Skip to main content

Frequently Asked Questions (FAQ)

This FAQ addresses common questions about AI Products and the AI Product Specification (AIPS).

Is an AI Product just a model API?​

No.
A model API may look like a product, but unless it satisfies the true AI Product characteristics, it is only an AI Asset.

For example, an LLM API without governance, prohibited use declarations, or monitoring cannot be considered a true AI Product.
Productness is earned, not assumed.

Why not just treat AI as a Data Product?​

While every AI system consumes and produces data, an AI Product is defined by capability, not content.
See Why Not a Data Product?.

Key differences:

  • AI Products are dynamic (they drift, retrain, and evolve).
  • They carry ethical and societal risks (bias, manipulation, unsafe autonomy).
  • They require explainability and risk declarations that go beyond Data Products.

What is the difference between an AI Asset and an AI Product?​

  • AI Asset = a building block (datasets, weights, embeddings, prompts, feature stores).
  • AI Product = a packaged, governed, discoverable unit of AI capability.

The difference is not about abstraction level.

  • A raw checkpoint can be turned into a Product if wrapped with metadata, governance, and deployment.
  • An API may still be only an Asset if it lacks the required characteristics.

See AI Asset vs AI Product.

Do AI Products have to include self-service?​

Yes.
Self-service is the load-bearing principle of AIPS.
If a consumer cannot access, evaluate, and integrate the AI Product without special arrangements, it fails the definition of a Product.

What about symbolic AI, hybrid AI, AGI, or ASI?​

AIPS is designed to be future-proof.
The Capability Types page defines categories from narrow models to agents, hybrids, AGI, and ASI.
All are valid AI Products — as long as they meet the characteristics required by AIPS.

Who is responsible for an AI Product?​

Every AI Product must declare:

  • An Owner (accountable for lifecycle and governance).
  • A Steward (responsible for day-to-day maintenance).

This aligns with the BPS principle of product ownership and ensures accountability.

Can AI Products produce Data Products?​

Yes.

  • Training data, embeddings, and generated datasets can be released as Data Products.
  • But the AI Product itself remains distinct — its value lies in its capability, not its content.

How do AI Products integrate into enterprise platforms?​

AI Products are first-class citizens in enterprise architecture.
They integrate with:

  • Data Platforms (for inputs and outputs).
  • Application Platforms (via APIs/services).
  • Governance & Compliance frameworks (risk, ethics, prohibited uses).
  • Observability platforms (monitoring drift, fairness, and performance).

See Architecture Alignment.

Summary​

  • Not all APIs are AI Products.
  • AI ≠ Data — AI Products need their own spec.
  • Self-service, governance, and discoverability are non-negotiable.
  • Capability types include today’s models and tomorrow’s AGI/ASI.
  • Ownership and accountability are essential.

Principle: AIPS ensures that AI is consumed as a governed, trustworthy, and future-proof Product — not just as an asset.