Investing in Temporal

Over the last two years, AI has amazed us with its ability to write, design, summarize, and code. And now, as AI systems begin to take the next step – proactive action in real-world contexts – the next big challenge is code execution. 

Agents are growing up from generating output to taking action. They’re moving data, writing and deploying code, updating records, and coordinating complex multi-tool, multi-step processes. It’s an exciting time; but if you’re responsible for code running correctly every single time, it’s also a nerve-wracking one. 

Reliable code execution isn’t a “nice-to-have” when agents are touching real systems of record, or taking hard-to-reverse actions. If an agent hallucinates a memo, that’s annoying. If it retries a payment incorrectly, that’s bad. If it screws up your customers’ business, that’s catastrophic.

Durable execution is higher stakes than ever. That’s why a16z is thrilled to lead Temporal’s Series D. 

Temporal is the code execution layer for critical workflows. Leading AI labs and AI-native companies, like OpenAI, Replit, Lovable, Abridge, and Anthropic all rely on Temporal. The whole AI ecosystem is converging on Temporal as their durable execution layer for both AI and non AI workflows, and we believe they’re already on their way to becoming a load-bearing component for all serious software engineering. 

“But it worked in the demo!”

The classic problem for agentic AI efforts is stalling at the pilot stage. The demo works, but production is unforgiving. Retries break logic, state disappears between steps, external APIs fail, and brittle orchestration code collapses under real-world conditions. After all, agentic systems are long-running and non-deterministic by design. They rely on probabilistic models, span multiple services, and often execute over minutes, hours, or even days. As coding agents move from autocomplete to multi-hour planning and execution loops, and as enterprise agents take on real operational responsibilities, orchestration complexity compounds. This isn’t traditional request-response software; it’s durable execution over time. 

This is already a familiar problem to anyone who watched software become more distributed and event-driven during the cloud era. Temporal got its start with solving these mission critical problems for huge enterprises and leading digital natives like JPMorgan Chase, Airbnb, Snapchat, and Stripe. Now that agents are taking on real responsibilities, agentic AI amplifies every distributed-systems challenge at once: reconciling financial transactions, updating CRM and ERP systems, orchestrating support workflows, and writing and testing production code. 

Temporal abstracts away that complexity by handling retries, state management, orchestration, recovery, and replay automatically, allowing developers to focus on business logic instead of the difficult plumbing of huge distributed systems. Developers define workflows in code using Temporal’s SDKs, and Temporal guarantees that each step executes correctly even in the presence of failures. If a service crashes, execution can be resumed exactly where it left off. If logic evolves, workflows can be replayed deterministically. State is durable, retries are safe, and recovery is automatic. For long-running agents operating over extended horizons, the durability that Temporal provides is the difference between a compelling demo and a production system. The underlying execution layer has become a central piece of the emerging AI agent stack.

And as agents take on longer and more autonomous workloads, they’ll be relying on Temporal to provide a durable execution layer. OpenAI uses Temporal to power ChatGPT Images, ensuring millions of image generation workflows run reliably to completion. Replit relies on Temporal to orchestrate coding agents that plan, write, run, and debug code over extended sessions.

Temporal: for certain

There are no two people better-suited to solve this problem than Temporal’s cofounders Samar Abbas and Maxim Fateev, who have spent their careers at the frontier of distributed systems — from building open source Cadence at Uber to leadership roles across Microsoft and Amazon, where they worked on the underlying Durable Task Framework that later powered Azure Durable Functions and on Amazon’s Simple Workflow Service. Their obsession with correctness, resilience, and developer ergonomics shows in every layer of the product.

We’ve had the privilege of getting to know them over the past five years. What stands out about this team is not just their technical depth but their long-term conviction: the team has been working on workflow execution since before AI was a big deal, and serve a range of customers that predate the agent age. Some founders feel uniquely prepared for the company they’re building. Samar and Maxim embody that rare alignment of experience and the opportunity to meet the moment. 

Every era of computing introduces new foundational layers. The cloud required new security platforms; mobile required new application architectures; the AI-native era requires infrastructure that can safely execute autonomous systems over long time horizons. As long-running agents become a primary driver of enterprise value, the execution layer beneath them becomes indispensable. Temporal wasn’t built in reaction to generative AI; it was built to make complex systems durable. But the agentic era has made that need undeniable.

Temporal is defining that layer, and we’re thrilled to partner with Samar, Maxim, and the entire team as they build the execution backbone for software that acts.