about. And if somebody wants to use AI to write their own CRM system and then maintain it themselves and be themselves responsible for all its mistakes, and have nobody to turn to when it's time to update it or protect it against the latest exploit, that's a very lonely road and I don't think many organizations are going to choose to take it.
John Furrier
>> It's a lonely and bad road actually. There's a lot of fire on that road. I want to get your reaction to some of the AI native capabilities that's emerging. In every way you see a shift. We've crossed that threshold here. We've seen a lot more capabilities from the models and kind of the configurations. You guys are modernizing apps so you know this space. I had an Apple executive on here, former Apple executive that worked during the iPhone day, pre-iPhone, and he answered the SaaSpocalypse question by saying that the products have to get better and that's the key, products where the consequences are clear, it's safe, et cetera. And he used the iPhone example and the iPod. The iPod was the number one selling product for Apple and the iPhone essentially cannibalized that, but the iPhone was an iPod, it was a computer, it was a phone, and an App Store. So the iPhone actually was better product. And so he brought that up in context to what AI is enabling in the modernization. What is your reaction to that? Do you see a similar thing happening where it's not just Salesforce been replicated? Because I could do that. I could also do, I could create Reddit, but I don't have millions of users. So replication is one thing, but actual product capability is another. This seems to be an AI native kind of thing going on. What's your reaction to this product change in the modernization product?
Matt Calkins
>> I think your guy has a good point. I think products will have to get better because now that AI is available, the ability to deliver better quality through software is in the hands of every software company, and those that succeed are going to outclass and outperform those that don't. So there's an arms race. That's how I'd characterize it amongst software companies. And those that can leverage AI to be far more powerful and valuable and low cost all at the same time have substantial rewards coming to them, and those that aren't dynamic or don't move fast enough or have a very conservative technology profile and are just sort of lurking on their installed base, I don't think this is going to turn out as well for them.
John Furrier
>> Well, I got you here, Matt. I'd love to get your thoughts on something we've been riffing on in theCUBE. We've been kind of talking about product market fit for a generation. "Oh, we got product market fit. That's a good business opportunity." We're kind of in the systems revolution right now where you got AI factories, you got applications talking to agents, you got physical AI. The system seems to be the new architectural construct. So maybe product market system fit. So the product system fit is a key thing. Do you agree with that? Do you see that happening? Because a lot of things have dependencies now. I mean, you know the process gain, if you're going to go from non-deterministic to deterministic that you now have a systematic feature.
Matt Calkins
>> Right. Okay. I think the phrase, "Product market fit," still makes a ton of sense and people should keep using it because when you start a new product or when you build something, you've got to be conscious of whether it matches the market, whether it's used, and whether it's creating value. So I still love it for that reason. But you're right that the fit itself is more complicated than it used to be. It's less scripted, it's what's rote, it's less predictable. We're going to see elements inside the enterprise connecting with each other in ways that even their authors did not intend. We're going to see instead of static silos with vertical connections and the same call made a million times, we're going to see lateral unexpected calls. We're going to see agents calling data sources that they've never called before. Or one application collaborating with another application spontaneously. The AI ecosystem is a dynamic and unscripted ecosystem. It's like the primordial soup instead of a fixed engine. And so any application that wants to take up space in the future enterprise should be ready to improvise. You should be ready to send API calls, receive API calls, MCP I'm talking. Be ready to collaborate in unusual ways. And above all, the golden rule, be useful.
John Furrier
>> Yeah. Great stuff.
Matt Calkins
>> And convey a value to those that you're working with.
John Furrier
>> Matt, thank you so much. Love that . and by the way, the domain specific product market fit is where the action is. That's the point of application, but it's a bigger thing. The pie is big. It's horizontal. I appreciate it and appreciate that comment. Appian and you guys are doing great work. Again, congratulations on your success. The world spun to your doorstep.
Matt Calkins
>> Thank you, John.
John Furrier
>> All right. I'm John Furrier. It's our Mixture of Expert series, breaking down the tech that's making it happen. As the AI wave continues, you're seeing a lot more intelligence being injected into businesses, and that's entire businesses. That's a domain line of business, that's a particular job, and you're seeing the system scaling it out. Obviously you need the data. We're doing our part to bring the data to you. Thanks for watching.
>> Palo Alto Studio connecting Silicon Valley and Wall Street. I'm John Furrier, host of theCUBE, here with Dave Vellante, my co-host.
Welcome back here to theCUBE. I'm John Furrier, your host here at our NYSE Studios. Market's going crazy. Technology is driving the market. A lot of big trades happening here today, but also this is our mixture of experts here. It's part of our NYSE Wired program, connecting Silicon Valley and Wall Street. Matt Calkins is here, CEO of Appian. Was recently on theCUBE at their event, Appian World. Matt, thanks for coming on theCUBE here remotely into our NYSE studio.
Matt Calkins
>> John, it's a pleasure to join you.
John Furrier
>> So as we uncovered with your conference and your company we've been covering for quite some time, you guys are kind of in the perfect position as the world spins to agents because agents are essentially a proxy for the future workforce, future processes, and the people process technology change management equation of IT is now being applied to business. And guess what? Everyone cares about governance, data, workflows, and process. All those things now are talked about more than ever. This is a great time for you guys as well as the industry as this next level of the stack emerges in generative AI to now agentic infrastructure. Take us through what was the keys from your event and where are you guys positioned today?
Matt Calkins
>> Yeah. Well, the number one thing everybody's got to know about agents is they occasionally make mistakes. And that's the same with AI generally. AI is probabilistic technology, and once in a while it's going to get it wrong. And so organizations that have an intolerance for error and the most important work in the world are generally unable to make use of AI. We're trying to bridge the gap and create a framework in which AI mistakes will be remediated before they take effect. The purpose of Appian today is to create a layer of process technology that ensures that agents output will be correct and successful so that organizations can use agents for the most important work that they do. So we've been focusing on that and that's what our annual conference was all about.
John Furrier
>> And you guys have a lot of industries you guys play in that require accuracy. Financial services, you name it, those are the ones that have good data. They've done analytics. Now they're in the agentic world. They got to get the data right. And the other trend that's happening is people are recognizing that the models are separating from glue layers, semantic layers, harmonization layers, and infrastructure themselves, which is enabling all that speed. So the role of the models now play a big deal. You're starting to see that with Anthropic and some of the cybersecurity areas. But also the coding agents need to be more disciplined. So you have coding advancements, you have pressure on cyber, and also you have the requirements for not being wrong, to your point about hallucinations. Now, that sounds very easy to say that, but scope the problem statement there because there's a lot going on. Make that work. Explain that concept because that's essentially the enterprise requirements.
Matt Calkins
>> Yeah. Let me start by saying that you and I use AI every day and it's easy and there's no problem, and so it's easy to think that everyone must be getting value equally, but they're not. The largest organizations doing the most mission critical things, they don't dare use AI because they can't afford the mistakes. It could be illegal, it could be disastrous, it could hurt their reputation. But if you look at the high end, and this is true in study after study, there's one recently by the HBR. PwC did one earlier this year. It finds that the biggest organizations in the most strategic use cases are getting nothing from AI, which is the problem because that's where the greatest value was to be had. What we need to do is make AI fit to be used in the most important and valuable use cases. And the way you do that, you're just getting to it there, John, the way you do that is to make it safe, make it reliable. Today, AI is going to be wrong a couple times out of 100. Doesn't seem like much of a deal when I'm just doing research on my computer. But if you're allocating loans or adjudicating insurance cases or sending humans into orbit, you just can't afford that. So instead, we need a reliability framework and that's the challenge in front of us. And I think it's actually the number one question in business in this entire year is how do we apply AI to strategic work? It's not just something that Appian's working on. This is a cross economy challenge. How do we apply AI to strategic work? The kind of work where you can't make errors. The kind of work where you've simply got to get it right. Well, there are ways. We just need what's known as a deterministic layer to counteract the probabilistic uncertainty that comes from AI. And provided you couch AI in a framework where everything does what it's supposed to do every time, then we can catch errors before they happen and a process is an ideal medium for doing that. Some people don't realize this, but process or workflow, if you prefer, was invented a couple of decades ago with the express purpose of remediating errors in human behavior. Humans would make mistakes. And when a person makes a mistake in a process, it gets reconciled and checked and escalated and it doesn't end up being a mistake that you make at the end of the chain of actions. You make a mistake in the middle and you fix it before you get to the end. That's exactly what we want to do with AI. And so process is the ideal vehicle for keeping AI accurate.
John Furrier
>> I was joking the other day on a podcast around. If Peter Drucker was alive, we would have a whole nother set of management principles because what you're getting at is the nature of work, process management has been around for a while. A lot of theory, but now you've got this non-determinist or probabilistic environment, we think in a very probabilistic way as humans, but the humans ultimately got to do the work. You got to have determinism in the process to execute. Those are workflows, so it's pretty deterministic if you notice. So explain how you guys do that, because I think this is important because you can do both. You can map non-deterministic things into deterministic workflow, which are workflows. This seems to be the linkage.
Matt Calkins
>> Absolutely. Let me give you an example. Yeah, you can absolutely do it. For example, we've got this product we call DocCenter, which is basically reading all the incoming documents into major organizations that we've deployed around hundreds of companies around the world, including some of the largest. And it uses AI to read the contents of incoming documents. And as you and I both know, AI is not going to be perfect. So we give it not to one AI, but to multiple, and we reconcile the results against each other and we follow back and check when there's a discrepancy and we assign it to a human if we need to. And in the end, the accuracy gets up to 99%. It's incredible what we can get, provided you use the safety of process with the magic of AI. You needed to do all those reconciliations and checks and comparisons and escalations in order to be sure of the accuracy.
John Furrier
>> This is a huge point because that's up and down the stack because you now have things like sovereignty, right? That's going to need to require accuracy of where to put things, where to do AI and where that value is being created, which is basically money, revenue, not just privacy. Then you've got the whole safety piece, because right now guardrails have been kind of the code word for, or dog whistle, or whatever word you want to say, for being ethical. But now AI safety is a bigger picture. Talk about your view on AI safety because the recent executive order around cybersecurity and Mythos and Anthropic, we're seeing a lot of activity with the models coming in adding value. So you have the intelligence layer feeding into this kind of deterministic kind of reconciliation technology. Safety becomes key because if you've got accuracy data and you could tie that to some KPIs, you can get that safety. Explain your vision there. I think this is something that you guys are sitting on right now.
Matt Calkins
>> John, it's incredible how fast the world is moving. And you're right, guardrails meant one thing a couple of years ago and they mean another now. But fundamentally they mean keeping us safe and making sure that we don't delegate authority to an entity that will misuse that authority. And in this day, we want to delegate authority to AI. We want to use AI for as much power as it can give us and we are restricted by the distance of trust, by how much we're able to trust AI is going to get it right. And so we can use guardrails in order to deploy AI into environments that it wouldn't have been qualified to be deployed in otherwise. That's why providing a guardrail these days could be an incredibly important feature that makes or breaks the ability of AI to address a specific problem. If you can give it a guardrail, if you can give it a process, if you can make it safe and reliable, then AI can go where AI has never been able to go before. So that's really exciting. Secondly, you mentioned Mythos. Mythos has changed the game around legacy applications. We've been in this business of converting legacy applications into modern applications for a decade, but it's been a small business. It's been a small business when only the applications that you most urgently need to port are the ones that you spend the money to switch from old platforms to new platforms. But Mythos has changed this. First of all, AI makes it faster and cheaper to convert a legacy application into a modern application. But secondly, it makes it necessary because Mythos can also crack old applications. And if you've got an old application from a vendor that's still doing business and it has a modern mindset, then you're going to be getting an update this month or next month. And I recommend that you do the update, right? Because they're informed by Mythos and they're going to patch their product so that it's not vulnerable. And that'll work for every product you're using that still has an ongoing vendor presence. But for most applications, and you know 70% of the Fortune 500 systems are more than 20 years old, for most of those systems, they are out of date and there will be no patch, and instead they are vulnerable to being cracked by Mythos or the latest from OpenAI or many other AI applications coming soon. Those applications which used to be an inert annoyance in the back closet of the enterprise are now a security limitation and a vulnerability. Something's got to be done. So organizations are looking for inexpensive ways to move their old applications to a new platform, a vital platform that's secure and updated all the time. However, the cost of making that move is more than just the cost of the AI or the code or the services. The hidden cost is the mistakes that you might make. Those old applications have lain inert for all this time because people didn't dare touch them and at least they worked. Now, you want to move it to a new platform and you run a risk. Maybe the new one won't work. Maybe you'll make some kind of mistranslation and it won't work exactly the way the old one did. So you've really got to get it right. And I caution people when they're interested in evacuating from their old applications, I say it's true, it's bad to be on old applications, but it'll be even worse if you invent a new one that doesn't do the job. So be very careful.
John Furrier
>> Yeah. And that really talks to the whole SaaS apocalypse kind of narrative which people are poo-pooing. I was poo-pooing it. I mean, my narrative around the SaaSpocalypse was bad software always dies. So that's kind of like a known thing. But what you're getting at is really more of the modernization wave. Okay, so AI can help with modernization. So take us through that piece of it and what you guys are doing, because this is what people are struggling with. Do I build or buy or retrofit or pave over that? Take us through the options. When is it better to do something new and when is it better to just modernize the existing app, whether you abstract away or harden it, whatever it's called. What's the choices for customers?
Matt Calkins
>> First of all, you mentioned the SaaSpocalypse. Let me address that topic and then move on to whether it's better to convert or rebuild or harden or any of those options. There has been a lot of concern lately that if AI allows code to be very cheap, if AI can make code at an extremely low cost, does that mean that everyone can now be their own SaaS competitor, that you don't need to buy a prominent SaaS product, like say Salesforce, you could just write your own? And there's been a lot of fear based on that and then worries that the whole SaaS industry will collapse. And I find that a little misguided because we've already been through a free software scare. It happened 20 years ago, it was called open source. Software was free. Software is free. Most of the world's software that runs today in big organizations is open source. And despite the fact that open source now makes a great deal of software free, the software industry has grown by five times over the course of those 20 years. So there's been plenty of growth. Software's still a great business. And the fact that code becomes cheap isn't the determining factor as to whether software is valuable. Mistakes are the real cost, not the code. You've got to get it right and the software business is the trust business. So first thing I want to say is I'm a skeptic when it comes to the SaaSpocalypse. Second of all, per your question, if you've got a lot of old applications in your enterprise, should you be updating them or hardening them? Well, ideally you update them but you don't want to go faster than you can ensure correct translation. So while you are ensuring correctness and while you're working with a reliable partner who's going to translate accurately all those old applications, there's going to be a while where you need to take care of those old applications and mind the vulnerabilities. And I suggest that you just not expose them. To the greatest degree, just don't allow them to be exposed to external inspection. Internal, you've cut your risk by a great deal. And so the truth is if you had somebody in your organization who wanted to hack it, they have other ways as well. They could log on legitimately and then do something wrong there. So I primarily would want to protect against external actors. And so the first thing I would do is just reduce access to these applications to the degree that you can.
John Furrier
>> I really appreciate, Matt, your comments on SaaSpocalypse. Totally agree. But the point about cost, the cost isn't about the code, it's about the mistakes and the consequences that can be quantified. A hack, misconfiguration, lack of attention to code.
Matt Calkins
>> Yeah, that's exactly right, and people don't understand that. They think the software industry is about making software the kind of way the hamburger industry is about making hamburger, and what they don't understand is we're in the business of accurate decisions. Software is just a list of instructions, but the output, the true output of our industry is good decisions, good actions, accurate responses to stimuli. That's what software is about. And if somebody wants to use AI to write their own CRM system and then maintain it themselves and be themselves responsible for all its mistakes, and have nobody to turn to when it's time to update it or protect it against the latest exploit, that's a very lonely road and I don't think many organizations are going to choose to take it.
John Furrier
>> It's a lonely and bad road actually. There's a lot of fire on that road. I want to get your reaction to some of the AI native capabilities that's emerging. In every way you see a shift. We've crossed that threshold here. We've seen a lot more capabilities from the models and kind of the configurations. You guys are modernizing apps so you know this space. I had an Apple executive on here, former Apple executive that worked during the iPhone day, pre-iPhone, and he answered the SaaSpocalypse question by saying that the products have to get better and that's the key, products where the consequences are clear, it's safe, et cetera. And he used the iPhone example and the iPod. The iPod was the number one selling product for Apple and the iPhone essentially cannibalized that, but the iPhone was an iPod, it was a computer, it was a phone, and an App Store. So the iPhone actually was better product. And so he brought that up in context to what AI is enabling in the modernization. What is your reaction to that? Do you see a similar thing happening where it's not just Salesforce been replicated? Because I could do that. I could also do, I could create Reddit, but I don't have millions of users. So replication is one thing, but actual product capability is another. This seems to be an AI native kind of thing going on. What's your reaction to this product change in the modernization product?
Matt Calkins
>> I think your guy has a good point. I think products will have to get better because now that AI is available, the ability to deliver better quality through software is in the hands of every software company, and those that succeed are going to outclass and outperform those that don't. So there's an arms race. That's how I'd characterize it amongst software companies. And those that can leverage AI to be far more powerful and valuable and low cost all at the same time have substantial rewards coming to them, and those that aren't dynamic or don't move fast enough or have a very conservative technology profile and are just sort of lurking on their installed base, I don't think this is going to turn out as well for them.
John Furrier
>> Well, I got you here, Matt. I'd love to get your thoughts on something we've been riffing on in theCUBE. We've been kind of talking about product market fit for a generation. "Oh, we got product market fit. That's a good business opportunity." We're kind of in the systems revolution right now where you got AI factories, you got applications talking to agents, you got physical AI. The system seems to be the new architectural construct. So maybe product market system fit. So the product system fit is a key thing. Do you agree with that? Do you see that happening? Because a lot of things have dependencies now. I mean, you know the process gain, if you're going to go from non-deterministic to deterministic that you now have a systematic feature.
Matt Calkins
>> Right. Okay. I think the phrase, "Product market fit," still makes a ton of sense and people should keep using it because when you start a new product or when you build something, you've got to be conscious of whether it matches the market, whether it's used, and whether it's creating value. So I still love it for that reason. But you're right that the fit itself is more complicated than it used to be. It's less scripted, it's what's rote, it's less predictable. We're going to see elements inside the enterprise connecting with each other in ways that even their authors did not intend. We're going to see instead of static silos with vertical connections and the same call made a million times, we're going to see lateral unexpected calls. We're going to see agents calling data sources that they've never called before. Or one application collaborating with another application spontaneously. The AI ecosystem is a dynamic and unscripted ecosystem. It's like the primordial soup instead of a fixed engine. And so any application that wants to take up space in the future enterprise should be ready to improvise. You should be ready to send API calls, receive API calls, MCP I'm talking. Be ready to collaborate in unusual ways. And above all, the golden rule, be useful.
John Furrier
>> Yeah. Great stuff.
Matt Calkins
>> And convey a value to those that you're working with.
John Furrier
>> Matt, thank you so much. Love that . and by the way, the domain specific product market fit is where the action is. That's the point of application, but it's a bigger thing. The pie is big. It's horizontal. I appreciate it and appreciate that comment. Appian and you guys are doing great work. Again, congratulations on your success. The world spun to your doorstep.
Matt Calkins
>> Thank you, John.
John Furrier
>> All right. I'm John Furrier. It's our Mixture of Expert series, breaking down the tech that's making it happen. As the AI wave continues, you're seeing a lot more intelligence being injected into businesses, and that's entire businesses. That's a domain line of business, that's a particular job, and you're seeing the system scaling it out. Obviously you need the data. We're doing our part to bring the data to you. Thanks for watching.
John Furrier