Why observability is part of the architecture
Whether a system is observable is decided at design time — not at the first incident.
What is this? · Perspective
A reasoned position — with a thesis, the trade-offs and the case in which we would decide differently. Not a neutral overview, but a stance. Go to overview
What this is about.
Logging, metrics and traces are often treated as an operations concern to be retrofitted later. In practice, observability is hard to add after the fact: what was never conceived as an event won't exist as a signal later on.
Why we see it this way.
You can only operate what you can see
A system whose failure cases produce no signals is flying blind in production. The question “why was this request slow?” has to be answerable before anyone asks it.
Observability is a design property
Correlation across services, meaningful traces and structured events come from decisions in the code — not from a dashboard bolted on afterward.
It changes how you build
Teams that factor in observability from the start draw boundaries and events differently — usually more clearly.
What it costs.
Observability has a cost: instrumentation, storage and the discipline to curate signals instead of emitting them indiscriminately. Too much telemetry is its own kind of noise.
For very small, short-lived systems, the effort can outweigh the benefit.
When we would decide differently.
For a throwaway prototype or a short-lived script, we deliberately forgo deep instrumentation.
Where data protection requires it, we reduce telemetry selectively and accept less insight.
Where we stand.
We treat observability as part of the architecture, not as an accessory. A system you can't understand in production isn't finished — it just isn't visibly broken yet.
Continue your engineering journey.
Related concepts, decisions, playbooks and perspectives — as one connected path, not a list of links.
Concepts
See it differently? Let's talk.
Perspectives are there to be challenged. If you see it differently — or have a concrete problem — talk to our management.
