Skip to main content

How AI Can Use This Architecture

In the CDD architecture, AI agents are treated as intent emitters and state observers, integrated via the same event pipeline and API interfaces as human users. We support two types of AI agents with distinct responsibilities and boundaries.

🤖 Two Types of AI Agents​

  1. Development AI Agents
    Internal agents that assist with building the product (e.g. generating tools, automating workflows).

    • Can emit new domain-specific events to reactive-events
    • Can create new handlers and APIs in core
    • Capable of helping humans design or extend the system itself

  2. User-Facing AI Agents
    External agents (e.g. via MCP) that assist end users with tasks.

    • Cannot mutate system architecture
    • Can only call public APIs exposed by core
    • Must follow the same permission, access, and validation pipeline as human actions

🗂️ Involved Repos​

core, reactive-events, ai-agent (in future), system-context

🔁 Event & Execution Flow​

This flow diagram shows how Development AI Agents can introduce new capabilities to the system:

In contrast, User-Facing AI Agents interact with the system through pre-defined APIs only:

🧱 Design Principles​

To keep both AI types safely decoupled from core logic:

  • Internal AI agents contribute to the event definition layer
  • External AI agents interact only through explicit public APIs
  • No AI can directly mutate shared state—they always go through the same reducers and validators
  • All actions are observable, reversible, and trackable

🧠 Summary​

  • AI is a first-class participant in the CDD system—no backdoors or privileged shortcuts
  • Internal AIs help evolve the system by working through reactive-events and core
  • External AIs (via MCP) operate safely within API boundaries
  • This makes AI a modular, testable, and predictable part of your architecture