DEV Community

bayu priatno
bayu priatno

Posted on

Introducing the NAEOS Reference Architecture

Founder Journal #11 — Designing an Operating System for AI Engineering

"Architecture is not about drawing boxes. It is about defining how engineering knowledge flows through a software system."


Every Great Platform Has a Reference Architecture

Modern software platforms rarely begin with implementation.

Instead, they begin with architecture.

Linux has a kernel architecture.

Kubernetes has a control plane.

Cloud platforms define networking, identity, storage, and compute layers.

These reference architectures provide something more valuable than code:

They provide shared understanding.

When thousands of engineers contribute to the same ecosystem, architecture becomes the language that keeps everyone aligned.

I believe AI Engineering needs the same foundation.

That belief became the starting point for the NAEOS Reference Architecture.


Why a Reference Architecture?

Many AI projects focus on features.

  • Better prompts
  • Better agents
  • Better plugins
  • Better IDE integrations

These are useful.

But features evolve quickly.

Architecture endures.

A reference architecture answers questions that features cannot:

  • Where does engineering knowledge live?
  • How does AI receive context?
  • Who enforces engineering policies?
  • How is project memory preserved?
  • How do multiple AI agents collaborate?
  • What remains stable when AI models change?

Those questions define the system—not the tools.


Thinking in Layers

One of the core design principles behind NAEOS is layered architecture.

Each layer has a single responsibility.

Each layer communicates through well-defined interfaces.

This keeps the system modular, replaceable, and understandable.

At a high level, the architecture looks like this:

┌────────────────────────────────────────────┐
│ Governance Layer                           │
├────────────────────────────────────────────┤
│ Engineering Constitution                   │
├────────────────────────────────────────────┤
│ Policy Engine                              │
├────────────────────────────────────────────┤
│ Context & Memory                           │
├────────────────────────────────────────────┤
│ Workflow Runtime                           │
├────────────────────────────────────────────┤
│ AI Agent Layer                             │
├────────────────────────────────────────────┤
│ Development Tools                          │
└────────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

Notice something interesting.

The AI model sits near the bottom.

Not the top.

Because intelligence should operate inside engineering—not above it.


Layer 1 — Governance

Everything begins with governance.

Governance defines:

  • Vision
  • Roadmap
  • Versioning
  • Ownership
  • Core engineering principles

This layer answers:

Why does this engineering system exist?


Layer 2 — Engineering Constitution

The Constitution defines rules that are not negotiable.

Examples:

  • Security principles
  • Documentation requirements
  • Architecture principles
  • Testing requirements
  • Quality standards

AI agents inherit these principles before they perform any engineering task.


Layer 3 — Policy Engine

Policies transform principles into enforceable rules.

For example:

Instead of saying:

"Testing is recommended."

The Policy Engine can require:

"Every Pull Request must include passing tests."

Architecture becomes operational.


Layer 4 — Context & Memory

This layer provides AI with engineering knowledge.

Including:

  • Documentation
  • ADRs
  • Domain knowledge
  • Architecture
  • Coding standards
  • Historical decisions

Instead of explaining the project repeatedly, AI retrieves persistent engineering context.


Layer 5 — Workflow Runtime

Engineering is a sequence of coordinated activities.

Planning.

Implementation.

Testing.

Review.

Deployment.

Monitoring.

The Runtime coordinates these activities while ensuring policies remain enforced throughout the lifecycle.


Layer 6 — AI Agent Layer

Only now does AI begin generating code.

The AI Agent does not operate independently.

It works within the engineering environment created by the previous layers.

Its intelligence is amplified by engineering knowledge.


Layer 7 — Development Tools

Finally come the tools developers interact with every day.

Examples include:

  • IDEs
  • Git platforms
  • CI/CD systems
  • AI Coding Assistants
  • Documentation portals
  • Package registries

These tools become clients of the engineering system rather than isolated products.


Why This Architecture Matters

Notice what remains constant if the underlying AI model changes.

Everything above the AI layer.

The Constitution remains.

Policies remain.

Architecture remains.

Documentation remains.

Memory remains.

Governance remains.

This is intentional.

Engineering knowledge should outlive individual AI models.


The Philosophy Behind the Layers

Each layer answers a different engineering question.

Layer Core Question
Governance Why are we building this?
Constitution Which principles cannot be violated?
Policy How are principles enforced?
Context What should AI know?
Memory What has already been learned?
Runtime How should work flow?
AI Agents How should tasks be executed?
Tools How do humans interact with the system?

Together, these layers transform AI from an isolated assistant into part of a complete engineering platform.


Architecture Drives Engineering

One of the guiding principles behind NAEOS is simple:

Architecture Drives Engineering.

Good architecture reduces ambiguity.

Good architecture enables collaboration.

Good architecture survives technology changes.

As AI models continue evolving, the engineering architecture should remain the stable foundation upon which they operate.

That is exactly what NAEOS is designed to provide.


Looking Ahead

Now that we've explored the overall architecture, the next articles will dive into each layer individually.

We'll begin at the very top:

The Engineering Constitution.

Why should every AI engineering project have a constitution?

And why might it become the most important document in an AI-native software organization?


Discussion

If you were designing an AI Engineering platform from scratch, which architectural layer would you build first?

  • Governance
  • Constitution
  • Context
  • Memory
  • Policy
  • Workflow
  • AI Agents

Or would you choose a completely different architecture?

Top comments (0)