The best developer infrastructure companies don’t get built because someone ran a market analysis. They got built because an engineer hit a wall, couldn’t find what they needed, and decided to make it themselves. Temporal is a prime example of that.
Samar Abbas and Maxim Fateev exemplify that. The idea behind Temporal didn’t arrive fully formed. It evolved through more than a decade of iteration: first at Amazon with the Simple Workflow service, then at Microsoft, where Samar built the Durable Task Framework that became Azure Durable Functions, then at Uber, where they built Cadence and proved the concept at massive scale. By the time they started Temporal in 2019, they were on their nth version of the same core insight. And they had never once been asked to build it from the top down. Every time, it was bottom-up. Engineers who needed this thing to exist, building it because nothing else was good enough.
That combination of deep founder-market fit, years of earned conviction, and a business that has grown into one of the most important pieces of infrastructure in modern software is the reason why Madrona has been a long-term partner to Temporal. It’s why we invested early, and it’s why today I’m proud to share that we’ve continued to back Temporal through their $300 million Series D at a $5 billion valuation, led by Andreessen Horowitz with participation from Lightspeed, Sapphire, and returning investors Sequoia, Index, Tiger, GIC, and Amplify.
A problem we know personally
When Samar and Max first walked us through what Temporal does, it resonated very personally. Earlier in my career at Microsoft, I spent years on .NET Framework and Windows Workflow Foundation to help developers build orchestrated, stateful systems without having to hand-wire every failure mode themselves.
What Temporal has built is, in a very real sense, what those early efforts were trying to become. Developers write code the way they normally would. Temporal takes care of what happens when machines crash, services time out, or processes stall mid-execution, automatically and durably, without developers having to build and maintain that recovery machinery themselves. As Max described it to us: you can use a debugger, use breakpoints, and write normal code. The complexity of the distributed system is hidden underneath. That’s the abstraction that changes everything.
Why agentic AI makes this unavoidable
We’ve spent years at Madrona writing about the rise of intelligent applications, software that doesn’t just deliver features but drives outcomes, adapts in real time, and acts on behalf of users. That vision is arriving, but not in the clean way anyone expected. Agentic AI systems run for hours, days, and sometimes even weeks. They call external services, branch into parallel paths, manage state across long time horizons, and operate in environments that will fail in unpredictable ways.
The result is that most agentic AI efforts stall before they reach production. Not because the models aren’t capable, but because the infrastructure around them doesn’t hold. Retries fail silently, and state gets lost. A workflow that worked in a sandbox falls apart under real-world conditions.
These aren’t new problems. They’re the same challenges distributed systems engineers have been wrestling with for decades, now suddenly unavoidable for every team building with AI. Temporal has been solving them for years. Building production-grade infrastructure of this kind takes time, iteration, and hard-won lessons that can’t be shortcut. Samar and Max have those lessons: they’ve earned them across five companies and more than a decade of work.
What the numbers are actually telling us
Temporal’s growth continues to validate their core thesis and the quality of the product they have built. Revenue is up more than 380% year over year. Weekly active usage has grown 350%. Monthly installations have surpassed 20 million.
These numbers are a signal of developer love as much as enterprise adoption. Over 11,000 organizations use Temporal today. Every major AI lab runs on it. So do global enterprises like ADP, Nordstrom, Yum! Brands, and Abridge. More than 9 trillion actions have been processed through Temporal Cloud. That represents millions of moments where a system recovered correctly, a transaction completed, and a workflow didn’t silently fail.
When OpenAI experienced a sudden spike to 150,000 actions per second with no advance notice, Temporal absorbed it. When a major cloud region went down, Temporal customers kept running. Founders who have lived through those failure scenarios, across five companies and a decade of iteration, have been obsessively preparing for this product and this moment. And we couldn’t be more impressed.
What comes next
We’re still early. Agentic AI today looks something like the first generation of personal computing, powerful in isolation, but not yet the distributed, interconnected reality it will become. As agents get more complex, longer-running, and more deeply embedded in consequential workflows, the execution layer they run on becomes more critical, not less. Temporal is building toward that future with Nexus, Serverless Execution, and Large Payload Storage, capabilities that will make it meaningfully easier for any team to build agents that actually work in production.
Samar and Maxim didn’t build Temporal for the AI wave. They built it because they knew this problem needed to be solved, and they kept building until they had something that genuinely solved it. That’s the kind of company we want to back, and one that tends to define categories rather than just participate in them.
We’re proud to continue this journey with Samar, Maxim, and the Temporal team.
The post Founders Who Built Until the World Was Ready: Quadrupling down on Temporal appeared first on Madrona.