Rolling Release

Software has traditionally been sold or distributed in versions. Every few years a new edition appears, and before that developers accumulate their changes. With Rolling Release, these large editions don’t exist. Instead, individual improvements flow to users continuously, often several times a week. The program formally remains the same, but it constantly changes a little bit. The name comes from the English word “to roll”: the software keeps rolling on, without stopping at fixed version numbers.

Why companies are moving away from big version jumps

Large versions have an unpleasant side effect. A finished improvement often has to wait months until the next edition is released. During this time, competitors may already have long had the feature. Those who deliver continuously, on the other hand, bring a new feature to users just a few days after its completion.

Security also speaks in favor of this model. A discovered security vulnerability is a malfunction that attackers can use to break into a system. Such vulnerabilities must be closed quickly. In a continuous delivery model, the path to doing so is already in place and is used every day anyway.

The price for this is a loss of reliability. A company managing a thousand computers wants to know exactly which program versions are running. If everything is constantly changing, this becomes difficult. That’s why many companies opt for a middle path: security problems are fixed immediately, while new features only arrive at fixed dates.

What lies behind the constant stream of updates

A largely automated testing process is a prerequisite. Every change made by a developer is run through test programs before anyone even sees it. If a test fails, the change is rejected. Only what passes this check goes into the delivery.

A second building block is staged rollout. A new feature first goes to one percent of users, then to ten percent, then to everyone. If the measurements in the small group show problems, the change is rolled back. The rest of the users never notice anything.

A common misconception is that Rolling Release means lower quality. The opposite is more likely true. Small changes are easier to oversee than a package of hundreds of new features. If something breaks after an update, the cause is usually found quickly. The model only becomes risky without the automation described above running in the background.

From the browser to the chatbot

The best-known example is the web browser. Chrome and Firefox update themselves in the background without anyone clicking anything. Version numbers rise on a weekly basis and practically no one cares anymore. Most apps on smartphones work the same way.

In the Linux world, there are operating systems that work on this principle, such as Arch Linux. You install it once and then keep updating it forever. The counter-model is Debian, with fixed editions maintained for years. Both approaches have their merits, depending on whether currency or stability is more important.

With AI services, the term becomes somewhat blurred. A chatbot in the browser is continuously refined without anyone noticing. The underlying language model, on the other hand, still receives version numbers, because its training is a completed process. In the news, therefore, the word is usually read in connection with the interface and the tools, less often in connection with the model itself.