Why Every AI Agent Needs Durable Execution | Temporal, Samar Abbas

This is a big platform shift happening. The cost of building software is going to radically change. The value is definitely shifting from software to eventually how do you operationalize or how do you put that software out there. So first set of things that we want to how do we make durable execution more approachable. How do we simplify? They don't need to be distributed systems experts to be able to adopt and leverage the technology. We are becoming the execution authority on how to make sure how work gets done. You and your co-founder have spent 20 years iterating on this problem and have built five iterations of the system. You're powering some of the biggest banks, the AI labs, and some of the biggest internet applications out there today. It's a new paradigm. It's a new way of building applications. We are creating a new category, essentially. World is not going to stop here. These things are going to get more and more smarter, and we are eventually going to move out of a single sandbox into a whole swarm of those agents. There is no way you can have that shift without a stable infrastructure like durable execution powering that. Hi, everyone. I'm Anushka Vaswani, a partner at Lightspeed, and this is The Investment Memo, the show where founders reveal the stories behind their businesses and how they will shape the future of their industries. I'm joined by Samar Abbas, co-founder and CEO of Temporal. Samar and his co-founder Max have been working on this problem for the last 20 years. Temporal is the culmination of their experience, enabling software to run reliably and execute to completion. It has become the de facto solution to power distributed systems and AI workflows and is being used by some of the most cutting-edge AI companies, from open AI to lovable. and largest enterprises like JPMorgan Chase. Today we'll talk about why durable execution has become one of the hardest problems in software, why Temporal has become mission critical for AI and agentic workflows, and what it takes to build infrastructure that developers trust with their most critical systems. Summer, welcome to the show. Yeah, thanks Anushka for having me. Tremendous congratulations. What an incredible milestone on your $300 million Series D. It's a stage very few companies get to. And so we wanted to reveal some of the investment memo behind our work and thinking and why we got so excited about Temporal and the opportunity. Yeah. Thanks a lot, Anushka. I am thrilled to be partnering with Lightspeed in this next phase of growth for the company. And then, yeah. super exciting time for us at Temporal. In our memo, we described Temporal as a durable execution platform, a category you created. For viewers less familiar with Temporal and durable execution, could you share a little bit about what that means? Yeah. So Temporal is an open source platform which ensures durable execution of your code. What that means is during an execution, if there is a failure and you lose all your state, Today, what Durable Execution provides is we remember all of the states, your local variable, your call stack and everything. And we are going to resurrect that execution on a different host and continue executing seamlessly from where it left off without you as a software developer writing a single line of code for it. In a nutshell, that's how I typically define what Durable Execution is. And today, almost... every AI company we speak to from Harvey to a bridge to the big AI labs are using Temporal when they want to put a product in production. Why have you seen so much pull in AI? What's happening that's made Temporal the de facto architecture for bringing AI into production? Let's imagine an application like managing an order system for a very busy kitchen yeah so as you can imagine on a busy night a kitchen is doing like hundreds of orders coming in the behavior they want is like oh the orders can come in with spikes or suddenly many customers walks in or some orders are fast to prepare some are slower uh at the same time there is all sorts of chaos going on in the kitchen or a station might go down or particular like chef might have to step out a new person needs to come in but at the end of the day what you are that kitchen is expecting is every order is prepared in exactly the same sequence of steps yeah and delivered exactly once to the customer in a timely fashion that's what they are looking at the business logic for that kitchen is oh taking an order and how to define break it down into steps and prepare it that's what they care about that's what their business logic for their system is. But they are today like handling spikes of orders coming in or all of the failure modes that I talked about and recovery from those failure modes and how to guarantee each and every order is delivered exactly once requires lots and lots of engineering. And so this is what durable execution is abstracting out. is we kind of helping, we are giving you a mechanism where you focus on your business logic. Because all of those other things slowly start sweeping in into your application logic and which makes your code bulky. And what we have seen in the past is 80% of your engineering goes into coding those rather than the actual problem that you are trying to solve. So this is the journey both me and Max have been on. We have seen all variations of that problem happening. And maybe just to crystallize that, you know, even before we get into AI and the tremendous pull you're seeing there, you know, maybe I love the kitchen analogy and maybe talk about kind of that kitchen analogy before and after a team is using Temporal. Because when we spoke to a number of your customers, people talked about it as night and day in terms of how their application ran before Temporal and what that took and what that looked like, and then just the shift that Temporal enabled them to realize. If you look at a lot of interactions in the world, very asynchronous. All sorts of chaos happens in real-world interactions. A lot of time, what we see is before Temporal, what people at least industry seems to have gravitated towards smaller services or microservices doing one aspect of your preparing your salad. Yeah. And but then you need an orchestrator to orchestrate calls across those services in a reliable fashion. That is the state of your order. Yeah. So what we typically see is people build these complex architectures through even driven systems by stitching together queues and databases and retry mechanisms and durable timers whatnot to build resiliency for that state management piece essentially so at the end of the day what like 80 percent of the engineering is building all of that glue rather than implementing the logic of how to process an order essentially and what temporal like after uh like once an organization starts adopting temporal what we see is like slowly they start moving away from those event-driven architectures or replace a queue or replace retry logic they start adopting it by solving smaller problems and then eventually they get on a journey to kind of replace the entire order management flow for a kitchen using temporal as a core and orchestrator essentially so we at the end of the day what they are doing is replacing all of that complex 80 of the glue code that they were hand coding themselves and replacing it with and like a fully production ready orchestrator which supports retries and recovery out of the box for you rather than you hand coding it for each and every application. Yeah, it was incredible. I think, you know, when we spoke to so many customers, people had a massive spaghetti code that they, you know, had to spend significant amounts of engineering time and resources constructing that ultimately they didn't have visibility over and ultimately couldn't scale that Temporal has completely changed the game on. You know, maybe then transitioning to kind of the AI. trend and this tremendous pull you're seeing, why have so many AI companies flocked to Temporal and what's happening in the industry such that if you want to take an AI application and put it in production today, all of these developers are standardizing on Temporal? Yeah. If you now talk about what has changed in AI, I think the core foundational change is previously what engineers used to do is take a business logic and then basically express that business logic in deterministic code essentially. And the core foundational shift with AI, especially agentic AI these days, is now these LLMs are smart enough where they can break down any complex problem into a sequence of steps on how you should be implementing it essentially. And so this is where a completely new form of application architecture is evolving, where the next generation of these agentic application is, you pick an LLM, you have a prompt, you give it some context and a set of tools, and just run this agentic loop to eventually get into an end state. So that's the core foundational shift which is happening in the world today. But at the end of the day, is what we are finding is once you build this app operationalizing this app like these llms are awesome it's telling you what to do but how things work get done who's the execution authority this is where temporal is kind of kicking in we are becoming the execution authority on how to make sure how work gets done so this is why all of the companies you are talking about are flocking to temporal as a core underlying orchestrator to basically guarantee as more and more workloads, as these LLMs get smart enough and these agents are taking more and more business critical applications, we are seeing these applications, these agents are getting more and more long-lived, more asynchronous, requires more complex human interactions, and more mission critical and more high throughput and high scale. At the end of the day, you are back to solving the exact same problem that we are trying to solve with Durable Execution for the event-driven architectures as evolving. So this is where Durable Execution has just found its ideal application in terms of AI, and the world is now coming around to it. And this is why a lot of developers in those organizations who already knew the concept of Durable Execution are now starting to use it as a core piece of infrastructure for powering their agentic applications. And maybe on that point, just to make it really concrete for everyone, talk through some of these incredible customer examples you're seeing and how some of these individuals are using Temporal to bring these agentic workflows to life. Yeah. So today, when you talk about the whole spectrum of AI, There is a lot of things, right? Like there are, of course, training models or like control plane or like schedulers for model training or like machine learning pipelines or basically your rags or your data or your inference engines where a lot of these models at the end of the day is like passing it the right context. And passing that context is pulling data from a wide variety of sources. It could be Slack messages. It could be an API. or it could be a Google Doc or all sorts of connectors to pull all of that data massage it, put it in drags and drive context for your prompts essentially. Then of course, they can take loops like orchestration of real business processes. We see Temporal for pretty much a whole wide spectrum of those use cases. To give you a few highlights, for example, OpenAI is a big user of Temporal. Codex is actually one of a very good example to talk about is today, Temple is used as the core orchestrator because what people are trying to do with Codex is actually a very good example of long-lived workflow. This long-lived workflow making it fully durable across various steps that you need to plan. You need to execute various tasks, you need to create your new pull request, you need to test it, you need to go through all of those stages, and you need to do it for millions of requests going out there simultaneously at a very large scale. Doing all of that work and guaranteeing execution for that, this is where we are adding value for the codex use case essentially. Same thing Sora. Image generation is actually another very good example. It's a perfect example of a data pipeline. Image generation also requires orchestrating calls across many services at the end of the day and doing that reliably and orchestrating all of that is we are the backbone or the execution authority for that. Same thing if you move to a few other examples like Replit. So Replit uses Temporal for their new agentic platform that they recently shipped. We are the control plane for managing a session, essentially. Every time you start an agentic session, we kick it off as a workflow and then the whole life cycle of that session and guaranteeing all of the work happening in that session happens reliably and durably. Temporal is the core backbone for those. A bridge is a very good example of all of these audio or video files and transcoding. That pipeline for each of those audio files are transcoded and summarized and all of the nodes being produced. Temporal is the orchestrator for managing that data pipeline in a reliable and a seamless fashion i can keep going on so like as you can see like people are leveraging the core capabilities of temporal to a pretty broad spectrum of uh use cases from the large labs to ai native companies to even the application layer which is now starting to show up that's exactly what we heard as well where so many ai pilots fail And when you want these agentic workflows to work in the real world and interact with systems and drive business critical processes, you need determinism. You need guarantees that things will execute as you need them to. And, you know, the only solution that powers this in the market today is temporal. So it's an incredible testament to what you've built. You know, very few companies have seen this type of pull from the AI ecosystem. How has this changed the way you're thinking about temporal and the temporal opportunity? We are kind of embracing AI heads on, essentially. If you look at our product roadmap right now, the core fundamental thing which is changing out there in the industry, basically is two things. First of all, a lot more developers are going to come in the fold. of building applications. We are super excited about that change. By the way, at the end of the day, Time Foral is a company which obsesses about developers. Our entire vision is how to give time back to those developers so they can innovate. what AI is enabling, kind of literally lowering the bar of what it takes to develop an application is super exciting for us. Even I've become a developer. Amazing, yeah. And so that's one thing is, first of all, how do we simplify? If you think about it, the whole history of me and Max comes from a background of hardcore distributed systems, backend engineer. in organizations like Amazon or Uber or Microsoft, like these digital native companies. But at the same time, I think what we are building is actually, it empowers all the rest of the developers, essentially, who really care about their application, who have an idea in their head, and they really want to build an application and bring that idea to life, essentially. And this is where they get bogged down by Eventually their entire task turns from building software or their application into becoming experts in distributed systems. So we've been democratizing distributed systems for a very long period of time. So first set of things that we want to, how do we make durable execution more approachable? How do we simplify? How do we dumb it down to a point where people can, they don't need to be distributed systems experts to be able to adopt and leverage the technology. uh for their especially for the agent uh development essentially so that's this is uh simplifying is a one big area of product investment for us essentially another thing that we are going to see is a lot of software is going to be written yeah really really fast the cost of building software is going to be radically changed in the next uh it's actually happening already and it's in the next year it's going to radically change which means there will be a lot of new software going to be written and i think the value is definitely shifting from software to eventually how do you operationalize or how do you put that software out there so there is a lot of focus on reliability how do we expand the spectrum the breadth of durable execution to be applicable to wider and wider class of use cases so there is a lot of focus we are doing A very good example is task queue priority and fairness feature. I think it solves a very complex problem which brings a completely new class of applications to be viable on top of durable execution as a core underlying platform. I think that's the biggest area is if a lot more software gets written, how do we basically turn that into a production ready? You can really quickly into even a larger enterprise. How can a larger enterprise adopt all of this? insane explosion of software and run it in a reliable and seamless fashion with all of the guardrails in those enterprises is a big focus for us. Maybe on that point, you know, Temporal is in the critical path of some of the most important workflows. You're powering some of the biggest banks, you're powering, you know, the AI labs and some of the biggest internet applications out there today. And you've always been from the very beginning, very thoughtful about how to construct Temporal's infrastructure and architecture such that people are willing to use it in their most mission critical applications and workflows. What decisions have you made that have driven that? Yeah. So I think the first, by the way, decision that we made, which was a little bit controversial, but both me and Max are a big believer, is open source. Yeah. I like if you look at it, this is probably the fifth time we are building a similar system. And I think that one of the things we felt is a software like Temporal is such a core requirement for any software developer out there where we need to make sure this software is available out there in the open source. Anyone can, because it's a new paradigm, it's a new way of building applications. The only way you make this paradigm shift happen is to make it viral, make it available to every developer on the planet. So one probably that's why like the whole DNA in the company is really open source first philosophy. Even today, every each and every feature that we do, we want we guarantee that feature works in our open source and not only on our cloud product, which we monetize essentially. To a point where we hold ourselves accountable that our cloud is positioned around being the best backend for running temporal workload rather than oh we hide features essentially from users we are not an open core company we provide full guarantee at an application level if you build an application on top of temporal today it is guaranteed to work seamlessly on open source to cloud literally you can migrate that application live with almost zero lines of code change from open source to cloud. And if you're not happy with cloud, we will actually help you migrate it back to open source with zero downtime. So we have actually invested into that migration capability. So that was one of the big decisions that we made. Another actually interesting decision that we made early on how we architected is we have clear separation of the backend and application. What is the state management, which is a temporal server, which has all of how you scale, reliability, and all of those promises come from the backend temporal server. And the programming model, which we have SDKs in seven languages, and this is what developers are using to build those applications today. There's clear separation, and you can run it anywhere. We are not in the business of running user code. This actually worked really well for us for enabling adoption in larger enterprises. because that's how we those enterprises were able to trust that oh our security posture was you run your code we don't run that your code um and and this actually enabled a lot of enterprise to adopt a temporal cloud at a very early stage of our journey um but at this is uh but at the same time if you look at the other segments like startups or mid-market or like smaller businesses they typically care about faster developer loop. Oh, I want to build something and try it out and iterate over my solution. So I think this is one area where today we don't provide a fully hosting solution. This is something which we are fixing and investing a lot this year to provide a fully packaged hosting solution as part of Temporal Cloud also, because we do see the need for these different segments. who just want a solution which solves it end-to-end rather than a separate decoupling of hosting and the backend, essentially. You alluded to this earlier in the conversation, but we're moving from using AI synchronously to now these agentic workflows running independently for hours, days, and potentially we'll start to see months or even longer at a time. Maybe talk about how that impacts Temporal. and why that trend is so interesting for the company? So at a high level, I kind of explained like the two things which are kind of changing by AI already, a lot more developers and a lot more applications. And another change now that you are talking about is as people at these models are getting smarter and smarter, these agents are taking on more and more work asynchronously on the back end. which is actually very exciting to see. At least one analogy that I was thinking about earlier is to me, it feels like right now, these agents are in the MS-DOS era, where if you think about it, it's like the architecture which has started to emerge already is, oh, I need a sandbox. These agents, who knows what they are going to do? Let's have a sandbox where these agents have full control. and cannot take actions to destroy something or delete everything on an enterprise or something like that. So people are kind of creating these sandboxes, empowering them as your prompt, your model, giving them access to file system or your data, and a set of tools. Oh, run a bash script here or do take these actions, but all in the... context of that sandbox. Even that people are finding it like for example, Claude Code which is a coding agent like that it actually works extremely well. People are finding this so useful already. But as these models get smarter and smarter, we eventually are going to run out of capability of a single sandbox. I think there will be whole swarm of these agents. I think that we are going to see an explosion of these agents. And very naturally, these agents will be talking to other agents. For example, like booking your vacation. Well, I am pretty sure there will be an API somewhere where these agents have to go outside of the context of that sandbox and eventually make an API called to book your airline ticket your hotel reservations or your activities or something and talk to those external services or maybe external other agents uh to for those purposes and so i clearly see we eventually we are uh in this very very early stage of ms dos era for agents and which people are finding it already pretty useful and it's already doing a very high value work or creating business outcomes already. But the world is not going to stop here. These things are going to get more and more smarter and we are eventually going to move out of a single sandbox into a whole swarm of those agents. And by the way, then you are back to solving distributed systems problems. My other co-founder, Max, he's very early Amazon. He saw the transformation of Amazon from a monolith into a service-oriented architecture. And he saw how those event-driven architectures, kind of queues and all of those enable that. And this is why we build durable execution to kind of simplify software. We believe that problem is going to about to go on steroids. with this explosion of those agents and explosion of those developers and explosion of those applications showing up. I honestly feel if durable execution was something which was needed, even in this cloud or distributed computing world, essentially, with the whole agentic platform shift which is happening, there is no way you can have that shift without a stable infrastructure like durable execution powering that. Switching gears a little. It would be great to talk about your personal journey and the culture at Temporal. You and your co-founder have spent 20 years iterating on this problem and have built five iterations of the system before starting Temporal. What was the initial insight for starting Temporal and then what made you take the leap to start a company? I think I heard Jensen, this earlier talk from Jensen is everyone can have a perspective. about the world, essentially. Who is an entrepreneur, by the way? It's about who have a perspective about something. And everyone can have it. It's not that hard. I think this is back in 2009-ish, 2010, when I joined the Simple Workflow team at AWS. Both me and Max were convinced software development for these cloud-native environments was broken. And we had this perspective, it's like why software needs to be so complex? Because we lived through that pain. We lived through like as a developer, suddenly my job changes from providing real business outcomes to the businesses to becoming experts in distributed systems. I for one is someone who actually enjoy distributed systems, deeper systems problem a lot, but I run into a lot of developers who don't enjoy that, but they are forced. to becoming experts in distributed systems because otherwise they cannot deliver those applications for real uh production scale and so we and and if you look at back then um the industry was they just accepted software is hard so but how all of the innovation that was happening is how to manage that complexity both me and max went on a different tangent okay why software is hard can we make it simpler? And we had this perspective on, oh yeah, we can, the whole idea which is now communicated as durable execution, we had it back then is like how, yeah, we can really make software development simpler. And I think so that perspective, like back then, there were really little believers, but then over the next 15 years, that constant iteration. And I think the approach we took it is like we had that perspective and then we iterated like crazy. We worked within the field with every developer, one application after another application. By the way, like people ask me a lot is like we come up with a, this is what a durable execution. We had vague abstract idea, but at the end of the day, the approach we took it is solving practical problems. is each and every feature in the product, I can eventually tie it to a real outage or real problem that I ran into myself or some other team. And we were trying to solve that. And that eventually resulted into this whole slew of Lego blocks that we built, which is now described as durable execution. What made you take the leap to start the company? It's funny that even 15 years ago, Both me and Max probably have a couple of screws loose. It was so obvious that this is the right answer. This is one thing I've seen in other founders or entrepreneurs is they have so much clarity that this thing cannot fail. It's an obvious answer. So for me, even like I've been, by the way, I've been iterating over this space because I know this is the right answer. And we've been taking that leap to starting Temporal with both me and Max somehow ended up at Uber. Where at Uber, basically, we saw in practice how this idea can enable a large organization which is moving really, really fast. All of the developer productivity wins and all of the reliability problems that we solved at the company was just insane. But at the end of the day, Uber was in the business of getting people. in the car essentially building a general purpose orchestration solution was not one of the core missions of the company and both me and Max kind of by then build enough belief that oh this problem is not specific for organizations like Amazon or Microsoft or Uber every modern organization is dealing with similar challenges and then we eventually took the leap and say okay let's try to do it go after the larger market yeah And when you started the company, you started as CTO and Max started as CEO and you've recently switched roles. How has your leadership style changed over your journey at Temporal? Yeah, it's been a humbling experience for sure. Both me and Max are very product people. Both can go to any level of detail in the product essentially and we love having product conversations. Building a company is... much much larger than the product itself of course like one of the core things you need to have an amazing product and which in the form of temporal we are super lucky to have such an amazing product with a great product market fit but there are many examples out there the great product doesn't necessarily translates into a great company also essentially um so i um i actually found myself i every like starting from just the two of us to we are now close to 400 people that journey like i essentially had to keep on transforming or reinventing myself yeah and one of the um between me and max typically the way like we have so much trust between each other max is amazing at taking a complex problem and distilling it in a way where where do we want to be in the next five years my role always has been what are the steps we need to take to deliver on that vision that's how both me and max are an awesome combination appear like that we work really well essentially what we found out like a couple of years ago is as the company got larger and larger it was becoming lesser about not having a product strategy it's about majority of the risk in the company was execution risk essentially And this is where actually Max himself came to the realization, asked me somewhere, I think it's the right time for the company for you to step in. And because the company needs a different flavor of leadership to kind of help us take through the next phase of that journey. So that kind of what led to the switch between us, essentially. By the way, that conversation was like less than 15 minutes. That's amazing. So few companies have been able to establish the relevance. that Temporal has in AI and the ubiquity Temporal has with developers. How have you driven a culture that's enabled that? So that is actually a tough, by the way, challenge. I honestly feel right now there's a big platform shift happening. Any organization out there which is not embracing AI, I honestly feel they will, in the longer run, they will not survive, essentially. Because the level of productivity the industry is about to see with AI is just going to be so orders of magnitude different. People who are not adopting this right now essentially will have a very hard time competing in the market going forward. So at least there are a few things that we have done is we first of all last year first thing remove friction. You need to have a culture of experimentation. in any organization. So I actually went with my head of infrastructure and told like, can we just enable everyone? And anyone coming in and want access to cloud or open AI or like any other tool, can we just enable everyone essentially? And so I think so that we started last year. So organically bottoms up, lots of interesting conversations started happening. Then you give them a forum to share what people have built with others in the company. And I think so this is where we have every two weeks, we have a sprint showcase and where we invite the entire company and then developers or engineers who are building cool things. Essentially, it's an opportunity for them to kind of showcase that to the rest of the organization. And I think really organically, what we see is like there was a theme in that if you look at our sprint showcase presentations or demos a year ago and now. there is already a pretty clear shift starting to happen more and more towards AI. For instance, one of the latest showcase, which actually I was super excited, is one of the engineers who recently joined built their own AI buddy to help them onboard, essentially. I love it. Yeah. And by the way, for a company which is growing so fast, essentially, it's super useful. And then suddenly, like through showcase, that person showed, oh, how he did it and how he's leveraging it. And then suddenly, other people have started using it essentially. So I think for this whole culture of experimentation in this special era where this whole space is evolving so fast, I think that's the muscle you want to build. Yeah, I absolutely agree. And how important is the capital stack you've assembled in this vision you're going after? I mean, you have some of the very best investors in the world on your cap table. How are you thinking about that? One thing I've learned is you raise capital when you don't need it. And I think, yes, of course, raising the capital is a big thing for us. By the way, we believe Temporal has a massive opportunity in front of us. And building that kind of infrastructure is going to come at a pretty big cost. For example, that's why we are investing a lot in R&D going forward. And I think this goes back to the same principle that me and Max had is we are big believers. We believe we can completely change the way how certain class of app, we are creating a new category, essentially. And the same rationale I had is why we left Uber to start this company is because we believe that this is such a gigantic opportunity. And if we stayed at Uber, we could never get the necessary investment to materialize that or capitalize on that opportunity. And I think that's exactly, even today, by the way, the company is bottlenecked on our ability to produce software. And that's why we are hiring with all of the enhancements on developer productivity. We are hiring a lot of developers still today, essentially, because we have so many ideas, so many dimensions where we can take temporal into. We are still kind of bottlenecked essentially on like on those investment areas in the product itself. So I think what this capital enables us is to basically to have a little bit higher risk appetite on investing in those areas a little more aggressively. Our entire partnership could not be more excited about working with you on this opportunity. Maybe share a little bit from your side around why Lightspeed? Yeah. So first of all, Anushka, like a big part of that is I've been working with you for almost a year and a half, like almost immediately after taking on the CEO role, I think the quality of conversations, the quality of like research Lightspeed has done is completely mind blowing on top of temporal. Temporal and durable execution in general is a very hard concept to grasp for developers, let alone like a VC kind of who's looking at who knows hundreds of companies out there. And I think that sophistication of conversations as I start building relationship with Lightspeed was definitely a big factor. Then on top of that, I think as I got the opportunity to work with Ravi also and then his thesis, that was just mind-blowing essentially. So a big part of the reason that we, from our side at Temporal, I think Lightspeed is just an outlier in terms of developer-led infrastructure companies. And it's just the whole Lightspeed ecosystem have so much experience with developer-led companies who have found a good product market fit and how to take those into really enterprise adoption. This is where we believe Lightspeed is very unique. out there and this is why I'm super excited to be partnering with Anoushka, Yu and Ravi to kind of get us, help us build in the next phase of this journey. We couldn't be more excited as well. I mean, you talked about the platform shift and there are only a couple of companies that really matter in this platform shift that have risen above the noise and what you and Max have built has, you know, you're one of the few of them. And so Temporal is a company that's been top of mind for us for many, many years. And I'm so excited that we finally made this happen. Yeah. You know, when you think about the journey ahead, what's exciting you the most? I am super excited about is so far, if you look at last year, durable execution, we were able to prove. If you look at one of the, some of the largest labs like OpenAI. are customers of Temporal. We have the next generation agent tech companies who are building platforms like Rapplet, Lovable. They are building systems on top of Temporal. Then even the application layer, which is starting to show up like Abridge or Evenup are really good examples of how they are able to find able execution and use it in a pretty meaningful way. for delivering value to their customers. I think what I feel durable execution in the form of AI has found their ideal application. That really excites me. We are in the path of business transactions, like every snap story, like, or every Coinbase transaction, or like every Yum Brands, KFC, Pizza Hut, Taco Bell, every time you place an order, it's a temporal workflow. And since we are building a consumption-based business, the business would be order of magnitude of those business transactions. So there was those category. Then we ran into Snap, where every Snap story, which is an internet scale, and you can imagine New Year's Eve, how many Snap stories being posted every second, essentially. So that is what drove me and Max, can we deliver durable execution at that scale? And we were able to kind of, if you look at it, we were able to deliver that like amazingly well. Then companies like OpenAI now have shown like these, I think now there is, first there was business transactions, then internet scale. Now with AI, we are talking a completely different scale, order of magnet scale. It is completely insane. It's mind blowing the scale and volumes that we are seeing. And I think that a couple of years ago, I think told Max, Max, I think we have over invested. into the scale characteristics of temporal cloud. That was the conversation I actually was having with Max. And now Max puts it back on me. Now we are having conversation. We need to figure out how to do 10x now, essentially, or 100x of that. And I think we are in just such an exciting period of this platform shift and how durable execution is so uniquely positioned to basically drive that shift is super exciting time for us. Awesome. I mean, Summer, thank you so much for joining me today. It's incredible to kind of hear your thinking on the business, where you guys are going, and how ubiquitous temporal is kind of becoming in the ecosystem. So really, really appreciate it again. Thanks. Thanks a lot, Anushka, for having me on the show. And then, yeah, I'm really excited to partner with Lightspeed in the next phase of our growth. We could not be more excited.

Learn more about Temporal at https://temporal.io/

Temporal co-founder and CEO Samar Abbas joins Lightspeed partner Anoushka Vaswani to unpack why Durable Execution  has become foundational infrastructure in the AI era. As agentic systems move from demos to production, Samar explains why reliability, not intelligence, is the real bottleneck, and why long-lived, asynchronous workflows require an execution authority that can guarantee completion, retries, and state recovery at scale.

Temporal, the open-source platform Samar and his co-founder Max have iterated on for nearly two decades, has become the orchestration backbone for distributed systems and AI workflows, from OpenAI’s Codex and Replit’s agent platform to large-scale data pipelines and enterprise workloads. They also dig into the company’s open-source-first philosophy, the architectural decisions that built deep developer trust (especially in security-sensitive enterprises), Samar’s leadership evolution, and why the next wave of AI will depend not just on smarter models, but on infrastructure that ensures work actually gets done.

ABOUT THE INVESTMENT MEMO PODCAST
The Investment Memo is a podcast interview series where founders see their original investment memo for the first time and revisit the deal that started it all. The memo becomes the spine of the episode — sparking honest conversation about early conviction, risk, growth and the relationship that formed between founder and investor.


Stay In Touch: 
Website: www.lsvp.com
X: https://x.com/lightspeedvp
LinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/
Instagram: https://www.instagram.com/lightspeed/

Chapters: 
00:00 Show & Founder Intro
01:31 AI Workflow Backbone
02:40 Durable Execution Explained
05:19 Glue Code Problem
09:16 Agentic AI Shift
15:25 Deterministic AI Systems
20:47 Open Source Strategy
26:47 Agent Swarm Future
30:36 Why Software Is Hard
33:30 Founder Role Transition
38:30 Capital & Scale Strategy

The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.