AI Artifacts, Assets, Model Products and AI Products
AI ecosystems contain many valuable things that are often described loosely as "AI Products". AIPS distinguishes them by asking two independent questions:
- Has the thing been productized?
- If it is a Product, does its Product Intrinsic Nature qualify it for the AI Product Family?
Productization alone does not establish AI Product status.
1. AI Artifact / Asset
An AI Artifact or Asset is a useful building block that has not necessarily been established as an independently managed Product.
Examples include:
- raw model checkpoints;
- datasets and evaluation sets;
- embeddings;
- prompt libraries;
- feature definitions;
- feature stores;
- knowledge bases;
- vector stores;
- rulesets;
- experimental pipelines;
- benchmark results.
An asset may be critical to an AI Product without being a Product itself.
2. Productized AI-related Capability
An AI-related asset can itself be productized without necessarily becoming an AI Product.
Examples include:
- Model Product — a governed, versioned, reusable learned computational capability;
- Model Service Product — a hosted service exposing generic model inference;
- Feature Platform Product — a governed feature-store capability;
- Knowledge Product — a governed reusable knowledge or retrieval capability;
- Tool / Platform Product — a reusable execution, orchestration, or AI-development capability.
These may be Business Products in the context of the organizations that sell them while remaining distinct from the AIPS AI Product Family.
3. AI Product
An AI Product is a Product whose value-bearing essence is a distinct governed artificial-intelligence capability and whose current constitution directly or transitively contains at least one qualifying AI Primitive.
Examples include:
- governed fraud-detection intelligence;
- an enterprise assistant with a defined intelligence proposition;
- a recommendation intelligence Product;
- autonomous robot intelligence;
- a Derived decision-intelligence Product composed from upstream AI Products.
See What is an AI Product? and AI Product Ontology.
4. AI-enabled Product
A Product may materially use AI while belonging intrinsically to another Product Family.
Examples:
| Product | Intrinsic nature | AI role |
|---|---|---|
| Smartphone with AI assistant | Physical / Engineered Product | Enabling |
| Streaming service using recommendation AI | Service Product | Enabling |
| AI-generated movie | Creative / Content Product | Constructional |
| Fully autonomous humanoid | Potential Compound Product | AI may be co-essential |
The presence of AI inside a Product's constitution does not automatically cause AI Product Family inheritance.
5. Typical Qualification Examples
| Artifact / capability | Typical classification | AI Product? | Why |
|---|---|---|---|
| Raw trained model checkpoint | AI Asset | No | Not independently productized |
| Governed downloadable LLM | Model Product | Usually no | Productized learned capability, but not necessarily a distinct governed intelligence proposition |
| Generic hosted model endpoint | Model Service Product | Depends | May remain generic model serving; can qualify only if the AI Product tests are satisfied |
| AI assistant / copilot | AI Product | Usually yes | Intelligence is typically the value-bearing essence and a distinct intelligence proposition exists |
| AI agent | AI Product or component | Depends | Must establish a governed intelligence proposition; orchestration alone is insufficient |
| Feature store | Data / ML platform capability | No | Supports intelligence production but intelligence is not its value-bearing essence |
| Rules engine | Platform / possible AI Primitive | Depends | Generic deterministic infrastructure is not an AI Product; a productized reasoning capability may qualify |
| Knowledge base / vector store | Knowledge or Data Asset/Product | Usually no | Supplies context or retrieval rather than a distinct intelligence proposition |
| Training dataset | Data Asset / Data Product | No | Input to intelligence construction, not itself AI intelligence |
| Evaluation dataset / benchmark | Evaluation Asset/Product | No | Supports assurance rather than providing intelligence |
| Prompt library | Prompt Asset/Product | Usually no | Construction/configuration artifact, not by itself a distinct intelligence proposition |
| RAG pipeline | Solution component or AI Product | Depends | Qualification depends on Product boundary, intelligence proposition, and AI Primitive evidence |
| Fraud detection intelligence | AI Product | Yes | Distinct governed intelligence proposition backed by qualifying AI Primitive(s) |
| Recommender intelligence | AI Product | Usually yes | Productized recommendation intelligence when governed as a Product |
| Autonomous robot intelligence | AI Product | Yes | Productized embodied intelligence capability; may participate in a larger Compound Product |
6. Model Product Example
Consider a company that sells downloadable LLMs through a marketplace.
For the vendor, the Model Product may be its Business Product. For the consumer enterprise, the same Model Product may be a procured/enabling Product and AI Primitive dependency used to construct a separate AI Product.
Business Product role and intrinsic AI Product qualification are therefore separate concerns.
7. Key Rules
- AI-related ≠ AI Product.
- Productized ≠ AI Product.
- Model Product ≠ AI Product.
- Uses an AI Product ≠ becomes an AI Product.
- AI Product qualification requires intelligence to be the value-bearing essence.
- AI Product qualification requires a direct or transitive qualifying AI Primitive.
- Business Product is a contextual role, not an intrinsic AI Product classification.