This manifesto outlines a paradigm shift towards AI-native software engineering, providing a playbook for developers in the LLM era. It redefines core engineering principles to leverage AI effectively, aiming to unlock new productivity and manage complexity.

How It Works

The approach advocates embracing AI-native principles, treating test cases as the sole compounding asset, and recognizing AI's stateless nature where the context window is paramount. It proposes transitioning from deep, vertical software stacks to wide, horizontal systems, underpinned by workflows like Plan–Act, Test–Code, and Doc–Code–Doc. Code is envisioned as tiny, isolated, AI-readable units, with AI IDEs' core value in intelligent context selection.

Quick Start & Requirements

This document is a manifesto and playbook, not a software project with direct installation or execution instructions.

Highlighted Details

• Presents "11 Hard Truths for 2025," asserting technology older than three years is obsolete, test cases are the primary asset, AI is stateless, and the context window is the most critical computational resource.
• Introduces AI-Native Engineering Practices, detailing concepts like AI statelessness, the Test–Code loop, debugging with AI, context selection, AI-native workflows, and the shift towards horizontal complexity.
• Hypothesizes a move towards 5,000 small, independent, AI-readable functions over deeply abstracted ones to better fit AI's context window limitations.
• Emphasizes AI IDEs should handle the full lifecycle, including deployment, requiring AST and symbol-level indexing for robust refactoring and navigation.
• Suggests AI agents should evolve from neutral tools to opinionated, personalized assistants.

Maintenance & Community

The document is actively revised, with updates noted for March 2025, May 2025, and December 2025, indicating ongoing development and refinement of AI-native engineering concepts.

Licensing & Compatibility

No specific software license is mentioned. This lack of information may pose compatibility concerns for commercial use or integration into proprietary systems.

Limitations & Caveats

The manifesto acknowledges AI cannot fully address the "first mile" (solution design) or "last mile" (real-world code correctness), necessitating continuous human-in-the-loop involvement. It notes AI struggles with novel tech stacks due to limited training data, and managing complexity in deep, vertical code chains remains challenging.