DEV Community

bayu priatno
bayu priatno

Posted on

The Seven Problems of AI Engineering

Founder Journal #6 — Defining the Design Requirements for the Next Generation of AI Engineering

"Every great engineering system begins by understanding the problems it must solve."


AI Is Advancing Faster Than Engineering Practices

Artificial Intelligence has changed software development at an extraordinary pace.

We now have models capable of reasoning, generating code, writing tests, reviewing pull requests, and even planning software architectures.

Yet despite these remarkable advances, engineering teams continue to face the same recurring challenges.

Projects become inconsistent.

Knowledge disappears.

Architectural decisions are forgotten.

AI behaves differently across repositories.

Every new project starts by rebuilding the same engineering foundation.

This suggests an important conclusion:

The biggest bottleneck is no longer AI capability. It is engineering capability.

Before designing NAEOS, I asked a simple question:

What are the fundamental problems preventing AI from becoming a true engineering partner?

After months of experimentation and observation, seven recurring problems emerged.

These seven problems became the design requirements for NAEOS.


Problem 1 — No Persistent Context

AI understands only what it receives.

Without structured project context, every conversation begins almost from zero.

Developers repeatedly explain:

  • Project goals
  • Architecture
  • Technology stack
  • Naming conventions
  • Business rules

The larger the project becomes, the more expensive this repetition becomes.

What NAEOS Should Provide

A persistent engineering context that survives conversations, sessions, and even model changes.


Problem 2 — No Shared Engineering Standards

Every team has standards.

But AI rarely knows them.

Examples include:

  • Folder organization
  • Naming conventions
  • Error handling
  • Logging strategy
  • Security requirements
  • Documentation style

Without explicit standards, AI fills the gaps with assumptions.

The result is inconsistency.

What NAEOS Should Provide

Engineering standards that are machine-readable, reusable, and automatically applied across projects.


Problem 3 — No Engineering Memory

Engineering is built on accumulated knowledge.

Teams constantly make decisions:

  • Why was this architecture chosen?
  • Why was another solution rejected?
  • Which trade-offs were accepted?

Unfortunately, much of this knowledge remains trapped inside meetings, chat messages, or individual memories.

AI cannot learn from decisions it never receives.

What NAEOS Should Provide

Persistent engineering memory containing Architecture Decision Records (ADRs), historical context, and institutional knowledge.


Problem 4 — No Workflow Awareness

Generating code is only one step in software delivery.

Engineering also includes:

  • Planning
  • Design
  • Review
  • Testing
  • Security validation
  • Deployment
  • Monitoring
  • Maintenance

Most AI assistants focus on implementation alone.

They rarely understand where a task belongs within the broader engineering lifecycle.

What NAEOS Should Provide

Workflow orchestration that enables AI agents to participate throughout the complete software development lifecycle.


Problem 5 — No Governance

Enterprise software cannot rely solely on "best effort."

Organizations require:

  • Security policies
  • Compliance requirements
  • Approval workflows
  • Quality gates
  • Auditability

Without governance, AI may generate technically correct code that still violates organizational rules.

What NAEOS Should Provide

A governance layer where engineering policies are enforced consistently before changes reach production.


Problem 6 — No Reusable Architecture

Many teams create excellent architectures.

Few teams make those architectures reusable.

Every project redesigns:

  • Service boundaries
  • Module organization
  • Communication patterns
  • Documentation structures

Architecture becomes a one-time artifact rather than reusable engineering knowledge.

What NAEOS Should Provide

Reference Architectures that encode proven engineering patterns into reusable templates for future projects.


Problem 7 — No Unified Engineering Environment

Today's AI ecosystem is fragmented.

Different models.

Different IDEs.

Different prompt formats.

Different tools.

Different workflows.

Developers spend considerable effort adapting themselves to tools.

The tools rarely adapt to a shared engineering system.

What NAEOS Should Provide

A unified engineering environment where multiple AI models and coding agents operate using the same architecture, standards, memory, and governance.


The Pattern Behind All Seven Problems

Although these challenges appear different, they share a common theme.

None of them are fundamentally about code generation.

They are about engineering systems.

AI Focus Today AI Engineering Needs
Prompt Persistent Context
Conversation Engineering Memory
Code Generation Workflow Orchestration
Individual Tasks End-to-End Lifecycle
Model Intelligence Engineering Governance
Local Decisions Shared Architecture
Tool Features Unified Engineering Environment

This shift—from isolated interactions to structured engineering systems—is the central philosophy behind NAEOS.


Designing NAEOS Around Problems, Not Features

Many software platforms begin by listing features.

NAEOS begins differently.

Every module exists because it addresses one or more of these seven engineering problems.

For example:

  • Context Engine → Solves Persistent Context.
  • Memory Layer → Solves Engineering Memory.
  • Policy Engine → Solves Governance.
  • Workflow Runtime → Solves Workflow Awareness.
  • Architecture Registry → Solves Reusable Architecture.

The architecture follows the problems—not the other way around.


Looking Ahead

Understanding the problems is only the first step.

The next question is even more important:

If prompts are not enough, what replaces them?

In the next article, we'll explore why Prompt Engineering—while valuable—is only one small component of a much larger engineering system.

We'll discuss the transition from Prompt Engineering to AI Engineering Systems and why this shift may define the next decade of software development.


Discussion

Which of these seven problems affects your projects the most?

  • Persistent Context
  • Engineering Standards
  • Engineering Memory
  • Workflow Awareness
  • Governance
  • Reusable Architecture
  • Unified Engineering Environment

Or is there another challenge you believe deserves to be on this list?

Top comments (0)