Every engineering era is named after what it pays attention to.

The DevOps era paid attention to deployment. The SRE era paid attention to reliability. The observability era paid attention to signals — logs, metrics, traces — and built a stack of tools to collect, store, and query them.

The next era is starting, and it is paying attention to something different.

The shift is from observing signals to engineering with behaviour.

What Signal-Based Engineering Looks Like

For more than a decade, the centre of gravity has been telemetry.

Instrument everything. Stream it somewhere. Store it. Query it. Alert on it. Build dashboards. Hire people who can read those dashboards.

It worked.

Incidents got shorter. On-call got better. Teams could see when things were wrong.

But the lens stayed the same:

Discrete signals, interpreted by humans, mostly during incidents.

Outside the incident, the signals went mostly unused.

• A developer changing code did not consult them
• An architect planning a migration did not query them
• A leader making a debt decision did not start there

The data was rich and the use was narrow.

What Behaviour-Based Engineering Looks Like

Behaviour-based engineering treats production as the source of truth across the full lifecycle, not just the incident.

It asks different questions, and asks them earlier:

• Before a change: which flows depend on this code?
• During a review: is this method actually used?
• During a migration: which services really talk to each other?
• During planning: where is the dormant code, and where is the hot code?
• Inside an AI agent: what production context should ground this edit?

The answers do not come from a dashboard.

They come from a behavioural view of the system — services, APIs, methods, flows — kept current by what is actually running.

Why Now

Two forces are pushing this shift.

1. Systems Are Harder to Reason About

More services. More async. More dependencies. More turnover on the teams that built them.

Tribal knowledge has aged out.

Diagrams cannot keep up.

2. AI Is Writing More Code

AI agents need ground truth that is not a stale wiki or a code search.

They need to know:

• What runs
• What does not
• What depends on what

Without that, faster code generation just means faster mistakes.

Signal-based tools tell you when something is wrong. Behaviour-based tools tell you how the system actually works.

The Category

This is not a feature on top of observability.

It is a different layer, with different users, different questions, and a different shape.

CodeKarma is building toward this layer.

So are others, in different ways.

The label matters less than the direction.

The next decade of engineering will not be won by whoever ships the most dashboards.

It will be won by whoever helps teams understand their systems the fastest.