Knowledge baseSoftware

CI/CD without buzzwords: why a release is not an event

1 September 20264 min read

"Deploying often? That sounds risky." It is the most common reaction to CI/CD, and it has things exactly backwards. Risk grows with the size of a release, not with its frequency. The release that puts months of work live in one evening is the risky one; the twenty-line change that goes live the same afternoon is the safe one. CI/CD is not a buzzword but the technique that makes the second possible: every change takes the same controlled route, and that is why a release with us is no longer an event.

What the pipeline does

For every change, small or large, the same things happen automatically:

  • Building on a fresh environment — the application is built from zero, without prior knowledge or leftover residue.
  • Testing — the test suite runs in full, every time. Not "just the important bits quickly", because there is no reason to: it costs no human time.
  • One artefact — what was tested is exactly what gets deployed. There is no manual step in between where something could still diverge.
  • Deploying by the description — the application's declarative description is executed; nobody has to "quickly go into the server".

The urgent fix and the big new feature take the same route. That is the point: there is no separate, less careful path for when it has to be fast.

Why big releases are the risky ones

A release of forty changes has forty suspects when something goes wrong, and rolling back means rolling back forty things, including the thirty-nine good ones. A release of one change has one suspect, and error tracking tells you within minutes whether it is misbehaving. Small steps do not just lower the chance of a fault; above all they make the answer to "what was it?" trivial.

Then there is the human side. A rare release is an event: an evening kept free, someone on standby, a runbook. Events get postponed, and postponement makes the next release bigger and therefore scarier. That is the spiral in which "we'll release next month" becomes the norm and every release feels a little more like a jump.

What you notice as a client

A reported error can be fixed and live the same day, instead of waiting for "the next release". A small request stays a small request, because there is no release overhead wrapped around it. And Friday afternoon is not a forbidden time slot: a change that has passed the pipeline on a Friday is no riskier than one on a Tuesday.

From our own kitchen

This website works the same way. Every knowledge base article, this page included, goes through a pipeline as a commit, with checks on metadata and both language versions before anything goes live. Not because an article is exciting, but because a route that is always there is also there when things are exciting.

Releases nobody notices

The best release is one nobody noticed, except the person waiting for the fix. That is how we set up software projects by default, and how we take on existing projects that still release per event. Curious what that would take for your application: get in touch.

Frequently asked questions

Frequently asked questions

Does continuous deployment mean things constantly change for my users?

No. Deploying and making visible are two different things; most changes are small and invisible, and what is visible can be made visible on a planned moment. The difference is that the technology no longer has to wait for a collection moment.

What happens when a test fails?

Then the change gets no further than the pipeline and never reaches production. That is not an incident but the system working: the error was caught at the cheapest possible moment, before any user was affected.

Is this not mainly something for large teams?

Rather the other way round. A large team can compensate for missing automation with people and procedures; a small team cannot. The pipeline is the colleague who checks every change, every time, without haste and without Friday-afternoon fatigue.

Answer not found?

Ask an engineer directly — we usually respond within one business day.