Composability
AI Products should support governed composition with AI Products, Data Products, Model Products, tools, and services. AIPS distinguishes system composition from the creation of a new Derived AI Product.
Composition alone does not create a new AI Product. A new Product requires a distinct governed intelligence proposition and Product boundary.
Composability Requirements
AI Products should declare:
-
Composable Interfaces
- input/output or action/effect contracts;
- compatible protocols and schema semantics.
-
Dependency Semantics
- dependency kind: AI Product, Model Product/AI Primitive, Data Product, knowledge/rule, tool/service;
- required vs optional dependencies;
- version/compatibility expectations.
-
Composition Behaviour
- sequencing, parallelization, delegation, aggregation, or other orchestration;
- whether the composition is merely operational or establishes a new Product proposition.
-
Extension Mechanisms
- adapters, plug-ins, tools, policies, or domain extensions.
Composition Patterns
Examples include:
- Pipelining — one Product's result feeds another;
- Parallel/Federated Use — several AI Products are used together without creating a new Product;
- Ensembling — multiple intelligence outputs are combined;
- Chaining — multi-step reasoning/generation across capabilities;
- Agentic Orchestration — an agent coordinates Products/tools;
- Productized Composition — composition establishes a new owned intelligence proposition.
Only the last pattern necessarily creates a new Product.
Derived AI Product Semantics
A new composed Product is Derived when:
- it itself qualifies as an AI Product;
- it owns a distinct governed intelligence proposition; and
- that proposition materially depends on one or more upstream AI Products.
Where multiple upstream AI Products materially contribute, the derivation topology may be described as Composite.
Classification:
APT = Derived
Derivation topology = Composite
Orchestration Is Not Automatically Productization
A router such as:
if request = translation → call Translation AIP
else → call Summarization AIP
may be useful system infrastructure without constituting a new AI Product. It becomes a Product only where a distinct reusable value proposition, ownership boundary, lifecycle, and governance are established; it becomes an AI Product only where the new Product also satisfies AI Product qualification.
Governance Integration
Composed systems must preserve:
- lineage and version binding;
- AI Product Type evidence;
- upstream risk/usage constraints;
- downstream impact analysis;
- policy compatibility;
- system-level risk evaluation where composition creates emergent behaviour.
Summary
- Composability enables modular AI ecosystems.
- AI Product dependency composition is distinct from BPS Compound Product intrinsic structure.
- Multiple upstream AI Products may create a Composite Derived AI Product only when a new intelligence proposition is productized.
- Pure orchestration or federated use does not automatically create another AI Product.
Principle: Composition explains how capabilities are used together; Productization determines whether a new owned intelligence proposition exists.