Skip to main content

Integration & Composability

AI Products may participate in workflows, systems, agent ecosystems, physical control architectures, and other Products. AIPDS describes the operational integration of those dependencies while preserving AIPROD's ontology distinctions.


Integration Requirements​

AIPDS should declare, where applicable:

  • supported interfaces and protocols;
  • runtime dependencies;
  • upstream AI Product version bindings;
  • Model Product / AI Primitive dependencies;
  • Data Product and knowledge dependencies;
  • tool/action interfaces;
  • synchronous, asynchronous, event-driven, agentic, or embedded integration patterns;
  • failure, timeout, fallback, and circuit-breaking behavior.

Versioned dependency bindings are strongly preferred for reproducibility and qualification-currentness analysis.


Composition Is Not Automatically Product Creation​

The following runtime graph:

AI Product A ─┐
AI Product B ─┼── orchestration/workflow
AI Product C ─┘

is not automatically a new AI Product.

A new Derived AI Product exists only where the composition establishes a distinct governed intelligence proposition with its own Product identity/version, AIPROD, qualification evidence, ownership, and lifecycle.

AI Product A ─┐
AI Product B ─┼──► New governed intelligence proposition
AI Product C ─┘ ↓
Derived AI Product D

Where multiple upstream AI Products materially constitute D, its derivation topology may be Composite.


Runtime Policy Propagation​

Composition must account for:

  • prohibited uses;
  • data and AI policy constraints;
  • risk escalations;
  • access/entitlement boundaries;
  • effect/action authority;
  • licensing and economic constraints;
  • upstream qualification currentness.

Policy inheritance is not assumed to be a simple union; conflicts must be resolved explicitly by governance rules.


Dependency Change and Requalification​

If an upstream AI Product version changes such that its AI qualification or Product Family changes, downstream deployments must not silently follow an unpinned latest dependency.

A new dependency binding should trigger impact analysis of:

  • APT;
  • qualification evidence;
  • intelligence behavior;
  • risk;
  • interfaces/effects;
  • monitoring and governance obligations.

Principle: AIPDS governs operational composition; AIPROD/Core determines whether that composition constitutes a new Derived AI Product.

Consumption through marketplace entitlements​

Marketplace entitlement controls who may resolve and use a Product port; it does not change composition semantics. A consumer that acquires several AI Products through the marketplace is performing federated/operational consumption unless a new governed intelligence proposition is intentionally productized.

Entitlement state must be checked together with Product currentness. A previously granted entitlement must not force continued access when the Product is suspended, retired, revoked, or no longer satisfies required qualification/governance policy.