Skip to content

Airbnb just gave away its Skipper

Is the newly Apache 2.0-licensed software an open-source threat to Step Functions and Temporal?

Airbnb just gave away its Skipper
Image credit: https://unsplash.com/@ourselp

Airbnb has released Skipper, the homegrown software that keeps its insurance claims, payments and video processing on track when systems fail. The company’s engineers open-sourced it this month under a permissive Apache 2.0 licence.

That means that any company building on Java or Kotlin can, if so inclined, adopt the same “lightweight workflow engine” without associated software licence fees. 

Companies that provide similar proprietary enterprise software, like Temporal, may want to be paying attention – with the obvious caveat that not every company has Airbnb’s engineering chops, scale, and small army of capable developers.

(Airbnb says it uses Skipper to power critical infrastructure automation, insurance, media processing, and payments workflows at Airbnb – and has handled peaks of 10,000 workflows per second at the $12 billion+ annual revenue digital native.)

What is Airbnb's Skipper?

Airbnb engineers Andriy Sergiyenko and Ricardo Gamba describe Skipper as an “embedded workflow engine” that “lets you express multi-step business processes as Java or Kotlin classes, persist their progress, and recover from crashes and transient failures. It runs inside your service and reuses its existing dependencies…”

Their approach is highly decentralised. Rather than a central service, it operates inside each team's own application, using a database it already has. The two said Skipper may “appeal to teams that have a robust operational environment and need a workflow engine that adapts to their existing ways of working.” (Storage support is limited to MySQL and SQLite for now, with Postgres and DynamoDB planned.)

It is especially useful, Airbnb’s team said in a recent blog, “when the alternative is maintaining custom retry jobs, state tables, and polling loops for each new process.”

The business need...

Many critical business processes are multi-step: insurance claims need to be validated, screened, estimated, paid and confirmed, sometimes over days and often while waiting on a human reviewer. Some payment operations run for weeks. 

If a server restarts halfway through, the business can lose track of what has already happened. Airbnb's engineers say that teams across the company had each been solving this on their own – with their own automations. The results were predictably expensive, hard to test and kept falling into the same traps around duplicates and recovery, with clean-up often left to fragile nightly reconciliation jobs.

Skipper's promise is simple: describe the business process once, as a series of steps, and the engine makes sure it finishes. Progress is saved after each step, so an interrupted process resumes where it stopped instead of starting over or stalling.

Airbnb says Skipper has been in production for over a year, supports over 15 use cases across insurance, payments, media, infrastructure, incentives and wallet teams, and – as mentioned earlier, hit peaks of 10,000 workflows per second.

How it compares

The idea, known as "durable execution", isn't remotely new. 

Temporal, the best-known option, counts Stripe, Netflix and Datadog among its users. Amazon's AWS Step Functions offers a fully managed equivalent billed at $25 per million steps, whilst Berlin-based Restate, founded by creators of the Apache Flink data platform, raised $20 million in September – building on the idea that AI agent use drives greater need for particular durable execution capabilities. 

As Restate put it: “The longer agents run and the more systems they touch, the more there is to preserve when something goes wrong. It is no longer only the execution that needs to survive, but the state, communication and interactions around it.”

Airbnb’s engineers argues that for its most critical services, a central system is a single point of failure: if it goes down, every dependent process stalls. 

Boo to the cloud...

As for buying cloud services to solve this problem? Spit! “Cloud-managed workflow services eliminate operational overhead but introduce vendor lock-in, regulatory and data-handling requirements beyond our own infrastructure, limits on execution, and the same fundamental concern: a critical external dependency”, as Gamba wrote in an earlier April 2026 blog on Skipper – published before the firm open-sourced it.

Skipper's closest relative is, perhaps, DBOS Transact, an open-source library that likewise slots into existing applications and saves progress in a company's own Postgres database, for languages including Python, TypeScript and Go. Skipper does the same for Java and Kotlin, the tech many banks, and large enterprises run on.

The company has strong open-source form.

In 2015 it open-sourced Airflow, now a widely used Apache project and the basis of Google's Cloud Composer*, as well as Astronomer’s services. (That’s right: The company of Coldplay camera shame fame. Ancient history now, of course.) 

For decision-makers, the appeal (if you have the chops to adopt something like this)is lower cost and lower risk. There is no per-transaction fee and no vendor to rely on. Because each process is written in one place, it is easier to see what it actually does, which helps in regulated industries. Airbnb also says it frees developers to focus on business rules rather than repetitive code written as a resilience safety net. 

Skipper is unlikely to dent Temporal's enterprise business soon, but for Java shops tired of reinventing the same plumbing, it offers a proven, free alternative…

Here’s the repo. We welcome readers’ views > ed@thestack.technology

*We check, whilst dropping a link in, and see that Google has now branded this “Managed Service for Apache Airflow” Not snappy, but it does what it says on the tin.

Add The Stack on Google