Temporal reveals a serverless option for its Durable Execution platform

We’ve all dreamed at one point or another about having safety guards on everything we do — such as keeping erratic drivers away in city traffic, avoiding illnesses, and making sure the food we eat doesn’t make us sick. Now there are real-time guardrails in software development that ensure all code works and doesn’t crash, no matter its state.

Durable Execution is a software development framework that makes code fault-tolerant by automatically persisting its state, allowing long-running processes to resume exactly where they left off after crashes, network failures, or restarts. It converts fragile code into crash-proof workflows by recording steps, ensuring reliability without manual error-handling logic. AI can be fallible, and these infrastructure guardrails provide the foundation needed for production to work efficiently. 

In an era of AI-driven applications, small, medium, and large language models, and increasingly heavy production workloads, this relatively new innovation in IT infrastructure is quickly becoming a go-to problem-solver. It’s like having a fully trustworthy, elastic, lifesaving net beneath all the computational activity of an enterprise IT system at all times.

“Production applications require a certain level of resiliency, scale, and observability.”
—Temporal CTO Maxim Fateev

Speaking this week at his company’s Replay 2026 conference in San Francisco, Temporal co-founder and CTO Maxim Fateev made safety a key message.

“Production applications require a certain level of resiliency, scale, and observability,” Fateev said at the conference, which drew more than 2,000 developers. “We invented this new abstraction in which we just preserve the full state of code execution all the time. This means that if any process crashes or another type of failure occurs, resurrected processes are executed in the same state and continue executing thereafter. 

“For example, you can write a function that runs for a year, and we can guarantee that this function will not die, because we preserve its state all the time.”

Temporal, founded in 2019 atop the open-source Cadence workflow orchestration engine originally developed at Uber, is growing rapidly, with more than 1,500 paying customers (including Nvidia, Netflix, Snap, and Stripe) and thousands more open-source users. 

Common usage examples

• Order processing/e-commerce: A multi-step purchasing process (inventory check, payment, shipping) that finishes correctly even if the application server restarts mid-process.
• Long-running data pipelines (ETL): Processing large data streams where individual steps may fail or take hours to complete.
• Agentic AI workflows: Complex AI agent actions involving multiple steps, database reads, and external API calls that can last for minutes or hours.
• Distributed transactions (Saga Pattern): Coordinating distributed microservices where, if one step fails, the system runs compensation logic for previous steps.

Key facts about Temporal

• Temporal ignores business errors: Temporal only handles infrastructure or platform failures (such as a process crash). Business logic errors (such as not having enough money in an account) are still up to the developer.
• It is mostly AI-agnostic: Temporal doesn’t have any special AI features. It simply makes any application, including AI applications (such as agentic loops or LLM-generated code), durable. AI code is merely code to Temporal, and it makes it reliable.
• LLM code loves Temporal: LLMs are great at writing business logic, but they’re bad at making code scalable and resilient. Temporal is the ideal target platform for LLM-generated code because it addresses all resiliency issues.
• Who uses it? Everyone, from tiny startups to huge hyperscalers and traditional customers, such as banks. HashiCorp and DataDog are well-known users for infrastructure automation.
• Business model: Temporal offers a cloud/SaaS backend cluster hosting service on a consumption-based model. The company guarantees full compatibility between the open-source model and the cloud versions. This reflects the company’s belief that every engineer should have access to the tools needed for successful, reliable code. 

News from Replay 2026

The Seattle-based company announced some innovation news at the conference that IT managers and developers will want to hear.

Temporal’s Serverless Workers enable users to run standard Temporal Workers on serverless compute platforms such as AWS Lambda. There are no servers to provision, no clusters to scale, and no idle compute to pay for, Fateev said. Temporal invokes the Worker when tasks arrive, and the Worker shuts down when the tasks are done.

Serverless Workers use the same Temporal SDKs (software development kits) as traditional long-lived Workers. Users register workflows and activities the same way. The difference is in the lifecycle: instead of running a long-lived process, Temporal invokes the Serverless Worker on demand when tasks arrive. The Worker starts, polls for available tasks, processes them, and exits when the task is done.

For a deeper look at how serverless invocation works, go here.

During the opening keynote, co-founder and CEO Samar Abbas introduced Workflow Streams for real-time observability and Standalone Activities for durable job processing, both aimed at helping developers reliably move AI systems into production. 

Abbas also described a possible partnership with OpenAI. That company’s VP of Application Infrastructure Venkat Venkataramani explained from the stage that “Temporal’s durable orchestration framework is critical to handling our massive scale, complex agentic workflows, infrastructure control plane, and data pipelines, reinforcing the platform’s importance for the next generation of AI products.” 

Standalone Activities enables Temporal Activities to run on their own, not just as steps inside a workflow. Standalone Activities let users improve the durability and debuggability of job processing, and eliminate complex queuing and retry logic by adopting the same model users already know. When a use case outgrows a single step, developers can leverage that same activity inside a workflow with no rewrite required.

Standalone Activities are now available in public preview for the Go, Python, and .NET SDKs, and in pre-release for the Java and TypeScript SDKs, and supported in Temporal Cloud.

Workflow Streams: This is durable streaming that uses Temporal’s Signal & Update primitives. These are mechanisms for interacting with running workflows. Signals are asynchronous, reliable messages used to send data to a workflow without waiting for a response. Updates are synchronous, blocking calls that allow mutating workflow state and receiving a return value, replacing complex Signal/Query patterns 

With Workflow Streams, users get token batches and application-level updates to power responsive UIs, live monitoring, and guardrails. Workflow Streams is designed for real-time user output while preserving Temporal’s reliability model. This is now in public preview.