At a Glance

  • AI is changing the unit of product design from screens and features to goals, decisions, and workflows.

  • Adaptive interfaces can improve relevance, but they must preserve predictability and user control.

  • Agentic products need clear permission boundaries, confirmation patterns, and recovery mechanisms.

  • The competitive advantage will come from trusted integration into real work, not from adding a generic chat interface.

For most of software history, the division of labor was clear.

The product displayed options. The user learned the structure, entered information, navigated through screens, and initiated every meaningful action.

The software waited.

AI is changing that relationship.

A modern product can interpret natural language, remember context, predict a likely next step, assemble information, and increasingly perform a sequence of actions on the user’s behalf.

The product is no longer only an interface. It is becoming an operating partner.

That transition affects far more than the visible screen. It changes how teams define features, structure workflows, manage data, establish trust, and measure value.

From Navigation to Intent

Traditional interfaces require users to translate a goal into the product’s structure.

A person who wants to plan a business trip must know where to search for flights, how to compare options, where to apply policy constraints, and how to coordinate the final itinerary.

An AI-enabled product can begin with the goal.

It may clarify the budget, identify appropriate options, compare trade-offs, and prepare an itinerary. The interaction moves from navigation toward delegation.

This does not mean menus and forms disappear.

Structured controls remain valuable when choices need comparison, accuracy, accessibility, or repeatability. Natural language is useful for expressing intent, but it can be inefficient when a person wants to select one date from a calendar or compare five prices.

The strongest products combine conversational input with visible, editable structure.

From Fixed Screens to Adaptive Experiences

Most digital products display the same hierarchy to everyone.

AI makes greater adaptation possible. A system can adjust recommendations, explanations, shortcuts, and level of detail according to context.

A new user may receive guided steps. An experienced user may see direct controls. A customer facing an unusual problem may receive a different path from someone completing a routine task.

Adaptation can reduce effort, but it creates a new design risk: unpredictability.

If the interface changes without a clear reason, users may struggle to form a stable mental model. They may not know where a function went or why the product made a particular recommendation.

Adaptive design therefore needs a stable foundation.

Navigation, terminology, status, and essential controls should remain understandable. Personalization should improve relevance without making the product feel like it has different rules every time it opens.

From Answers to Actions

The first generation of generative products focused on producing information.

Agentic products can act across systems. They can create records, schedule meetings, update workflows, prepare purchases, or coordinate tasks.

Every increase in autonomy also increases the need for design discipline.

The product must make several things clear:

  • What it plans to do.

  • Which systems and information it will access.

  • Which actions require confirmation.

  • Which actions can be reversed.

  • What happened after execution.

  • What to do if the action fails.

A helpful assistant does not hide consequential work. It communicates intention and preserves the user’s authority.

Autonomy should be earned progressively. A user may first allow the system to prepare a recommendation, then draft an action, and later execute low-risk actions within defined boundaries.

From Personalization to a Data Relationship

More relevant products generally need more context.

They may use past activity, location, preferences, business information, or data from connected services. That context can create value, but it also changes the relationship between product and user.

According to Salesforce’s recent global customer research, 71% of customers said they were becoming more protective of their personal information even as personalization improved. See the connected-customer findings.

Trust depends on more than a privacy policy.

Users need to understand what information is being used, why it matters, and how they can change or remove it. The product should avoid collecting context it cannot connect to visible value.

In Indonesia, the Personal Data Protection Law provides the national legal foundation for handling personal data. Read Law No. 27 of 2022. Internationally, obligations such as the EU AI Act are also pushing AI products toward greater transparency.

Compliance defines a minimum. Product trust requires understandable choices.

The Five Layers of an AI Product

1. The Value Layer

Which customer outcome becomes meaningfully better?

If the answer is limited to “the product now has AI,” the value proposition is unfinished.

2. The Interaction Layer

How will users express intent, review output, edit assumptions, and recover from mistakes?

Conversation is one interaction pattern, not a complete experience strategy.

3. The Intelligence Layer

Which model or system capability is appropriate for the task? What are its known limitations? How will output be evaluated?

4. The Operational Layer

How does the capability connect to data, tools, business rules, human reviews, and exception handling?

This is where a demonstration becomes a working product.

5. The Trust Layer

How are permission, privacy, security, transparency, monitoring, and accountability handled?

Trust cannot be applied as a visual treatment after the system has been built.

Design for Calibrated Trust

Apple’s current guidance for generative AI experiences recommends telling people where AI is used, explaining capabilities and limitations, and offering curated suggestions for open-ended features. It also warns against making people believe they are interacting with human-authored content when they are not. See Apple’s generative AI design guidance.

This principle can be described as calibrated trust.

Users should not trust the product more than its performance deserves. They should also not reject a valuable capability because the experience fails to explain it.

Calibrated trust comes from:

  • Clear scope.

  • Visible sources when relevant.

  • Appropriate confidence signals.

  • Easy correction.

  • Confirmation for consequential actions.

  • Consistent behavior.

  • Honest communication about limitations.

A Leadership Agenda for AI Products

Product leaders should ask five questions before adding AI to a roadmap.

1. Which user goal becomes easier or newly possible?

2. What context does the system need, and why should users provide it?

3. What should the system recommend, prepare, or execute?

4. Where must a person remain in control?

5. How will the product learn without becoming unstable or intrusive?

These questions move the conversation away from features and toward the relationship between user and system.

The Interface Is Becoming a Contract

A button once represented a predictable command. The user pressed it and expected a defined result.

AI introduces interpretation. The system may infer, recommend, adapt, and act. This flexibility makes products more capable, but it also makes the relationship less obvious.

The interface now needs to communicate a contract: what the system understands, what it intends to do, and where the user retains authority.

The most successful AI products will not be those that appear most intelligent.

They will be the ones that help people achieve meaningful outcomes while remaining understandable, controllable, and worthy of trust.