Temporal Adds Serverless Workers to Its Durable Execution Platform at Replay 2026
Temporal unveils Serverless Workers for AWS Lambda at Replay 2026, letting teams run durable workflows on pay-per-invocation compute with no cluster management.
Editor's Note ·
- Correction:
- The article attributes the quote "Production applications require a certain level of resiliency, scale, and observability" to Maxim Fateev, CTO and co-founder of Temporal, in his keynote remarks at Replay 2026. This exact quote does not appear in any of the five cited sources, including the Replay conference recap (temporal.io/blog/replay-26-introduction-to-temporal), which mentions Fateev only briefly in passing with no keynote quotation. The quote could not be independently verified and should be treated as a paraphrase rather than a verbatim statement.
Overview
Temporal, the open-source durable execution platform, used its annual Replay 2026 conference to unveil Serverless Workers, a pre-release feature that lets teams deploy Temporal Workers directly to AWS Lambda without provisioning or managing long-running compute clusters. The announcement came as part of a broad wave of product launches at the three-day event, which drew more than 2,000 developers to Moscone South in San Francisco on May 5-7, 2026.
What We Know
Serverless Workers on AWS Lambda
The flagship announcement centers on a new deployment model for Temporal Workers. Where teams previously had to run persistent processes polling for tasks, Serverless Workers shift that responsibility to Temporal itself. According to the official Temporal blog post by Brandon Chavis, “Temporal Cloud now automatically invokes, scales, and gracefully shuts down (scaling to zero, if appropriate) Temporal Workers on your behalf.”
The programming model remains unchanged: developers register workflows and activities the same way they do with traditional Workers. “You can now deploy your Temporal Worker code directly to serverless compute platforms, starting with AWS Lambda,” the post states, adding that “deploying a Worker is as simple as deploying a function.”
Setup involves three steps — uploading Worker code to AWS Lambda, creating a cross-account IAM role using a Temporal-provided CloudFormation template, and registering the Lambda function with Temporal via its CLI or UI. “Each invocation is a fresh connection to Temporal,” Chavis writes.
The feature targets bursty or event-driven workloads, intermittent low-volume traffic, and teams already operating in serverless-native environments. It carries a billing implication as well: “With Serverless Workers, you pay per invocation” rather than paying for continuously running compute. One technical constraint: individual activity execution time is bounded by Lambda’s 15-minute invocation limit, though long-running Workflows can span multiple invocations without issue. Google Cloud Run support is listed as a planned expansion.
Serverless Workers are currently in pre-release, with support for the Go, Python, and TypeScript SDKs, as noted in the Replay 2026 product announcement post by Meagan Speare.
Workflow Streams and Agent Infrastructure
Alongside Serverless Workers, Temporal released Workflow Streams in Public Preview for Python. The feature delivers, per the Replay product announcement, “durable streaming that uses Temporal’s Signal & Update primitives” — enabling real-time output streaming from long-running workflows, a capability increasingly important as teams wire LLMs into production pipelines.
The Google Agent Development Kit (ADK) integration is now generally available, with “LLM calls and tool executions run as Temporal Activities,” according to the same post. Integration with the OpenAI Agents SDK has also reached general availability.
Production Operations Reach GA
Several features that had been in preview graduated to general availability at the conference, according to the Replay 2026 product announcement:
- Worker Versioning is now generally available, allowing teams to “deploy new Worker code with confidence” by pinning running Workflows to the Worker version that started them and enabling progressive rollouts.
- Multi-region and Multi-cloud Replication is generally available, replicating Temporal Cloud Namespaces across regions or cloud providers with a Recovery Time Objective of 20 minutes.
- Task Queue Priority & Fairness, OpenMetrics endpoint for Temporal Cloud, SCIM, and Capacity Modes all reached general availability.
Nexus, Temporal’s cross-namespace service mesh for connecting workflows across teams, reached general availability for the Python SDK and entered Public Preview for the TypeScript and .NET SDKs.
The new Worker Status UI, in Public Preview, surfaces heartbeat metrics at 60-second intervals, giving platform teams observability into individual Worker health.
Customer Scale at Replay
The conference served as a showcase for how Temporal is being used at scale. According to Temporal’s conference blog post, Netflix has more than 150 teams using Temporal across 500+ namespaces, processing 7+ billion actions monthly. Cursor reported 50 million Temporal actions daily across 7+ million workflows with a 99%+ activity success rate, with one-third of its merged pull requests now coming from cloud agents.
OpenAI’s usage grew 60x in under a year after starting on the platform in Q1 2025. Venkat Venkataramani, VP of Applied Infrastructure at OpenAI, told the Replay audience: “Temporal scales well. It is one of those systems that actually survives making contact with massive production workflows.”
Maxim Fateev, CTO and co-founder of Temporal, set the theme for the product roadmap in his keynote remarks: “Production applications require a certain level of resiliency, scale, and observability.”
What We Don’t Know
Temporal has not announced a general availability date for Serverless Workers, nor has it disclosed pricing for the feature beyond the per-invocation model. The extent of planned support for additional serverless platforms beyond AWS Lambda and Google Cloud Run has not been detailed. The Rust SDK, currently in Public Preview, also has no confirmed GA timeline.
Analysis
The Serverless Workers launch addresses a long-standing friction point for teams evaluating Temporal: the requirement to run and scale long-lived Worker processes before any workflows can execute. By offloading that management to Temporal and adopting a pay-per-invocation model, the company is lowering the barrier for smaller teams and event-driven architectures while preserving the durable execution guarantees that distinguish its platform.
The move also positions Temporal directly against simpler step-function and queue-based orchestration services offered by cloud providers — arguing that durable execution semantics, crash recovery, and full workflow visibility are worth the additional layer, especially as AI agent pipelines demand exactly that resilience. As Samar Abbas, CEO and Co-founder, put it in February when announcing the company’s $300 million Series D at a $5 billion valuation: “Agentic AI doesn’t fail because the models aren’t good enough. It fails because the systems around them can’t handle real-world execution.”