The same foundation model can support a writing assistant, a maintenance copilot or a robot. Calling all three "AI products" hides the decisions that matter. Their authority, failure consequences, operating environment and evidence needs are different.
This taxonomy is a product-management tool. It is not a legal classification and it does not replace a system-specific safety, cybersecurity or regulatory assessment.
The short definition
Generative AI creates or transforms information for a person to assess.
Operational AI informs or coordinates a business or operational workflow while a person or deterministic process retains decision authority.
Physical AI perceives, decides or recommends within a system that can change the physical world. Its output is connected—directly or through an operator—to a machine, vehicle, robot, energy asset or industrial process.
The boundary is authority, not branding. A language model can appear in all three categories. A computer-vision model can be operational in one product and physical in another.
Generative AI — content remains under human judgement
A generative system drafts text, code, images, audio or structured information. The user reviews the output before it becomes consequential.
Product decisions concentrate on usefulness, quality, provenance, access, intellectual property, privacy and the cost of review. The system can still cause harm, but the normal interaction assumes a person decides what to accept and what to do next.
Examples include document drafting, design exploration, code assistance and content transformation.
Evidence focus: task quality, user review behaviour, provenance, data handling and failure disclosure.
Operational AI — the product shapes a workflow
Operational AI classifies, predicts, recommends or coordinates inside a business or field workflow. It may prioritize a maintenance queue, summarize fleet signals, recommend an inspection or help an operator diagnose an issue.
The output can materially influence work even when it does not command a machine. Product discovery therefore has to examine how people interpret recommendations, how exceptions are escalated and how accountability is preserved.
Examples include predictive-maintenance planning, operator assistance, demand forecasting, work-order prioritization and fleet exception triage.
Evidence focus: decision quality, workflow fit, human oversight, escalation, traceability and operational ownership.
Physical AI — the product participates in a physical system
Physical AI connects perception or learned decision logic to physical action. The model may control an actuator, propose a trajectory, trigger a machine response or guide an operator whose next step has a physical consequence.
The product is never only the model. It includes sensors, compute, connectivity, deterministic controls, human authority, update paths, observability, degraded behaviour and incident response.
Examples include robotic manipulation, autonomous navigation, machine vision connected to line action, driver-assistance functions and intelligent equipment that adapts to its environment.
Evidence focus: operating envelope, latency, system behaviour, safe state, human authority, hardware variation, field acceptance, update control and lifecycle ownership.
A recommendation can cross the boundary
Consider one maintenance model.
- If it drafts a technician note for review, it behaves as generative AI.
- If it prioritizes work orders, it behaves as operational AI.
- If it changes machine operation or triggers an automated physical response, it participates in Physical AI.
The model did not change. The intended use, integration and authority did. That is why model-centric classification produces weak product decisions.
The six classification questions
Use these questions at the start of Strategy and again before Productization:
- What can the system observe? Define sensors, data sources, context and blind spots.
- What can the system decide or recommend? Write the output in operational terms.
- Who or what receives the output? A user, workflow engine, controller or actuator creates a different authority path.
- What can change in the world? Include indirect changes through an operator.
- What overrides the system? Name deterministic controls, human authority and safe state.
- What evidence permits release and change? Define evaluation, field acceptance, monitoring and update approval.
If the team cannot answer those questions, the product has not been classified deeply enough to choose architecture, evidence or ownership.
The product implications
Strategy
Generative products often compete on usefulness and workflow integration. Operational products also depend on decision ownership and adoption. Physical AI products add a coupled hardware-software-data system, a field environment and physical consequences. The opportunity thesis must reflect that whole product.
Discovery
A user interview is not enough for a Physical AI product. Observe operators, maintainers, installers and the environment. Record edge cases, constraints, workarounds and acceptance authority.
Productization
Translate the taxonomy into architecture. Decide where inference runs, how state moves, what remains deterministic, how the system degrades and what evidence accompanies a release.
Launch
Define installation, acceptance, support, incident response and change authority. A successful demonstration does not establish those capabilities.
Scale
Track whether value, deployment and economics repeat across sites, hardware variants and operating conditions. Do not infer repeatability from one bounded test.
Relationship to regulation
This taxonomy can identify questions for legal and conformity analysis, but it does not assign an EU AI Act risk class or a product-safety route. Classification depends on intended purpose, integration, role and applicable law. Use current primary regulatory text and qualified counsel for a shipping decision.
The practical value of the taxonomy is earlier alignment. Product, engineering, safety, operations and commercial leaders can see when they are discussing different products under the same "AI" label.
The Physical AI Product System turns that classification into decision gates across Strategy, Discovery, Productization, Launch and Scale. Use the Product Decision Review when authority, product boundary or evidence ownership is unresolved.
Source anchors and boundary
Generative, Operational and Physical AI are Hyperion product categories, not statutory or standards-defined classes. Article 3 of the EU AI Act defines an AI system partly through outputs that can influence physical or virtual environments and defines intended purpose in context. The NIST AI RMF Core frames risk through intended context, lifecycle, evaluation and management. Those primary sources anchor the authority-and-context lens; they do not convert this taxonomy into a legal classification.
