01
AI-native engineering
The software-production organization itself is structured around agent teams with delegated authority, explicit responsibilities, deliberation, routing, independent review and governance.
AI-NATIVE ENGINEERING · PRODUCT · ORGANIZATION
Building systems where humans govern and AI organizations decide and execute.
I’m Ali Babaei, CTO at MelkOnline and solo architect & builder of a human-governed, agent-executed AI-native engineering organization.
I design engineering organizations and products in which AI is foundational—not a bolt-on feature and not just a coding assistant.
Solo Architect & BuilderHuman-GovernedAgent-ExecutedAI-Native by Design
ABOUT
I don’t use AI to assist a normal engineering team. I design and build the engineering organization itself around AI.
My work sits first in AI-native engineering organization design, and also in AI-native product design. Humans retain purpose, strategic authority and governance boundaries. Within those boundaries, AI-agent teams can deliberate, decide, plan, build, review and operate; owner intervention is reserved for structural choices, policy changes and true escalations.
At MelkOnline, I initiated the AI-first direction before we adopted the term AI-native, then architected and built the engineering organization behind that transition from first principles.
The resulting system treats software production as an organization: responsibilities, context, memory, decision processes, routing, independent review, quality gates, lifecycle, authority and feedback loops are designed as first-class parts of the system.
ORIGIN
The direction existed before the label.
The current AI-native direction at MelkOnline began before we were using the term “AI-native.”
I had built the earlier conventional version of MelkOnline myself. In April 2025, I proposed an AI-first direction for MelkOnline: AI should be foundational both in the product and in the way the product was engineered.
The proposal then went through a period of consideration and preparation. From July into August 2025, we continued building the conventional product and launched its pilot.
In August 2025, the AI-first direction was adopted for MelkOnline. We began using AI directly in development, initially including work on the existing application. In the same month, I proposed the architecture for the new AI-first product—conversational for users and AI-centered across core workflows such as data understanding and matching—and began architecting the agentic engineering organization that would support end-to-end software production.
In today’s terminology, that direction was AI-native in intent across both product and engineering.
The terminology evolved. The architectural direction did not.
THE MODEL
An AI-native engineering organization with human governance and agent execution
1Human purpose, strategic authority & governance
2Lead-agent interpretation, deliberation & coordination
3Agent planning, team / mission design & delegated decisions
4AI-agent execution workforce
5Independent review, test & evaluation
6Release, observation & feedback
02
AI-native product
AI belongs in the core user journey and decision/workflow architecture—not as a feature attached to a conventional product.
03
Feedback-driven redesign
Defects and observations can feed back into design, mission structure or execution strategy—not merely into another patch.
TIMELINE
From an AI-first direction to an AI-native engineering organization with delegated agent authority.
I proposed an AI-first direction for MelkOnline across both product and engineering.
The proposal was broader than adding AI features: AI would become foundational to the product experience and to how the product itself was built.
The direction remained under consideration while we prepared the next phase.
This was an interim period rather than the start of the new architecture in production.
We continued building the conventional product and launched its pilot.
The existing product path remained active while the larger AI-first direction had not yet become the operating direction.
The AI-first direction was adopted, AI entered MelkOnline development, and I began the new product and engineering architecture.
We first used AI directly in development, including work on the existing application. In the same month, I proposed the new AI-first product architecture—AI-native in today’s terms—and began architecting the agentic engineering organization for end-to-end software production.
I began turning that direction into a persistent multi-agent operating model.
The work moved beyond using AI tools toward designing continuity, delegation, coordination and explicit operating rules across agent roles.
I formalized the direction as an explicit AI-native engineering architecture.
The system moved toward defined team roles, agent nodes, routing and role-specific operating procedures, turning the AI-first direction into an operating engineering architecture.
I moved the engineering model up a level—from operating agents to engineering the layer that manages them.
The focus became delegation, durable coordination and management across specialized agents rather than manually driving individual implementation sessions.
I expanded the management layer into a role-based AI engineering organization.
Persistent responsibilities, delegation, reporting paths and independent review became organizational concerns rather than ad-hoc session behavior.
I added explicit deliberation and decision architecture for delegated agent autonomy.
The system was designed to reframe problems, compare alternatives and resolve decisions inside delegated scope without defaulting to owner intervention, while preserving a clear escalation boundary for owner-level choices.
I pushed the operating model further toward delegated authority across planning, decision and execution.
My role is concentrated on purpose, structural choices, governance boundaries and true escalations; within delegated scope, agent roles can plan, decide, build, review and operate without waiting for implementation-level direction.
FIELD NOTES
Lessons from building AI-native organizations in practice
GOVERNANCE
Human-governed ≠ human approval for every decision
Governance defines purpose, authority and escalation boundaries. Within them, agent roles can deliberate, decide and execute without waiting for human approval on every step.
ORGANIZATION
Operating through lead agents changes the human role
The human role moves up to purpose, boundaries and structural choices; lead and specialist decision roles can handle planning, coordination and delegated decisions below that line.
QUALITY
Independent review is organizational
If implementation and review share the same context and incentives, review can collapse into self-confirmation. Independence has to be designed.
More field notes will follow as the product and engineering organization evolve.
THESIS
Software engineering is no longer centered on writing code. Its frontier is engineering AI organizations.
My work treats the engineering organization itself as a system to be designed: authority, decision processes, context, memory, coordination, verification and escalation paths—so AI-agent teams can decide, build, review, operate and improve software inside explicit boundaries.
The unit of engineering is no longer only code. It is also the AI organization that can decide, build, review and operate it.