Temporal use cases and design patterns
Across industries, Temporal applications fall into a handful of shapes: a transaction that must not half-complete, a process that waits days for a person, an entity that accumulates state for years, a pipeline that fans out across GPUs. This page groups those shapes, names companies running each one in production, and points to the pattern or Guide that implements it.
Why Temporal covers what the platform replaces, and Understanding Temporal covers the vocabulary. Design Patterns is the code-level catalog, and Guides are full implementation walkthroughs built on top of those patterns.
Temporal in production
These are the use cases where Temporal runs at scale today. For more companies and a deeper look at each, see Temporal in use.
Transactions
Operations across two or more parties where a partial completion leaves money or data in the wrong state.
Start with the Saga pattern to compensate for steps that already succeeded, and Non-Retryable Errors to fail fast on a rejection that no retry will fix. See distributed transaction patterns to choose between them.
Business processes
A sequence of steps that ends in a service or product delivered to a customer, often with a person in the middle.
- Bookings at Turo
- Orders and logistics at Maersk
- Marketing campaigns at Airbnb
- Human-in-the-loop at Checkr
Use the Approval pattern for the steps that block on a decision, and Updatable Timer for deadlines that move. The reliable document approvals Guide walks through a full implementation.
Entity lifecycle
A long-lived business entity — an account, a device, a loan application — that accumulates state over months or years.
Give each instance one Entity Workflow and drive every state transition with a Signal or an Update, then use Continue-As-New to keep Event History bounded. The loyalty points and gaming sessions Guides show the pattern end to end.
Operations and infrastructure
Repeatable internal automation: provisioning, deployments, and scheduled maintenance.
Run recurring work on a Schedule, wait on external state with the Polling pattern, and route work to the right hardware with Worker-Specific Task Queues. The route specialized workloads and temporary rate limit increases Guides cover the routing and capacity cases.
Data and AI pipelines
Multi-step extraction, embedding, transcoding, or batch inference work that fans out and must resume without reprocessing what already finished.
Fan out with Child Workflows or a MapReduce Tree, page through unbounded input with the Batch Iterator, and cap concurrency with a Sliding Window. When pipelines share Workers across models or tenants, Task Queue Priority and Fairness dispatches urgent work ahead of bulk work and keeps one tenant from starving the others.
AI agents
Agent loops are long-running and stateful, and the calls inside them fail: model requests time out, tools rate-limit, approvals take a day. The loop is a Workflow, each model call and tool call is an Activity, and a crash resumes the conversation instead of restarting it.
- Reliable, observable agents at Lindy
- Long-running, durable agents at Dust
- Account summaries with agents at ZoomInfo
Durable AI is the entry point for this work: runnable AI Cookbook recipes for tool calling, MCP, structured output, and human-in-the-loop, plus the agent framework integrations for each SDK.
Common application patterns
Several shapes recur regardless of industry. Each one below has a reusable implementation in the Design Patterns catalog.
Human in the loop
Some steps can't proceed until a person acts: an onboarding form, an invoice approval, a manual fraud review. The wait is the hard part, because it can last days and the connection between the person and the rest of the system is unreliable.
A Workflow handles this without a callback table or a polling job. It blocks on a Signal or Update, sets a durable Timer for the reminder or the deadline, and keeps its place across Worker restarts in between. See the Approval pattern for the reusable version and the background checks example for a complete application in Go.
Background jobs and long-running work
To run a single unit of work durably, use a Standalone Activity: a top-level Activity Execution started directly by a Client, with no Workflow around it. It gives you retries with backoff, an Activity ID and Run ID you can cancel or query, and deduplication at submit time, without running a broker. This is Temporal's job queue, and Standalone Activities are Generally Available. Teams moving off Celery, Sidekiq, or a homegrown queue can start with the Celery migration Guide.
Jobs that take longer than a broker's poll timeout are where this matters most. A message queue with a five-minute visibility timeout hands the same message to a second consumer while the first is still working; Temporal tracks the Activity Execution itself, so a long job doesn't get duplicated. Report progress with an Activity Heartbeat so a job that dies mid-run is detected before its Start-To-Close Timeout expires, and resumes from its last reported checkpoint. See the Long-Running Activity pattern.
Polyglot systems
Temporal has eight SDKs — .NET, Go, Java, PHP, Python, Ruby, Rust, and TypeScript — and Workflows written in different languages can call each other, so teams keep the language they already use.
The polyglot sample shows Workflow Executions in Go, Java, PHP, and TypeScript sending messages to each other, including how errors propagate across the language boundary. For calls between teams that don't share a Namespace, deploy schedule, or database, Temporal Nexus puts a service contract between them.
State machines
A state machine changes a system's behavior in response to changes in its state. Applying one to a real business process usually means a transition table, a persistence layer for the current state, and a job to detect rows that got stuck — none of which is the process you set out to model.
In Temporal, the current state is the position of a running Workflow. Express transitions as ordinary control flow and hold state in local variables, and Temporal persists it for you. The Entity Workflow pattern is the general form, and entity lifecycle patterns covers the variants. For the longer argument, see the State Machines Simplified whitepaper.
Find an implementation
Design patterns
Reusable Workflow and Activity techniques with runnable code across SDKs, grouped by problem domain.
Guides
Step-by-step walkthroughs for specific use cases, written and verified by Temporal staff.
Durable AI
Agents, processing pipelines, and model training on Temporal, with cookbook recipes and framework integrations.
Code Exchange
Community and Temporal-built sample applications you can run and adapt.
To write your first Workflow, follow a getting started tutorial, then work through the project-based tutorials and example applications.