Front Matter: The AI-Native SDLC

What software development looks like when the machine writes it

The AI-Native Canon - Volume VI Year: 2026 Author: Alpesh Nakrani

A practical reconstruction of software development lifecycle discipline for teams that use AI to draft, modify, test, review, and sometimes operate code.

An AI-native SDLC is a software delivery lifecycle rebuilt for machine-written code: intent first, provenance captured, risk routed, behavior evaluated, ownership explicit, and production learning wired back into the system.

Key Takeaways

• The AI-native SDLC does not remove engineering discipline; it makes intent, provenance, and evidence more important.
• Generated code should be managed as one artifact inside a chain that includes specs, prompts, tests, ownership, and rollback.
• The ACCEPT loop gives teams a practical operating model for making machine-written change survivable.
• The book is for leaders and engineers who need generation speed without losing accountability.

Research spine

This chapter uses: DORA, State of AI-assisted Software Development 2025; NIST SP 800-218 Secure Software Development Framework; SLSA: Supply-chain Levels for Software Artifacts.

Preface

The software development lifecycle was built around a human author. Requirements were written by people, implementation was written by people, tests were written by people, review was performed by people, and production incidents were investigated by people. AI does not remove this lifecycle. It breaks some of its assumptions.

When a machine can generate a meaningful share of the implementation, the team needs stronger, not weaker, engineering discipline. The diff may be larger. The author may not understand every line. The generated code may pass local tests while violating architecture, security, or product intent. The prompt that created the change may be as important as the code. The spec may be the only durable human artifact. The review process must move from "do I like this code?" to "does this change satisfy a traceable intention under a defined risk class?"

This book is for engineering leaders, staff engineers, platform engineers, product engineers, and AI-native builders who need to operate software delivery when generation is abundant. It proposes the ACCEPT loop: Articulate, Create with provenance, Check mechanically, Evaluate behavior, Prove ownership, Trace in production. The point is not to slow AI-assisted development down. The point is to make its speed survivable.

Contents

• Chapter 1: The Generated PR That Passed
• Chapter 2: Specs, Prompts, and Provenance as Lifecycle Artifacts
• Chapter 3: The ACCEPT Loop
• Chapter 4: Risk-Classifying Generated Diffs
• Chapter 5: AI-Native CI
• Chapter 6: Human Review When the Author Is a Model
• Chapter 7: Agent Permissions, Sandboxes, and Supply Chain Trust
• Chapter 8: Testing Behavior, Not Just Code Paths
• Chapter 9: Release, Observability, Rollback, and Incident Learning
• Chapter 10: The Operating Playbook