Airbnb has open-sourced Skipper, the internal workflow engine that keeps its insurance claims, payments and media processing running when servers fail. Airbnb Skipper is now available under the permissive Apache 2.0 licence, so any company building on Java or Kotlin can adopt it without licence fees.
Airbnb says Skipper has run in production for more than a year. It supports over 15 use cases across insurance, payments, media, infrastructure, incentives and wallet teams, and it has peaked at 10,000 workflows per second.
What Skipper does
Engineers Andriy Sergiyenko and Ricardo Gamba call Skipper an embedded workflow engine. A developer writes a multi-step business process as plain Java or Kotlin classes, and the engine stores its progress and restarts it after crashes or temporary faults.
The engine lives inside the host service and shares its dependencies. No central server is needed: every team’s application carries its own copy and keeps state in a database the team already runs.
Database choice is narrow at launch, with MySQL and SQLite supported and Postgres and DynamoDB still to come.
The problem it replaces
Plenty of business processes are slow and have many stages. Take an insurance claim: someone has to check it, screen it, estimate the damage, pay out and confirm, and a human reviewer may hold things up for days. Certain payment operations last weeks.
A server that restarts mid-process can leave the company unsure what already happened. According to Airbnb’s engineers, each team had built its own workaround. Those were costly and hard to test, kept producing duplicates and failed recoveries, and usually relied on brittle nightly reconciliation jobs to tidy up.
The pitch is simple: define the process once as steps and let the engine see it through. Every completed step is saved, so a process that gets interrupted resumes from that point. Airbnb says the gain is biggest for teams that would otherwise write bespoke retry jobs, state tables and polling loops each time.
How it compares with Temporal and others
Skipper enters a crowded field. The approach, called durable execution, is well established. Temporal leads it, with Stripe, Netflix and Datadog among its customers. AWS sells a managed equivalent, Step Functions, at $25 per million steps. Restate, a Berlin company started by people behind the Apache Flink data platform, raised $20 million in September, betting that AI agents will make durable execution more in demand.
The nearest comparison is DBOS Transact, an open-source library that likewise embeds in an application and records progress in the company’s own Postgres. It targets Python, TypeScript and Go. Skipper serves Java and Kotlin, which many banks and big enterprises rely on.
For its most sensitive services, Airbnb rejects a shared central engine. If that system fails, everything depending on it halts. Managed cloud services remove operational work, Gamba said in an April 2026 post on the project, but they bring vendor lock-in, data-handling requirements beyond a company’s own infrastructure, limits on execution and the same dependence on an outside system.
What it means for engineering teams
For decision-makers the draw is cheaper and safer operation: nothing is billed per transaction and no supplier can change terms. A process defined in a single place is easier to inspect, which regulated industries value. Airbnb adds that developers spend their time on business rules instead of boilerplate meant to survive failures.
The catch is capability. Skipper suits teams with a strong operational setup, and not every company has Airbnb’s engineers or scale. Temporal’s enterprise business is unlikely to feel it soon. Java shops weary of rebuilding the same plumbing, however, get a free option that has already carried live payments and insurance traffic.
Airbnb has released code before. Airflow, which it open-sourced in 2015, is now a popular Apache project and underpins Google’s managed Airflow service and Astronomer’s products.
What to watch next
The first thing to track is storage. Postgres and DynamoDB support would open Skipper to many more teams, since MySQL and SQLite cover only part of the market. The second is outside adoption: whether companies beyond Airbnb run it in production and contribute fixes back. Airbnb has published the code and the design reasoning, and the next signs of traction will show up in contributions and in which teams trade their retry jobs for an engine like this.
Source: The Stack