A structured, semantic, and deterministic workflow for AI‑assisted software development
HCMD is a discipline that separates meaning, structure, and implementation into explicit, human‑authored layers.
It provides a stable, reproducible alternative to prompt engineering and ad‑hoc agent behavior by introducing a clear semantic pipeline:
HUT → AML → STS → PSC → Implementation
HCMD is designed for developers and scientific professionals who want AI assistance without losing architectural control, determinism, or reproducibility.
It works naturally with Windows, Visual Studio 2026, VS Code, and the Copilot App.
Current AI coding tools are powerful, but they lack:
- semantic grounding
- explicit task decomposition
- deterministic structure
- reproducibility across sessions
- stable regeneration
- separation of meaning from mechanics
HCMD addresses these gaps by externalizing reasoning into structured, human‑authored artifacts:
- HUT — human meaning preparation
- AML — semantic meaning
- STS — procedural meaning
- PSC — deterministic structure
- HMD — unified meaning + structure
This allows humans to define what the system means and how it is shaped, while AI handles the mechanical implementation.
HCMD now uses a five‑stage conceptual pipeline:
HUT → AML → STS → PSC → Implementation
- HUT clarifies human intent
- AML expresses semantic meaning
- STS expresses procedural meaning
- PSC expresses deterministic structure
- Implementation executes structure mechanically
This pipeline is one‑directional and regeneration‑safe.
A major evolution of HCMD:
PSC replaces local training.
PSC is the training.
PSC acts as the structural bytecode of the domain.
It ensures deterministic behavior across cloud Copilot, local LLMs, and future AI systems.
HCMD clarifies the roles of human and machine:
- prepares meaning through HUT
- expresses intent through AML
- validates STS
- remains the semantic authority
- transforms STS into PSC
- generates implementation
- executes structure mechanically
This separation preserves meaning while enabling determinism.
HCMD is organized into three layers:
concept/ → environment/ → implementation/
meaning operation execution
Contains the complete conceptual foundation of HCMD:
- HCMD Manifesto
- HCMD Overview and Principles
- AML, STS, PSC, and HMD conceptual definitions
- The Three‑Layer Model
- Conceptual Workflow
- Conceptual Example
This layer defines the discipline before any tooling or implementation enters the picture.
Explains how HCMD is executed using:
- Copilot App (semantic environment)
- GitHub Copilot (implementation environment)
- Windows (stable execution substrate)
It includes:
- operational workflow
- example sessions
- regeneration guidelines
- PSC‑first training integration
Planned additions:
implementation/— schemas, templates, generators, and HMD toolingexamples/— complete HCMD projectsdocs/— public documentation for practitioners
Today’s work completed a full modernization of the concept/ layer:
- updated HCMD Overview
- updated HCMD Principles
- updated HCMD Manifesto
- updated Three‑Layer Model
- updated AML, STS, PSC, HMD concepts
- updated Conceptual Workflow
- updated index.md
These documents now reflect the 2026 HCMD architecture.
The file Conceptual_Example.md remains intentionally minimal.
Upcoming work includes:
- updating the
environment/layer - aligning the
implementation/layer - syncing HCMD‑Tutorials with the new conceptual foundation
- adding PSC‑first training examples
- adding regeneration‑safe implementation templates
HCMD is my personal research project, developed independently.
I remain affiliated with LBNL and can be reached at H_Nishimura@lbl.gov.