Why are AI coding agents still hard to use in real engineering?

I’ve been building NAEOS around a question:

What happens when AI coding agents move from “write some code” to actually participating in software engineering?

Most AI coding tools are getting very good at generating code.

But production engineering involves much more:

• architecture

• engineering standards

• security

• testing

• documentation

• governance

• consistency

• decisions made across multiple sessions

• knowing which context is authoritative

This creates an interesting problem.

An agent can generate technically valid code while still making the wrong engineering decision.

That led me to build NAEOS — an Engineering Operating System for AI-assisted software development.

The idea is simple:

Don't just give agents more context. Give them an explicit engineering system.

NAEOS uses things like:

→ Engineering Constitution
→ Architecture rules
→ Policies
→ Standards
→ Playbooks
→ Templates
→ AI instructions
→ Quality gates
→ Multi-agent workflows

I'm building it in public and experimenting with how far this approach can go.

I'd love to hear from other builders:

What is the biggest problem you've encountered when using AI coding agents on real production projects?

Context management?
Architecture drift?
Code quality?
Consistency?
Security?
Agent coordination?

Curious to hear what others are seeing.

#AIEngineering #CodingAgents #SoftwareEngineering #BuildInPublic