theCUBE + NYSE Wired: C-Suite Business Transformation | Samar Abbas, Temporal

Clips
More from theCUBE + NYSE Wired: C-Suite Business Transformation

Samar Abbas

Co-Founder & CEO

Temporal

play_circle_outline Temporal: open-source durable execution platform ensuring code never loses state.

play_circle_outline Enabling AI agents to take long-running actions reliably and securely.

play_circle_outline Auditable Agent Governance: Guardrails, Human Approvals, and Telemetry for Safe Execution Histories

play_circle_outline Project Nexus: industry-standard asynchronous communication between agents and services.

play_circle_outline APIs vs Durable Stateful Agents: Reliable Microservices Guaranteeing Workflow Completion for Money Transfers

play_circle_outline AI Resilience with Multi-Region, Multi-Cloud Availability for Revenue Protection and Operational Continuity

Info

Samar Abbas, Temporal

Samar Abbas

Co-Founder & CEO Temporal

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street. I'm John Furrier, co-hosting with Dave Vellante, my co-host. Hello, I'm John Furrier, your host of theCUBE. We are here in theCUBE's NYSE studio. Of course, we have our Palo Alto studio on the West Coast connecting Silicon Valley to Wall Street. This is our Transformation Edge series. We talk to the leaders who are building technologies and are part of the massive transformation that's being enabled by AI and AI infrastructure, but more importantly, the software that's being generated to run the AI will change our lives for the better, to be safe, reliable. That's the focus today is to talk more about not just the productivity, but the safety. Samar Abbas is here. He's the co-founder and CEO of Temporal Technologies. Temporal just recently announced some big news, big funding. Thanks for coming on. It's great to see you.

Samar Abbas

>> Yeah, thanks a lot, John, for having me. Excited to be here.

John Furrier

>> We were just talking a little history lesson down memory lane almost a decade and a half ago about Amazon Web Services and the rise of the cloud. Now we have the rise of AI infrastructure, which is helping everybody. The rising tide lifts all boats. Yeah, the hyperscalers, AWS, Azure, all winning more revenue there. You've got the rise of the neo clouds, now called AI clouds. They don't like the neo cloud name, some called Neo Cloud Labs. Now you have the enterprises eyeing this next wave where they saw the value of cloud. Agents are coming right along fast in the enterprise, so they're going to leaning into deployment strategies for CapEx. We're seeing that clearly. I mean, they're not going to spend billions of dollars like a neo cloud, but they will connect to and put on-premises AI factories that'll spawn new applications, create new user expectations that everybody wants the AI. You guys are in the middle of it.

John Furrier

>> Yep.

John Furrier

>> Your valuation is really a significant, I think, template and a sign of what's coming. Explain the news and why that happened.

Samar Abbas

>> Yeah. So before I go into the news, let me give you a high-level intro. About what we are, because then it will explain things. So Temporal is an open-source platform which ensures durable execution of your code. What that means is today, if you woke up and published a Snap story or ordered a Taco Bell or used OpenAI's Codex agent, all of that was powered by Temporal. And so this is where what we've been trying to do is, especially as software, when it runs for longer periods of time, it interacts with the outside world. The outside world has lots of failure modes there. And during the execution of that software, people can lose their— software has always been losing their place, essentially, when those failures happen. So now what we do with Temporal is if that software is built on top of us, they will never lose their place.

John Furrier

>> Explain what people are buying, because I think this is a good point. So we'll get to the news. I think the momentum and the value proposition justify the big numbers. What is the reason? What's— what is actually being powered? Are you powering the app, are you powering the infrastructure? What specifically are you guys enabling and accelerating for your customers?

Samar Abbas

>> So, especially if you look at the last 18 months, what has fundamentally happened AI has stopped talking and started doing stuff. And this is exactly where Temporal is super differentiated, is as these models get smarter and smarter, when we talk to an organization, you always hear that when an agent runs longer, takes more action, does more stuff, it creates more value. It's super clear to those enterprises. And guess what? Temporal is the layer which is making sure that these agents can do that reliably and most importantly, securely take those actions to make sure they don't cause any harm. And this is exactly the reason why we have seen an amazing growth of those actions on top of our platform. Even in the month of August alone, we have processed over 1.9 trillion actions, which is a 350% increase compared to a year ago.

John Furrier

>> So you're talking about the talking, meaning the chatbots. We know when we first used it, hey, tell me about this, write the story for me, give me some marketing material, search for things. That's search. Yeah, you're giving it more work, okay, go figure this out. Complex problem. Organize my email, call my friend, book a flight. More multi-step activities that require touching more resource.

Samar Abbas

>> 100%. A very good example is, for example, coding agents. As these models are getting smarter, what we are seeing is these coding agents are now taking on more and more complex tasks, which run for hours. And during those hours, all sorts of failures can happen. Machines can restart, or downstream tools can fail or time out, or even models can return an error. These models have a pretty high error rate. So how do you remember all of that state? This is exactly where Temporal comes in to manage all of that state for your agents so they don't lose track of it and don't redo your stuff. And it actually helps us with saving tokens also.

John Furrier

>> You know, I was talking to an engineer who works at Stripe recently and this was not on camera. It was like an off-the-record conversation, so I won't say his name. He said these agents are great. People are deploying them like crazy. They're seeing value, as you pointed out. He says, but there's a big problem. I go, what? He goes, well, they're basically calling the main models and paying for the tokens and they're running it in a production environment. That's like us building a payment gateway and running it in production.

John Furrier

>> Yeah.

John Furrier

>> The sandbox, the test dev environment. So he's like, it's all broken. It's exciting. People are deploying and are leaning into it. They're enthusiastic, but they have to use production-like resources and costs to do the test dev. That kind of brings up the IT transformation kind of connection, which is, hey, I want to just build this app and I want to automate a lot of these things. I want to create that value, but I got to observe it. I got to have telemetry. I need to do all these, tech things under the covers. It sounds like that's what you guys are doing. Explain what that means for you guys. Because that's the problem that everyone's trying to solve right now.

Samar Abbas

>> And I hope, by the way, that everyone starts solving because this is becoming the problem. If the industry does not get it right, we might screw up the entire AI platform shift that we are seeing. And by the way, this is not a new problem. This is like when the cloud platform shift happened, we saw the exact same excitement with the cloud platform shift. But then industry came around and to your point, made all of the infrastructure to make sure you can run an application in a cloud with all of the necessary guardrails and protections which won't cause any harm. And I think the AI industry, as AI is now moving from just providing you answers to actually taking real actions on behalf of a user, they absolutely need to start putting in all of those guardrails.

John Furrier

>> The press loves to talk about, oh, it's off the rails, agents gone rogue. I've heard that we need a kill switch. We don't need a kill switch. So there's a lot of negativity around some of the failures in agents, but there's a lot of positive examples happening. What's your reaction to that? Because you're seeing a bifurcation of people going, well, I am scared or doom and gloom. And then you have another half saying this is the best thing we've ever seen. It's life-changing experiences. Societal benefits are off the charts. What's your reaction when you see this kind of polarization?

Samar Abbas

>> So let me first address the negativity comment that you mentioned. I think a lot of fear which is coming in is— I completely understand where it's coming from. But if you double-click on especially, let's say, the last few months, all of the incidents which have been coming out in the news, for instance, the Hugging Face hack which happened during an OpenAI testing, which is all over the place now.

Samar Abbas

>> Yeah, everyone uses that one. Yeah.

Samar Abbas

>> So if you kind of dig into it, one thing becomes absolutely clear. There is a pattern emerging that when these AI agents take actions, one of the things which we don't have in place is auditability of what that agent is doing. And in my opinion, that's one of the biggest gaps which is out there right now. So this is exactly the place where Temporal is differentiated, is when you build an agent using Temporal as an underlying platform, each and every step or action an agent takes, it's durably recorded in what we call an execution history of that agent. And nothing— there is no way anyone can come in and interfere with it because that's how the platform works natively. Basically, nothing is off the books. Everything is recorded durably. And every time that agent is taking an action, before it takes an action, it's durably recorded there. And then you have an audit log which can be used by any auditors to actually chase what that agent was doing when it goes rogue. And then the second capability Temporal adds is all of those actions are recorded before they happen, which means it gives an opportunity to that organization to put on their own guardrails. Which could be a human kind of approving that action or a business rule executed somewhere which might reject it. And so that's how you have this chain of who did what. And this is the role that's like rules of engagement too.

John Furrier

>> You have everything laid out. What's interesting about when I hear you talking, I get excited because there's a lot of parallels and it's kind of a history lesson to be learned here. If you look at AWS, early days of cloud where you have history, a lot of the higher-level services that Amazon ended up coming out with weren't even invented yet. the basic building blocks of AWS were a dream scenario for a startup. Didn't have any cash, put their credit card down. Everyone knows that story. Shadow IT emerged. You could actually get stuff done with those simple things. But then higher-level services came out. When you're talking about agents, I could see new things emerge like agents for agents. I talked with Jensen and NVIDIA team all the time about how they're using simulations and synthetic data, you can almost predict things. So you start to see as it gets out there and more adopted, what's your vision on that? What do you see that's possible to make things more reliable, more predictable? I think that's—

Samar Abbas

>> yeah, that's exactly the direction that we are headed. We have some customers who are already running over 5,000 namespaces on top of Temporal Cloud. Namespace is our unit of isolation where people run their own workloads. So what that means is we are already starting to become the API boundaries on how those organizations talk to each other. We have a project called Project Nexus, which is trying to create an industry-wide standard, how asynchronous communication happens in those larger enterprises. So as these agents get more and more long-lived, This is exactly where our investment on Project Nexus will be super differentiated because it becomes the backbone for how agents talk to each other.

John Furrier

>> I think you're the first interview guest I've had on theCUBE that used the word namespace. So congratulations, you get the award for namespace. Namespace is, for the folks watching, it's a very simple definition. Think of it like an address book, like your address at home, it knows things, knows where things are. And that's important. I want to double click on that because I think it's a nuanced tech point, but it's the science behind the computer science where if you understand the domains and the addressing system, discovery, reliability, paths, efficiency, explain the importance of having a good namespace system or addressing system and why is that super important?

Samar Abbas

>> So think of namespace as an envelope where we give you an envelope and use that envelope to run your specific application within that envelope. And it gives you complete isolation from anything else. And it's especially important because Temporal Cloud is a multi-tenant cloud service. What that means is today we are completely hiding all of the infrastructure from you. You come in and create an account on Temporal Cloud, and then when you want to build an application, the first thing you build is a namespace, which gives you a box where you can now start running your workflows and activities inside. The reason it's important is how do you guarantee one namespace does not interfere with the workload or performance characteristics of another namespace? And that's where this envelope becomes really, really interesting because it gives you complete isolation from a performance standpoint. And you like—

John Furrier

>> you can get things delivered. It gets to the endpoint. Yeah. I want to talk about your history. You and your co-founder have— I think it's what I love about this era. It's not talked about a lot. In the media is the history of the computer industry and specifically open source has a saying standing on the shoulders of giants, meaning because it's open source code, you can build on top of code. Great. We love open source, but you start to see the giants' shoulders like Uber, like AWS. You work at Uber. A lot of this stuff has been solved for these early first movers that had to build their own system. Let's take Uber, for example. Yeah. they literally had to build the app that we all love and use. Lyft did the same thing. it's basically you log in to the cloud, multi-tenant cloud. You actually have to hail a driver who doesn't even work for the company. Yeah, you got to track that person with GPS. You have an account, a database for all your stuff. You got to bring all these things together in real time. Now that's a state-of-the-art problem. Go back when they originally founded it— that was really hard. But that was also a marker. Trail marker for what would be coming to the mainstream. So in a way, a lot of the innovation that you're doing and others have been there and the market just hasn't evolved yet. And now it has. Explain the importance of this. You didn't just wake up one day and say, hey, I'm going to start Temporal.

John Furrier

>> Yeah.

John Furrier

>> Talk about the importance of the trajectory of the industry and why those other advancements were and what it means for the AI world. What happens next?

Samar Abbas

>> 100%. So very good question. So by the way, there are founders who sometimes wake up one day, they have an idea and then they build a company around it. Both me and my co-founder Maxim, we've been iterating over this problem for at least 20+ years now. It started actually at Amazon AWS. So where we saw the transformation of Amazon from a monolith into a service-oriented architecture. And this is where we built Simple Workflow as an underlying application platform to help with those massive service-oriented architectures to become the reliability layer for building those applications. Then I continued that at Azure where I was part of a team at Azure called Azure Service Bus, which owns the messaging stack. And I very quickly saw similarities on how engineers are using messaging as a core underlying primitive to build essentially reliability or resiliency into their apps. So I continued my same journey and built Durable Task Framework at Azure. And then out of coincidence, both me and Max actually ended up at Uber, where we, among other things—

John Furrier

>> Randomly at Uber? Or did you call each other up? You're going to Uber? Let's go together as a pair. Random?

Samar Abbas

>> It was complete luck, essentially. yeah. And we didn't actually stay in touch after AWS, and it was just a coincidence, essentially. So among other things, we built an open source platform there by the name of Cadence. The whole idea of Cadence is engineers who are composing low-level primitives like queues, databases, retry mechanisms, and durable timers to build resiliency into their applications. And 80% of the engineering was actually going into composing or stitching those things together rather than focusing on the actual problem that you are talking about, which Uber needs to solve to provide the best ridesharing experience. And this is where both me and Max looked at that and said, I think we can fix this problem.

John Furrier

>> And I think the service-oriented architecture is a great way to talk about AI now because AI is essentially a bunch of services. These agents are services. And a lot of the microservices stuff in cloud native actually map nicely to how agents need to behave. What resource am I touching? Is it the right resource? These are big questions. You got to think these through. So let's talk about the success. You got $550 million, over half a billion dollars in capital raised at a $12.55 billion valuation. Great success. Obviously, you guys have the track record. You're solving a big problem. What was that funding round like? Did people lean in? Do they understand it? What was the pitch? What was the pitch when you were doing the funding? Because there's a transformation story here. There's kind of an IT on steroids story. There's an agent story. It's a lot of things happening in your world.

Samar Abbas

>> Yeah. So let's take a step back, look at what has happened in the last 7 months since our previous round. We have crossed over 250 million of ARR. Our NDR, like net dollar retention, is over 200%, which means our customers running on top of our cloud are doubling their spend on us.

Samar Abbas

>> They're using more and putting more in. Yeah, they're expanding.

Samar Abbas

>> They are expanding. And then we have over 4,300 now customers running as cloud customers, essentially. So I and by the way, we don't get paid for AI enthusiasm. The thing we make money on is when these AI agents essentially, or applications, take real actions in the outside world. And this is where August alone we processed over 1.9 trillion actions. In a single month, which is up 350% compared to a year ago. So I think that

John Furrier

>> is—the pitch is more're the connective tissue for reliable, resilient applications running on enterprise or an application.

Samar Abbas

>> And this is exactly what drove a lot of interest from these investors at this point in time. Temporal has kind of become the default durable execution engine for powering the AI wave.

John Furrier

>> Yeah, it's just I will say that, in 17 years of doing theCUBE, the word resilience was always either a backup and recovery discussion or security. Resilience now is built into all applications as a standard feature because like more mentioned, these agents can't go rogue. You gotta, they gotta be resilient. They gotta react. They gotta make good decisions. It's not a search paradigm. It's like a re— it's a work paradigm. And so res— the word resilience has kind of been moved into the mainstream, cyber resilience, AI resilience. How would you define the word resilience? If someone says, hey, I need AI resilience, what would be your response to that? To someone, if you had to explain that.

Samar Abbas

>> So you have made an interesting point. As an engineer, this is a struggle I have seen throughout my— I've been in the industry for 26+ years. Every engineer, when they are tasked with a challenge, this concept of resilience that you are talking about has always been bolted on afterwards. And I think what this AI wave or platform shift is showing us is the world cannot operate in that world, especially what's been happening in the last few months, is people have to think about reliability of their systems from the ground up. When they are building it rather than bolting it on later. Because otherwise, that's how you see all of the incidents that you are seeing in the industry right now.

John Furrier

>> It's exciting because a lot of the old terms come back in vogue. Grid computing, now that's energy resilience. Now that's embedded into everything. You're starting to see self-healing come up. I heard that a few times. Self-healing networks. Remember back in the day, these agents have to heal and react on their own and make decisions. Repair. Oh, I don't do that, but roll back, do this, talk to this agent on that namespace. So a lot of the computer science has moved up the stack. What, what's your vision on how you take this funding and go forward on the build side and your customer side? What's the focus?

Samar Abbas

>> Yeah, first of all, a lot of funding is going to go into R&D, which is we need to keep on doubling down on durable execution, although we are clearly the leader in that space, but we are really aggressively investing into more and more primitives in durable execution so people can use it for a wider and wider class of applications. And, for example, what's the constraint there?

John Furrier

>> What's the constraint on durable execution? Just more primitives.

Samar Abbas

>> So let me give you an example. We just shipped a feature called Standalone Activities because now People see the power of durable execution, and it is a no-brainer when you are working on it for a workload which is highly critical or high value. But then people love the programming model. They want to use it for low-value use cases also. But sometimes cost comes in the way. So we are creating lighter-weight primitives so people should be able to adopt durable execution for pretty much every workload.

John Furrier

>> Yeah, durable execution, that's an easy sell. When you come and say, We provide durable execution. Like, sign me up. That's like, who doesn't want better durability? Explain what it means. What does durable execution actually mean?

Samar Abbas

>> Amazing. So the way I would describe durable execution is today when an engineer writes code or write and during an execution of a function, if a failure happens, you lose all your state. And what durable execution does is If you now implement the same function using Temporal's durable execution capabilities, that function can never fail. Which means if you run a money transfer application, which is a debit from an account and then credit into a target account, what happens if the transaction fails? A machine restarts when it's in the middle. That money just simply disappeared into thin air. What durable execution gives you a guarantee is That function can never fail, which means your transfers can never fail. We are— we will guarantee that your run, your transfer runs to completion.

John Furrier

>> If people aren't technical, it means that you have to start over.

John Furrier

>> Yeah.

John Furrier

>> And or it just doesn't work. It fails.

Samar Abbas

>> Basically start over in this money transfer case is not even an option because you have already lost that money into thin air because you did debit it from the—

John Furrier

>> we could probably put you in our Mixture of Experts series that we have too, because You're— we're getting in the weeds here in a good way. Talk about the importance of durable execution as it relates to agents and compare the importance of what APIs meant in the cloud era, because, you know, APIs didn't really have a lot of state. They were critical for the cloud growth for the services. Now you got state in agents as you cross these namespaces. It's like preserving the envelope. Talk about the importance of the data state and having that durability. And is it as big of an impact as APIs were for the cloud?

Samar Abbas

>> I think it will be even bigger impact because in the agent world, especially as I mentioned earlier, the agents are now graduating from just giving you answers to taking actions on your behalf. You give examples like, oh, booking your next reservation or your vacation or canceling an order or making a money transfer. These are actions agents have started to take on a person's behalf. And when these agents are taking actions, you want this durability layer backing those agents because you want each and every one of those agents to make sure they don't lose their thread. While they are in the middle of running a really critical task for you.

John Furrier

>> This transformation series to wrap up, it's really originally started around CFOs and business model transformation. It's evolved to essentially AI transformation, competitive edge because when you get into tokens and agents, you talk about money, not just money transfers, there's now value lines of sight into the economics of the results. It's not an IT project. People aren't just getting their apps loaded faster on the screen or some other IT benefit in the past. This is actual real money. And so this is a big transformation story. What's your advice to a CFO out there when they say, well, just bottom line me, do I make more money with you guys or not? What would be your answer?

John Furrier

>> So is it, sorry, making money?

Samar Abbas

>> Because if the system's running, it's going to produce revenue.

Samar Abbas

>> Oh, 100%. Yeah. And by the way, for instance, last year in October, I believe, there were a bunch of cloud provider outages happening. And this is where the impact of not having a reliable system was so clear because then, for example, Alaska runs their bookings on top of us. While their systems were down, they were not taking any bookings. Taco Bell, every order that you place at a Taco Bell is a Temporal Workflow. If you have a system which is down, you are not bringing in new revenue essentially. So this is exactly—

Samar Abbas

>> and they can stay up with you guys.

Samar Abbas

>> And this is where we have invested into features like Multi-region Namespaces, which allows a user with just a click of a button to completely migrate their workload to be powered out of a completely different region. And actually we support it across different cloud providers also. So depending on the availability requirements of whatever underlying workload you are powering and how critical it is, we give you all sorts of availability options to run the workload with the right level of nines that you are looking for.

John Furrier

>> Congratulations on the big funding, $550 million, half a billion, over half a billion. Validation, really valuation shows validation. Congratulations. Thanks for sharing on theCUBE.

John Furrier

>> Thanks.

John Furrier

>> Talking namespaces, agents, really the connective tissue is powering the agents. Everyone wants agents. You want that automation, but you have to have reliability. You need the resilience, you need the safety. That's the top conversation we're seeing today in AI and how that will come in. It'll come in through the entrepreneurs and the engineers working hard. This is theCUBE's Transformation Edge series. Thanks for watching.

Samar Abbas, Temporal

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street. I'm John Furrier, co-hosting with Dave Vellante, my co-host. Hello, I'm John Furrier, your host of theCUBE. We are here in theCUBE's NYSE studio. Of course, we have our Palo Alto studio on the West Coast connecting Silicon Valley to Wall Street. This is our Transformation Edge series. We talk to the leaders who are building technologies and are part of the massive transformation that's being enabled by AI and AI infrastructure, but more importantly, the software that's being generated to run the AI will change our lives for the better, to be safe, reliable. That's the focus today is to talk more about not just the productivity, but the safety. Samar Abbas is here. He's the co-founder and CEO of Temporal Technologies. Temporal just recently announced some big news, big funding. Thanks for coming on. It's great to see you.

Samar Abbas

>> Yeah, thanks a lot, John, for having me. Excited to be here.

John Furrier

>> We were just talking a little history lesson down memory lane almost a decade and a half ago about Amazon Web Services and the rise of the cloud. Now we have the rise of AI infrastructure, which is helping everybody. The rising tide lifts all boats. Yeah, the hyperscalers, AWS, Azure, all winning more revenue there. You've got the rise of the neo clouds, now called AI clouds. They don't like the neo cloud name, some called Neo Cloud Labs. Now you have the enterprises eyeing this next wave where they saw the value of cloud. Agents are coming right along fast in the enterprise, so they're going to leaning into deployment strategies for CapEx. We're seeing that clearly. I mean, they're not going to spend billions of dollars like a neo cloud, but they will connect to and put on-premises AI factories that'll spawn new applications, create new user expectations that everybody wants the AI. You guys are in the middle of it.

John Furrier

>> Yep.

John Furrier

>> Your valuation is really a significant, I think, template and a sign of what's coming. Explain the news and why that happened.

Samar Abbas

>> Yeah. So before I go into the news, let me give you a high-level intro. About what we are, because then it will explain things. So Temporal is an open-source platform which ensures durable execution of your code. What that means is today, if you woke up and published a Snap story or ordered a Taco Bell or used OpenAI's Codex agent, all of that was powered by Temporal. And so this is where what we've been trying to do is, especially as software, when it runs for longer periods of time, it interacts with the outside world. The outside world has lots of failure modes there. And during the execution of that software, people can lose their— software has always been losing their place, essentially, when those failures happen. So now what we do with Temporal is if that software is built on top of us, they will never lose their place.

John Furrier

>> Explain what people are buying, because I think this is a good point. So we'll get to the news. I think the momentum and the value proposition justify the big numbers. What is the reason? What's— what is actually being powered? Are you powering the app, are you powering the infrastructure? What specifically are you guys enabling and accelerating for your customers?

Samar Abbas

>> So, especially if you look at the last 18 months, what has fundamentally happened AI has stopped talking and started doing stuff. And this is exactly where Temporal is super differentiated, is as these models get smarter and smarter, when we talk to an organization, you always hear that when an agent runs longer, takes more action, does more stuff, it creates more value. It's super clear to those enterprises. And guess what? Temporal is the layer which is making sure that these agents can do that reliably and most importantly, securely take those actions to make sure they don't cause any harm. And this is exactly the reason why we have seen an amazing growth of those actions on top of our platform. Even in the month of August alone, we have processed over 1.9 trillion actions, which is a 350% increase compared to a year ago.

John Furrier

>> So you're talking about the talking, meaning the chatbots. We know when we first used it, hey, tell me about this, write the story for me, give me some marketing material, search for things. That's search. Yeah, you're giving it more work, okay, go figure this out. Complex problem. Organize my email, call my friend, book a flight. More multi-step activities that require touching more resource.

Samar Abbas

>> 100%. A very good example is, for example, coding agents. As these models are getting smarter, what we are seeing is these coding agents are now taking on more and more complex tasks, which run for hours. And during those hours, all sorts of failures can happen. Machines can restart, or downstream tools can fail or time out, or even models can return an error. These models have a pretty high error rate. So how do you remember all of that state? This is exactly where Temporal comes in to manage all of that state for your agents so they don't lose track of it and don't redo your stuff. And it actually helps us with saving tokens also.

John Furrier

>> You know, I was talking to an engineer who works at Stripe recently and this was not on camera. It was like an off-the-record conversation, so I won't say his name. He said these agents are great. People are deploying them like crazy. They're seeing value, as you pointed out. He says, but there's a big problem. I go, what? He goes, well, they're basically calling the main models and paying for the tokens and they're running it in a production environment. That's like us building a payment gateway and running it in production.

John Furrier

>> Yeah.

John Furrier

>> The sandbox, the test dev environment. So he's like, it's all broken. It's exciting. People are deploying and are leaning into it. They're enthusiastic, but they have to use production-like resources and costs to do the test dev. That kind of brings up the IT transformation kind of connection, which is, hey, I want to just build this app and I want to automate a lot of these things. I want to create that value, but I got to observe it. I got to have telemetry. I need to do all these, tech things under the covers. It sounds like that's what you guys are doing. Explain what that means for you guys. Because that's the problem that everyone's trying to solve right now.

Samar Abbas

>> And I hope, by the way, that everyone starts solving because this is becoming the problem. If the industry does not get it right, we might screw up the entire AI platform shift that we are seeing. And by the way, this is not a new problem. This is like when the cloud platform shift happened, we saw the exact same excitement with the cloud platform shift. But then industry came around and to your point, made all of the infrastructure to make sure you can run an application in a cloud with all of the necessary guardrails and protections which won't cause any harm. And I think the AI industry, as AI is now moving from just providing you answers to actually taking real actions on behalf of a user, they absolutely need to start putting in all of those guardrails.

John Furrier

>> The press loves to talk about, oh, it's off the rails, agents gone rogue. I've heard that we need a kill switch. We don't need a kill switch. So there's a lot of negativity around some of the failures in agents, but there's a lot of positive examples happening. What's your reaction to that? Because you're seeing a bifurcation of people going, well, I am scared or doom and gloom. And then you have another half saying this is the best thing we've ever seen. It's life-changing experiences. Societal benefits are off the charts. What's your reaction when you see this kind of polarization?

Samar Abbas

>> So let me first address the negativity comment that you mentioned. I think a lot of fear which is coming in is— I completely understand where it's coming from. But if you double-click on especially, let's say, the last few months, all of the incidents which have been coming out in the news, for instance, the Hugging Face hack which happened during an OpenAI testing, which is all over the place now.

Samar Abbas

>> Yeah, everyone uses that one. Yeah.

Samar Abbas

>> So if you kind of dig into it, one thing becomes absolutely clear. There is a pattern emerging that when these AI agents take actions, one of the things which we don't have in place is auditability of what that agent is doing. And in my opinion, that's one of the biggest gaps which is out there right now. So this is exactly the place where Temporal is differentiated, is when you build an agent using Temporal as an underlying platform, each and every step or action an agent takes, it's durably recorded in what we call an execution history of that agent. And nothing— there is no way anyone can come in and interfere with it because that's how the platform works natively. Basically, nothing is off the books. Everything is recorded durably. And every time that agent is taking an action, before it takes an action, it's durably recorded there. And then you have an audit log which can be used by any auditors to actually chase what that agent was doing when it goes rogue. And then the second capability Temporal adds is all of those actions are recorded before they happen, which means it gives an opportunity to that organization to put on their own guardrails. Which could be a human kind of approving that action or a business rule executed somewhere which might reject it. And so that's how you have this chain of who did what. And this is the role that's like rules of engagement too.

John Furrier

>> You have everything laid out. What's interesting about when I hear you talking, I get excited because there's a lot of parallels and it's kind of a history lesson to be learned here. If you look at AWS, early days of cloud where you have history, a lot of the higher-level services that Amazon ended up coming out with weren't even invented yet. the basic building blocks of AWS were a dream scenario for a startup. Didn't have any cash, put their credit card down. Everyone knows that story. Shadow IT emerged. You could actually get stuff done with those simple things. But then higher-level services came out. When you're talking about agents, I could see new things emerge like agents for agents. I talked with Jensen and NVIDIA team all the time about how they're using simulations and synthetic data, you can almost predict things. So you start to see as it gets out there and more adopted, what's your vision on that? What do you see that's possible to make things more reliable, more predictable? I think that's—

Samar Abbas

>> yeah, that's exactly the direction that we are headed. We have some customers who are already running over 5,000 namespaces on top of Temporal Cloud. Namespace is our unit of isolation where people run their own workloads. So what that means is we are already starting to become the API boundaries on how those organizations talk to each other. We have a project called Project Nexus, which is trying to create an industry-wide standard, how asynchronous communication happens in those larger enterprises. So as these agents get more and more long-lived, This is exactly where our investment on Project Nexus will be super differentiated because it becomes the backbone for how agents talk to each other.

John Furrier

>> I think you're the first interview guest I've had on theCUBE that used the word namespace. So congratulations, you get the award for namespace. Namespace is, for the folks watching, it's a very simple definition. Think of it like an address book, like your address at home, it knows things, knows where things are. And that's important. I want to double click on that because I think it's a nuanced tech point, but it's the science behind the computer science where if you understand the domains and the addressing system, discovery, reliability, paths, efficiency, explain the importance of having a good namespace system or addressing system and why is that super important?

Samar Abbas

>> So think of namespace as an envelope where we give you an envelope and use that envelope to run your specific application within that envelope. And it gives you complete isolation from anything else. And it's especially important because Temporal Cloud is a multi-tenant cloud service. What that means is today we are completely hiding all of the infrastructure from you. You come in and create an account on Temporal Cloud, and then when you want to build an application, the first thing you build is a namespace, which gives you a box where you can now start running your workflows and activities inside. The reason it's important is how do you guarantee one namespace does not interfere with the workload or performance characteristics of another namespace? And that's where this envelope becomes really, really interesting because it gives you complete isolation from a performance standpoint. And you like—

John Furrier

>> you can get things delivered. It gets to the endpoint. Yeah. I want to talk about your history. You and your co-founder have— I think it's what I love about this era. It's not talked about a lot. In the media is the history of the computer industry and specifically open source has a saying standing on the shoulders of giants, meaning because it's open source code, you can build on top of code. Great. We love open source, but you start to see the giants' shoulders like Uber, like AWS. You work at Uber. A lot of this stuff has been solved for these early first movers that had to build their own system. Let's take Uber, for example. Yeah. they literally had to build the app that we all love and use. Lyft did the same thing. it's basically you log in to the cloud, multi-tenant cloud. You actually have to hail a driver who doesn't even work for the company. Yeah, you got to track that person with GPS. You have an account, a database for all your stuff. You got to bring all these things together in real time. Now that's a state-of-the-art problem. Go back when they originally founded it— that was really hard. But that was also a marker. Trail marker for what would be coming to the mainstream. So in a way, a lot of the innovation that you're doing and others have been there and the market just hasn't evolved yet. And now it has. Explain the importance of this. You didn't just wake up one day and say, hey, I'm going to start Temporal.

John Furrier

>> Yeah.

John Furrier

>> Talk about the importance of the trajectory of the industry and why those other advancements were and what it means for the AI world. What happens next?

Samar Abbas

>> 100%. So very good question. So by the way, there are founders who sometimes wake up one day, they have an idea and then they build a company around it. Both me and my co-founder Maxim, we've been iterating over this problem for at least 20+ years now. It started actually at Amazon AWS. So where we saw the transformation of Amazon from a monolith into a service-oriented architecture. And this is where we built Simple Workflow as an underlying application platform to help with those massive service-oriented architectures to become the reliability layer for building those applications. Then I continued that at Azure where I was part of a team at Azure called Azure Service Bus, which owns the messaging stack. And I very quickly saw similarities on how engineers are using messaging as a core underlying primitive to build essentially reliability or resiliency into their apps. So I continued my same journey and built Durable Task Framework at Azure. And then out of coincidence, both me and Max actually ended up at Uber, where we, among other things—

John Furrier

>> Randomly at Uber? Or did you call each other up? You're going to Uber? Let's go together as a pair. Random?

Samar Abbas

>> It was complete luck, essentially. yeah. And we didn't actually stay in touch after AWS, and it was just a coincidence, essentially. So among other things, we built an open source platform there by the name of Cadence. The whole idea of Cadence is engineers who are composing low-level primitives like queues, databases, retry mechanisms, and durable timers to build resiliency into their applications. And 80% of the engineering was actually going into composing or stitching those things together rather than focusing on the actual problem that you are talking about, which Uber needs to solve to provide the best ridesharing experience. And this is where both me and Max looked at that and said, I think we can fix this problem.

John Furrier

>> And I think the service-oriented architecture is a great way to talk about AI now because AI is essentially a bunch of services. These agents are services. And a lot of the microservices stuff in cloud native actually map nicely to how agents need to behave. What resource am I touching? Is it the right resource? These are big questions. You got to think these through. So let's talk about the success. You got $550 million, over half a billion dollars in capital raised at a $12.55 billion valuation. Great success. Obviously, you guys have the track record. You're solving a big problem. What was that funding round like? Did people lean in? Do they understand it? What was the pitch? What was the pitch when you were doing the funding? Because there's a transformation story here. There's kind of an IT on steroids story. There's an agent story. It's a lot of things happening in your world.

Samar Abbas

>> Yeah. So let's take a step back, look at what has happened in the last 7 months since our previous round. We have crossed over 250 million of ARR. Our NDR, like net dollar retention, is over 200%, which means our customers running on top of our cloud are doubling their spend on us.

Samar Abbas

>> They're using more and putting more in. Yeah, they're expanding.

Samar Abbas

>> They are expanding. And then we have over 4,300 now customers running as cloud customers, essentially. So I and by the way, we don't get paid for AI enthusiasm. The thing we make money on is when these AI agents essentially, or applications, take real actions in the outside world. And this is where August alone we processed over 1.9 trillion actions. In a single month, which is up 350% compared to a year ago. So I think that

John Furrier

>> is—the pitch is more're the connective tissue for reliable, resilient applications running on enterprise or an application.

Samar Abbas

>> And this is exactly what drove a lot of interest from these investors at this point in time. Temporal has kind of become the default durable execution engine for powering the AI wave.

John Furrier

>> Yeah, it's just I will say that, in 17 years of doing theCUBE, the word resilience was always either a backup and recovery discussion or security. Resilience now is built into all applications as a standard feature because like more mentioned, these agents can't go rogue. You gotta, they gotta be resilient. They gotta react. They gotta make good decisions. It's not a search paradigm. It's like a re— it's a work paradigm. And so res— the word resilience has kind of been moved into the mainstream, cyber resilience, AI resilience. How would you define the word resilience? If someone says, hey, I need AI resilience, what would be your response to that? To someone, if you had to explain that.

Samar Abbas

>> So you have made an interesting point. As an engineer, this is a struggle I have seen throughout my— I've been in the industry for 26+ years. Every engineer, when they are tasked with a challenge, this concept of resilience that you are talking about has always been bolted on afterwards. And I think what this AI wave or platform shift is showing us is the world cannot operate in that world, especially what's been happening in the last few months, is people have to think about reliability of their systems from the ground up. When they are building it rather than bolting it on later. Because otherwise, that's how you see all of the incidents that you are seeing in the industry right now.

John Furrier

>> It's exciting because a lot of the old terms come back in vogue. Grid computing, now that's energy resilience. Now that's embedded into everything. You're starting to see self-healing come up. I heard that a few times. Self-healing networks. Remember back in the day, these agents have to heal and react on their own and make decisions. Repair. Oh, I don't do that, but roll back, do this, talk to this agent on that namespace. So a lot of the computer science has moved up the stack. What, what's your vision on how you take this funding and go forward on the build side and your customer side? What's the focus?

Samar Abbas

>> Yeah, first of all, a lot of funding is going to go into R&D, which is we need to keep on doubling down on durable execution, although we are clearly the leader in that space, but we are really aggressively investing into more and more primitives in durable execution so people can use it for a wider and wider class of applications. And, for example, what's the constraint there?

John Furrier

>> What's the constraint on durable execution? Just more primitives.

Samar Abbas

>> So let me give you an example. We just shipped a feature called Standalone Activities because now People see the power of durable execution, and it is a no-brainer when you are working on it for a workload which is highly critical or high value. But then people love the programming model. They want to use it for low-value use cases also. But sometimes cost comes in the way. So we are creating lighter-weight primitives so people should be able to adopt durable execution for pretty much every workload.

John Furrier

>> Yeah, durable execution, that's an easy sell. When you come and say, We provide durable execution. Like, sign me up. That's like, who doesn't want better durability? Explain what it means. What does durable execution actually mean?

Samar Abbas

>> Amazing. So the way I would describe durable execution is today when an engineer writes code or write and during an execution of a function, if a failure happens, you lose all your state. And what durable execution does is If you now implement the same function using Temporal's durable execution capabilities, that function can never fail. Which means if you run a money transfer application, which is a debit from an account and then credit into a target account, what happens if the transaction fails? A machine restarts when it's in the middle. That money just simply disappeared into thin air. What durable execution gives you a guarantee is That function can never fail, which means your transfers can never fail. We are— we will guarantee that your run, your transfer runs to completion.

John Furrier

>> If people aren't technical, it means that you have to start over.

John Furrier

>> Yeah.

John Furrier

>> And or it just doesn't work. It fails.

Samar Abbas

>> Basically start over in this money transfer case is not even an option because you have already lost that money into thin air because you did debit it from the—

John Furrier

>> we could probably put you in our Mixture of Experts series that we have too, because You're— we're getting in the weeds here in a good way. Talk about the importance of durable execution as it relates to agents and compare the importance of what APIs meant in the cloud era, because, you know, APIs didn't really have a lot of state. They were critical for the cloud growth for the services. Now you got state in agents as you cross these namespaces. It's like preserving the envelope. Talk about the importance of the data state and having that durability. And is it as big of an impact as APIs were for the cloud?

Samar Abbas

>> I think it will be even bigger impact because in the agent world, especially as I mentioned earlier, the agents are now graduating from just giving you answers to taking actions on your behalf. You give examples like, oh, booking your next reservation or your vacation or canceling an order or making a money transfer. These are actions agents have started to take on a person's behalf. And when these agents are taking actions, you want this durability layer backing those agents because you want each and every one of those agents to make sure they don't lose their thread. While they are in the middle of running a really critical task for you.

John Furrier

>> This transformation series to wrap up, it's really originally started around CFOs and business model transformation. It's evolved to essentially AI transformation, competitive edge because when you get into tokens and agents, you talk about money, not just money transfers, there's now value lines of sight into the economics of the results. It's not an IT project. People aren't just getting their apps loaded faster on the screen or some other IT benefit in the past. This is actual real money. And so this is a big transformation story. What's your advice to a CFO out there when they say, well, just bottom line me, do I make more money with you guys or not? What would be your answer?

John Furrier

>> So is it, sorry, making money?

Samar Abbas

>> Because if the system's running, it's going to produce revenue.

Samar Abbas

>> Oh, 100%. Yeah. And by the way, for instance, last year in October, I believe, there were a bunch of cloud provider outages happening. And this is where the impact of not having a reliable system was so clear because then, for example, Alaska runs their bookings on top of us. While their systems were down, they were not taking any bookings. Taco Bell, every order that you place at a Taco Bell is a Temporal Workflow. If you have a system which is down, you are not bringing in new revenue essentially. So this is exactly—

Samar Abbas

>> and they can stay up with you guys.

Samar Abbas

>> And this is where we have invested into features like Multi-region Namespaces, which allows a user with just a click of a button to completely migrate their workload to be powered out of a completely different region. And actually we support it across different cloud providers also. So depending on the availability requirements of whatever underlying workload you are powering and how critical it is, we give you all sorts of availability options to run the workload with the right level of nines that you are looking for.

John Furrier

>> Congratulations on the big funding, $550 million, half a billion, over half a billion. Validation, really valuation shows validation. Congratulations. Thanks for sharing on theCUBE.

John Furrier

>> Thanks.

John Furrier

>> Talking namespaces, agents, really the connective tissue is powering the agents. Everyone wants agents. You want that automation, but you have to have reliability. You need the resilience, you need the safety. That's the top conversation we're seeing today in AI and how that will come in. It'll come in through the entrepreneurs and the engineers working hard. This is theCUBE's Transformation Edge series. Thanks for watching.