From dynamic reports to AI-powered experience systems.
Rethinking product design in the age of generative interfaces - how AI can generate user experiences dynamically while preserving business logic, permissions, data integrity, and design consistency.
For years, product teams have focused on helping users configure software. When users want to create a report, they select data sources, choose metrics, configure filters, select chart types, and arrange layouts. The software provides building blocks. The user assembles the experience.
AI is fundamentally changing this interaction model. Instead of configuring interfaces, users can now express intent:
"Create a report showing jobs created this month, new customers acquired this month, task distribution by team member, and revenue trends over the last three months."
At first glance this looks like a dashboard generation feature. In reality, it exposes a much larger design challenge: how can AI generate user experiences dynamically while preserving business logic, permissions, data integrity, and design consistency?
The configuration burden.
Traditional reporting workflows place the assembly burden on the user. Every step - from data source selection to layout - requires manual configuration. This creates friction, a high learning curve, and poor accessibility for non-technical users.
Most AI-powered reporting products stop at prompt interpretation. User prompt goes in; dashboard comes out. This approach works for demos. It breaks down in real products.
Questions quickly emerge: Which metrics are available? Which visualizations are appropriate? What permissions should apply? How should widgets be arranged? What happens when the request is ambiguous?
AI cannot safely answer these questions without understanding the product itself. The challenge is no longer generating UI. The challenge is generating valid experiences.
- Metric availability.AI doesn't know which data sources are connected or which calculations are defined for this workspace.
- Permission constraints.A generated dashboard may expose data the requesting user isn't authorized to see.
- Visualization appropriateness.Without rules, AI may select charts that misrepresent the data or contradict business conventions.
- Ambiguous intent.Prompts are underspecified. Without a structured model of the domain, disambiguation is guesswork.
Building the Experience System.
To support AI-generated reports, four foundational models were designed - together they form the Experience System layer that guides AI without removing its generative power.
A shared vocabulary for reasoning about reports.
Instead of treating reports as visual elements, reports are modeled as structured business objects. This gives AI the concepts it needs to interpret intent accurately.
- Data Sources - Jobs, Customers, Revenue, Tasks
- Metrics - Count, Revenue, Completion Rate
- Dimensions - Date, Team Member, Status
- Visualizations - KPI Card, Line, Bar, Table
Chart selection as constrained decision-making.
AI should not arbitrarily select charts. A mapping layer defines relationships between data structures and appropriate visual representations - turning generation into governed choice.
- Revenue Trend → Line Chart
- Category Distribution → Pie Chart
- Team Comparison → Bar Chart
- Single Metric → KPI Card
Layout hierarchy that survives any prompt.
Generating widgets is not enough. AI must also understand layout hierarchy - so the result is a dashboard that remains coherent and consistent regardless of prompt complexity.
- KPI metrics appear first
- High-priority insights occupy dominant positions
- Related metrics are grouped together
AI as a trustworthy system, not a creative generator.
AI-generated experiences require governance. Without it, AI can generate technically valid but operationally invalid experiences. Governance defines the boundaries of what can be generated.
- Available data constraints
- Permission-based visibility rules
- Business rules and restricted actions
Reports were never the point.
The dynamic report problem was never really about reports. Reports are one type of experience. The same architecture - ontology, decision models, composition rules, governance - applies to dashboards, alerts, recommendations, agents, workflows, and summaries.
Traditional software is built around screens. AI-native products are built around intent. The interface becomes an output rather than the primary artifact.
In the traditional model: Page → Component → Interaction.
In the AI-native model: Intent → Context → Experience → Interface.
This is not a feature change. It is a change in what product design produces. Designers are no longer primarily crafting screens - they are architecting the systems that determine what experiences AI can generate, under what conditions, and governed by what rules.
Future users will not request charts. They will express business objectives:
- "I want to improve sales performance."The system determines required insights, workflows, agents, dashboards, and recommendations.
- "I want to reduce customer churn."The system generates the most appropriate combination of experiences automatically.
- "I want to identify operational bottlenecks."Intent maps to experience architecture, not to a specific screen.
The future of product design is not about generating interfaces. It is about designing systems that enable AI to generate trustworthy experiences.
As AI becomes a first-class participant in product experiences, designers must move beyond screen design and begin architecting the systems that shape how experiences are generated.
-
01
Experience Ontologies. A structured vocabulary of experience types, their purposes, and their relationships - the conceptual layer AI reasons over.
-
02
Intent Models. Frameworks that map user objectives to appropriate experience configurations - bridging natural language and product logic.
-
03
Context Architectures. Definitions of what contextual signals - user, role, moment, history - should influence which experiences are generated.
-
04
Governance Systems. Rules determining what AI can and cannot generate, owned by the design system, not bolted on afterward.
-
05
AI-readable Design Systems. Design systems that carry semantic meaning alongside visual definition - readable and actionable by language models.
If you're working on AI-native product design - particularly at the intersection of generative interfaces and enterprise governance - I'd like to hear from you. Get in touch →