The Architecture of Intent: Zones, Boundaries, and Baselines in the Cloud
Security architecture has always started with separation.
Long before cloud, security teams were already thinking in layers, segments, and perimeters. The first firewall was not just a filtering device. It was a way of expressing structure: which parts of the environment were internal, which were external, and what was allowed to move between them.
In other words, it helped establish zones.
That basic instinct has not changed. Security still depends on defining meaningful zones, understanding what belongs inside them, and controlling what can cross between them. It is one of the oldest ideas in security, and still one of the most important.
What changed is the environment.
In older environments, the perimeter was easier to recognize and easier to manage. There were fewer edges, fewer systems, and a smaller number of network paths to control. A lot of the architecture could be expressed through network segmentation and a relatively contained set of devices.
In the cloud, that model became much harder to hold onto. The perimeter did not disappear. It expanded.
Now it lives across network paths, identities, resources, and service-level relationships. A boundary is no longer expressed only through a firewall rule or a routing decision. It may also depend on who can assume a role, which service can access which resource, what can be deployed into production, or which systems are allowed to interact across environments.
That is why cloud security often feels more complex than traditional segmentation, even when the underlying goal is the same.
The goal is still to define zones. The goal is still to establish a perimeter. The goal is still defense in depth.
But now those ideas have to be expressed through many different kinds of cloud-native controls, across multiple providers, and kept true as the environment keeps changing.
The problem isn’t a lack of controls.
Cloud providers already give organizations a large set of security building blocks. The problem is not that the controls do not exist. The problem is that they are fragmented, low level, provider-specific, and difficult to connect back to the architecture the organization is actually trying to build.
A security architect usually does not think in terms of individual primitives.
They think in architectural statements like these: A CNAPP vendor should only have narrow read access into production. A CI/CD system should be able to deploy, but not freely destroy. Internet-originated access should stop at specific entry points. Sensitive applications should live in tighter segments than general-purpose workloads. Non-production should not quietly become a side door into production.
These are architectural statements.
But implementing them usually means dropping into a tangle of identities, network controls, policies, permissions, organization rules, and resource constraints. The organization usually knows the outcome it wants. What is missing is the layer that translates secure-by-design principles into architectural intent, and intent into something operational and durable.
Security still needs zones.
This is why zones matter so much in the cloud. A zone is not just a grouping mechanism. It is a way of making security posture coherent.
A production zone. A non-production zone. An internet-facing zone. A sensitive data zone. A shared services zone. A zone for systems that can deploy into production. A zone for external connectivity. A central security zone. A central FinOps zone. A zone for AI services and model access. A zone for agentic workflows and the systems they are permitted to reach.
These are not just organizational labels. They are security statements. They express that the things inside the zone should share a posture, and that the interactions with other zones should be governed intentionally.
Once you start there, the architecture becomes easier to reason about. Not “what random controls do we have?” But “what are our zones?” What is allowed to enter them? What is allowed to leave them? What standards must hold inside them? Where are the crossings, and which of them are deliberate?
That is the language security teams have wanted all along.
Boundaries define the lanes.
Once zones exist, the next question is how they interact. Every architecture has lanes. Some are expected and necessary. Others are risky and should be narrowed or closed.
Traffic from the internet into a public entry layer is one kind of lane. Access from a CI/CD zone into production is another. A connection from a branch environment into internal cloud systems is another. A third-party service reading specific resources is another. An AI-enabled internal application invoking model services and touching sensitive data is another.
These are not all the same kind of interaction. Some are network paths. Some are identity paths. Some are resource access paths. Some are service-to-service paths.
But from an architectural perspective, they all behave like boundaries.
A boundary is the rule that governs what can cross from one zone to another, or from one actor to another. It defines the crossing, the conditions, and the constraints. In older environments, many of those boundaries lived mainly in the network. In cloud environments, they are distributed across the full control plane.
That’s why the perimeter is harder to see now. It is no longer one line. It is a system of boundaries.
Baselines define the floor inside each zone.
Boundaries are only half the architecture. The other half is the baseline inside each zone.
A production zone may require stronger encryption defaults, stricter deployment constraints, tighter identity assumptions, more controlled egress paths, and higher standards for operational hardening. A non-production zone may allow more experimentation, but still require clear limits. A sensitive data zone may need much stronger data protection baselines than a general shared-services zone.
That is the difference between the shape of the architecture and the standard it must hold. Boundaries define the shape. Baselines define the floor.
Both are necessary if the architecture is going to hold.
A zone with weak boundaries is exposed. A zone with weak baselines is brittle. A secure architecture needs both.
This is where living infrastructure breaks the old model.
New services appear. Identities shift. Resources move. Exceptions accumulate. The environment grows through thousands of small changes, each one reasonable on its own, but together they can blur the architecture.
So instead of asking what the production boundary should be, teams argue about individual controls. Instead of defining the baseline for a sensitive zone, they chase point-in-time findings. Instead of reasoning about the approved lanes between zones, they manage scattered permissions and exceptions.
The result is that enforcement accumulates, while the architecture becomes harder to read.
That is the danger in living infrastructure: the system keeps evolving, but the intent does not stay legible.
You can have many controls and still not be able to clearly answer:
What are our zones?
- What is the perimeter of each?
- What are the approved lanes between them?
- What baseline does each zone need to meet?
- Where are the gaps?
That’s the missing layer from secure by design, from architectural intent to enforcement.
AI agents are now actors in your architecture.
This gap has always existed. What changed recently is that it is now urgent, not just important.
AI agents are now operating inside cloud environments. Not as passive components, but as actors: invoking services, querying data stores, triggering deployments, and calling external model endpoints. They operate at machine speed. They inherit the permissions of whatever identity they run under. And in most organizations, nobody has clearly defined which zones those agents are allowed to operate in, which crossings are approved, or what baseline governs their behavior.
The pattern is familiar. Engineers move fast, as they should. They wire up integrations that work. Nobody is being careless. And then, at some point, there is an environment full of agents that are already running, already crossing zone lines, already load-bearing. Every unapproved integration that works becomes a legacy actor. Fixing that is hard.
The answer isn’t a separate AI security layer on top of an unarchitected environment. The answer is the same answer it has always been: define the zones, govern the crossings, enforce the baselines. Before the agents run, not after. Native is built around that layer.
That's the shift we are building toward in Native.
We're helping organizations define and operate a secure cloud architecture through zones, boundaries, and baselines.
That means giving teams a way to express the architecture in the way security already thinks about it: which actors operate in the environment, which zones exist, which lanes between zones are allowed, which boundaries govern those crossings, which baselines should hold inside each zone, and how all of that is enforced through the native controls already present in the cloud.
This matters because the hard part is not writing down intent.
Most organizations already have an intuitive sense of what they want: production should be harder to touch, sensitive systems should be more protected, external access should be narrow, and trusted systems should still operate within clear limits.
The hard part is turning that into an operating model that survives scale, exceptions, multi-cloud sprawl, and constant change.
From a static perimeter to living architecture.
That is why this cannot just be a static design exercise.
Cloud environments are living systems. New workloads appear. New access paths get created. New services are introduced. New actors show up. Existing actors gain new capabilities. AI gets embedded into systems that did not have it before.
And when it does, it does not arrive as a passive feature. It arrives as an actor: one that calls services, crosses boundaries, and operates under permissions that were configured for a different purpose. The architecture has to account for that actor the same way it accounts for every other one, with defined zones, governed crossings, and enforced baselines.
So the architecture has to stay alive too.
Zones need to reflect the real environment. Boundaries need to stay aligned with how systems interact. Baselines need to keep holding as the environment evolves.
That is the difference between a diagram and an operating model. And that is why this layer matters so much.
Security still begins with separation. It still depends on defining meaningful zones. It still needs clear perimeters, even if those perimeters are now made of more than networks alone. It still relies on defense in depth.
What changed is that the cloud turned those ideas into a broader control-plane problem.
So the challenge now is not whether you have controls. It is whether your controls still express the architecture you intended. And whether that architecture is ready for the actors now operating inside it.
That’s the gap we help close.