What is an agile release train? SAFe ART explained
7 min read
An agile release train (ART) is a long-lived team of agile teams, usually 50 to 125 people, that plans, builds and delivers one or more products together on a shared schedule. The term comes from the Scaled Agile Framework (SAFe). Every team on the train plans together at the start of each planning interval of 8 to 12 weeks, shows integrated work at the end of every iteration, and reviews the whole interval at its end. The name refers to the timetable: the events happen on fixed dates, while releases to customers can happen whenever the business decides.
| Element | What it is | When |
|---|---|---|
| ART | A long-lived team of agile teams, 50 to 125 people | Permanent |
| Planning interval (PI) | A timebox of four or five development iterations plus one innovation and planning iteration | Every 8 to 12 weeks |
| PI planning | A planning event for the whole train, typically two days | Start of each PI |
| System demo | An integrated view of the features all teams delivered | End of every iteration |
| Inspect and adapt | PI system demo, measurement review and problem-solving workshop | End of each PI |
| Release | Making the work available to customers | On demand |
What is an agile release train in SAFe?
It is the main unit of delivery in SAFe: several agile teams that share one mission, one backlog of features and one calendar, and that stay together over time instead of being assembled per project. The train contains every skill needed to define, build, test and release its product, so it does not hand work to other departments to finish.
SAFe’s page on the agile release train defines it as “a long-lived team of Agile teams that incrementally develops, delivers, and often operates one or more solutions,” and gives the usual size as 50 to 125 people. The page describes ARTs as cross-functional, with “all the capabilities needed to define, build, validate, release, and, where applicable, operate” the solutions they own.
Who is on an ART?
Mostly the agile teams themselves. Around them, four roles work at the level of the whole train: Product Management, the System Architect, the Release Train Engineer and the Business Owners.
Product Management manages and prioritizes the ART backlog, so it decides what the train builds next. The System Architect defines and communicates a shared technical vision for the systems the train builds. Business Owners are the stakeholders with primary responsibility for return on investment, governance and compliance. SAFe describes the Release Train Engineer as “a servant leader and ART coach who facilitates ART events and processes,” and you can think of it as a scrum master for the whole train: it runs PI planning, tracks flow across teams and pushes improvement.
How does an agile release train work?
It runs on a fixed rhythm. A planning interval opens with PI planning, continues through several short iterations in which each team builds and integrates its work, and closes with a review of the whole interval. Then the next interval starts.
SAFe’s article on the planning interval calls it “an 8 to 12-week timebox,” usually four or five development iterations followed by one innovation and planning iteration, and presents that pattern as a suggestion. Before SAFe 6.0, a PI was a “program increment,” as a Scaled Agile blog post notes, which is why older material uses that name.
The events follow from that rhythm:
- PI planning. SAFe’s PI planning page calls it “typically a 2-day event” for the entire train. It ends with committed objectives from each team and an ART planning board that shows feature delivery dates and dependencies between teams.
- Iterations. Each team plans and runs its own short iterations inside the interval.
- System demo. At the end of every iteration, the system demo gives stakeholders “an integrated view of new features” delivered by all the teams on the train.
- Inspect and adapt. At the end of each PI, inspect and adapt combines a PI system demo, a review of quantitative and qualitative measures, and a retrospective with a problem-solving workshop.
Does an ART release on a fixed schedule?
Development runs on a fixed cadence, and releases need not follow it. In SAFe the planning dates are fixed and the release decision is separate, so a train can release several times within one planning interval, or hold finished work until the business is ready.
SAFe’s Release on Demand article separates releasing from the work of deploying, and treats release on demand as one of the four aspects of its continuous delivery pipeline. That is the same split between shipping code and exposing it to users that release vs deployment explains in detail.
What is the difference between a release train and an agile release train?
A release train, outside SAFe, is a shipping schedule: a product releases on fixed dates with whatever is ready. An agile release train is a group of people organized around a product. A company can have either one without the other.
Browsers offer a well-known example of a plain release train. The Chromium team announced in 2021 that Chrome would release a new milestone every 4 weeks, starting with Chrome 94, and added an Extended Stable option with milestone updates every 8 weeks for enterprise administrators and Chromium embedders who need more time. Some mobile teams ship on a weekly train too. The release management process article compares scheduled trains with continuous deployment and change-managed releases.
How should an ART handle release notes?
Write each entry when the change merges, publish it when the feature is released, and give one person the final say on what customers read. Collecting notes at the end of a planning interval turns them into guesswork, because ten teams’ worth of work is summarized from memory weeks after it was done.
Three practices keep notes accurate on a train:
- Draft at merge, per team. Each team writes or approves the entry for its own change while the context is fresh. Changeloop works this way by default: every user-facing pull request that merges turns into a draft for someone on the team to approve (a per-repository auto-publish setting skips that step), so drafts pile up during the iteration and get edited, not written from scratch. The docs describe where approved entries are published.
- Review at the system demo. The demo already gathers the features finished in the iteration, which makes it a natural moment to check that every customer-visible change has an entry.
- Publish on release, not on PI end. If a feature is released in week three, its note goes out in week three. Product Management usually owns the public wording; the release notes template gives a structure that works across teams.
Trains also need notes inside the company, because teams on the same train depend on each other’s changes. Internal release notes covers what to tell support, sales and neighboring teams, and how to write release notes covers the customer-facing side.
When is an ART the wrong fit?
When there are not enough teams to coordinate. An ART exists to align several teams that depend on each other, so a single team, or two teams with separate products, gets the overhead of the events without the benefit.
The events cost real time: a two-day planning session every 8 to 12 weeks for up to 125 people, plus demos and the inspect and adapt workshop. That cost pays for itself when dependencies between teams are the main reason work is late. A small team that deploys continuously and releases behind flags usually gets more from a light release process than from a train.
FAQ
How long is a planning interval? SAFe describes it as an 8 to 12-week timebox, usually four or five development iterations followed by one innovation and planning iteration. The exact pattern is a suggestion, and trains adjust it.
How many people are on an ART? SAFe gives the typical size as 50 to 125 people, organized into several agile teams plus the train-level roles of Product Management, System Architect, Release Train Engineer and Business Owners.
What does a release train engineer do? The release train engineer coaches the train and facilitates its events and processes, including PI planning. The role also helps optimize flow across teams with metrics and drives improvement, much like a scrum master working at the level of the whole train.
Does PI planning have to be in person? SAFe prefers it, saying that where possible everyone is physically together. Its PI planning page also says virtual planning has proven effective when physical presence is not possible, provided the whole train takes part.
The technical claims in this article have not been independently reviewed. If something here is wrong, tell us and we will correct it.