AI Unified Process

Requirements-driven, AI-native software development.

Specs at the center. AI handles the rest.

unifiedprocess.ai · Methodology · Tools · Tutorial · Articles · Deutsch

The AI Unified Process (AIUP) is a spec-driven development methodology in which requirements — not code — are the source of truth. Every project starts with a written vision and proceeds through a requirements catalog, an entity model, and use case specifications before a single line of code is written. Code, tests, and documentation are then generated, regenerated, and refactored around those specs by AI.

The phases follow the Rational Unified Process — Inception, Elaboration, Construction, Transition — adapted for AI-driven workflows.

How this differs from other spec-driven tools. In most of them, a developer writes a spec to describe the code they are about to write: an elaborate prompt for one feature, effectively discarded once the code ships. AIUP specs describe the behavior of the system — what it must do, not how the code does it. They are living artifacts owned by requirements engineers (or by developers deliberately doing requirements engineering), and they outlive any implementation.

Nothing gets built without a use case. Nothing reaches production without tests traceable to requirements.

Inception          Elaboration                           Construction
─────────────────  ───────────────────────────────────   ──────────────────────────────────
/requirements  →  /entity-model  →  /use-case-diagram  →  /use-case-spec  →  /flyway-migration
                                                                          ↘  /implement
                                                                          ↘  /browserless-test
                                                                          ↘  /playwright-test

Each step picks up the files the previous one produced — docs/vision.md, docs/requirements.md, docs/entity_model.md, docs/use_cases.puml, docs/use_cases/UC-*.md, docs/test_cases/TC-*.md — and you can inspect or edit any of them before continuing.

Inheriting a legacy codebase? /reverse-engineer walks the existing code, configuration, and schema and produces the same artifacts the forward workflow would have, giving you a documented baseline.

1. Write a docs/vision.md describing the product, its users, and its goals — the richer it is, the better everything downstream.

2. Install the plugins in Claude Code:

   /plugin marketplace add ai-unified-process/marketplace
   /plugin install aiup-core
   /plugin install aiup-vaadin-jooq        # for a Vaadin + jOOQ project
   /plugin install aiup-angular-jpa        # for an Angular + JPA project
   

   Using a different stack? Install aiup-core on its own — the methodology skills are stack-agnostic. On another AI coding agent, install via Tessl instead: tessl install aiup/aiup-core.

3. Run /requirements and let the workflow carry you from there.

The tutorial walks through the whole chain on a real application, and the videos show it in motion.

Workshops, coaching, and enterprise rollout support are available — see Enterprise or get in touch.