AIPCH04 — Declarative Intelligence Composition

“AI Product Is Composed via Declarative, Intent-Driven Expression”

What AIPCH04 is really asserting​

AIPCH04 is not asserting that:

“Users can configure AI systems using forms, prompts, or low-code tools.”

It is asserting that:

The AI Product is created through a declarative composition of intent, behavior, constraints, and constituent capabilities — where the creator expresses what the intelligence should do, and the platform deterministically compiles how it is realized.

This is a shift from:

engineering AI systems
to
composing intelligence as a product

The Essence (HDIP + AIPS Interpretation)​

An AI Product satisfies AIPCH04 if and only if:

1. The creator expresses the product through intent, behavior, and constraints
2. The composition is complete enough to define the capability
3. The platform acts as a compiler of intelligence, not a toolkit

If creation requires:

• designing pipelines
• selecting models
• orchestrating components
• writing prompts as logic
• assembling workflows manually

then AIPCH04 is not met, even if a UI is used.

Important Distinction​

AIPCH04 must be clearly separated from:

🔹 AIPCH03 — Declarative Specification​

👉 AIPCH04 ensures:

The creator composes intelligence — the platform performs it.

Positive Criteria — When AIPCH04 is met​

AIPCH04 is met when all of the following are true:

1. Creation is expressed entirely in intent, not implementation​

The creator defines:

• what decision or capability is required
• expected behavior and outcomes
• constraints (risk, policy, safety, explainability)
• usage context and expectations

The creator does not define:

• model types
• pipelines
• orchestration logic
• prompts as execution scripts

2. Composition includes capability dependencies (where applicable)​

The AI Product may reference:

• other AI Products
• Data Products
• tools or external capabilities

But:

• these are referenced at capability level, not wired procedurally
• composition is declarative, not orchestrated manually

3. Platform performs deterministic compilation​

The platform:

• selects appropriate models and techniques
• orchestrates execution (agents, pipelines, tools)
• provisions infrastructure
• binds policies and governance
• generates AIPDS and runtime artifacts

The creator does not influence how this is done.

4. No human translation layer exists​

There is no step where:

• engineers interpret requirements
• data scientists implement models manually
• platform teams translate intent into pipelines

The system is:

intent → compiled execution

5. Composition is reproducible​

The same declarative composition:

• produces consistent behavior
• can be versioned and replayed
• does not depend on hidden manual steps

Negative Criteria — When AIPCH04 is not met​

AIPCH04 is not met if any of the following are true:

❌ Composition requires procedural design​

Examples:

• building DAGs or workflows
• chaining models manually
• writing orchestration logic
• defining step-by-step execution

This is engineering, not composition.

❌ Prompting is used as implicit orchestration​

Examples:

• prompts encode logic, flow, or control decisions
• system behavior emerges from prompt engineering rather than declared intent

This is fragile and non-declarative.

❌ Creator must choose technical implementation​

Examples:

• selecting models (e.g., GPT vs XGBoost)
• tuning hyperparameters
• choosing tools or frameworks

This violates abstraction.

❌ Composition is hidden in code or scripts​

Examples:

• notebooks define behavior
• pipelines define interactions
• orchestration is implicit in implementation

This breaks declarative composition.

Edge Cases (Important Guidance for Agents)​

Case 1: “Low-code workflow builder”​

❌ Not met

Rationale:

• user still designs flow
• still procedural thinking
• abstraction is incomplete

Case 2: “Prompt-driven AI with structured templates”​

⚠️ Partial

Rationale:

• intent partially captured
• but execution logic often embedded in prompts
• not fully declarative

Case 3: “Intent + constraints → platform compiles full system”​

✅ Met

Rationale:

• creator defines what
• platform defines how
• no engineering leakage

Case 4: “Agent-based systems with manual orchestration”​

❌ Not met

Rationale:

• composition is explicit and procedural
• violates declarative model

Evidence Signals an Agent Should Look For​

Authoritative evidence:​

• intent-driven composition interface
• absence of user-facing model/pipeline selection
• AIPROD capturing intent, behavior, and constraints

Supporting evidence:​

• automatic generation of AIPDS and orchestration
• no code authored by domain for composition
• platform logs showing compilation decisions

Red flags:​

• prompts used as control logic
• workflow diagrams authored by users
• model selection exposed in UI
• reliance on engineering teams for composition

How an Agent Should Decide​

Decision rule (simplified):

If the creator must think in terms of implementation, orchestration, or execution steps instead of intent, behavior, and constraints, AIPCH04 is not met.

Why AIPCH04 Is Non-Negotiable​

Without AIPCH04:

• AI creation remains engineering-bound
• composability does not scale
• governance cannot be embedded at creation time
• democratization is limited to usage, not creation

AIPCH04 enables:

• scalable AI creation
• true self-service capability composition
• consistent governance through compilation
• future-proof support for agents and composite systems

Canonical Statement (for AIPS)​

AIPCH04 is satisfied only when an AI Product is created through declarative composition of intent, behavior, constraints, and constituent capabilities, with all implementation fully abstracted and deterministically compiled by the platform, without requiring procedural design or engineering intervention from the creator.