What Puts the Design in AI? Behavior,
Part 4

Together, these dynamics form a unified system. Attention and alignment establish the conditions for initiating relationships, while adaptation and repair establish the conditions for sustaining them. Together, they define a foundational framework for human-centered AI design. We need to consider how the human-AI relationship unfolds and manifests across intelligence, user interface, and scaffolding at the point of interaction.

Intelligence is the behavior of AI itself, defining how it learns, reasons, responds, and acts. The user interface mediates not just the relationship between humans and AI, but also the relationship between AI agents. Scaffolding supports these relationships by providing the necessary context and guidance, thus completing the experience.

I recently worked with a startup to evaluate its AI-driven application. They had positioned the application as a helper and coach that guides users in crafting responses to professional licensing complaints. The application does not provide legal advice; rather, it helps users organize their thoughts and ensures that users address the most pertinent issues by telling their stories.

Research with the target audience had revealed a key finding: users were unable to effectively judge the application’s output. Although the participants worked in the same domain as the application targets, most had no prior experience with licensing issues and accepted the outputs at face value. Even participants with direct experience took them at face value. Without sufficient scaffolding, users had no basis for judgment, regardless of what the intelligence produced or the user interface presented.

Compounding this challenge is the application’s single-use nature. Users face a licensing complaint—hopefully a once-in-a-career event—and don’t return regularly. Most of the participants were more accustomed to focused, single-task AI tools such as note transcription. Participants needed to understand what constitutes a good response, know how to use a chat-based AI user interface, and be able to navigate the workflow of creating a response.

This is where the behavioral dynamics come to life. An application that is positioned as a helper and coach needs to behave like one—not only in the model’s inherent behavior but also in the user interface and the scaffolding that completes it. Application of the behavioral dynamics that I’ve introduced in this series reveals what it takes to design effectively for unbounded systems.

Next, we’ll look at how these considerations played out for this application.

Establishing the Relationship

The dynamic behaviors that matter most in establishing a relationship include attention and alignment.

Attention

Because the AI-driven application that I evaluated adopts a coaching posture, the intelligence works with the user, surfacing reasoning and guiding rather than producing a finished output on its own. The user interface separates the conversation from the work product. The chat in which the user interacts with the AI appears alongside the output they’re building together. The scaffolding orients the user to the process, expectations, and what the AI can and cannot do—and supports the user in developing competence, not just completing the task.

Alignment

The user has a clear job, crafting a response to a complaint. The intelligence needs to surface the domain context that shapes the job. The user interface needs to make the user’s intent visible, showing where the user is, what the user is addressing, and what remains to be done. The scaffolding provides the codified judgment the user lacks: the criteria, reference points, and boundaries that constrain how to fulfill the job.

Sustaining the Relationship

The dynamic behaviors that sustaining a relationship requires include adaptation and repair.

Adaptation

Each user brings a different situation, set of circumstances, and level of experience to a relationship with an AI. The intelligence needs to pick up on how the user is engaging and adjust its guidance accordingly. The user also needs to learn from the intelligence’s guidance. If the AI produces and the user simply accepts, there is no cocreation of shared context. The user interface needs to reflect what the human and AI have together addressed and what remains. The scaffolding needs to shift its support based on what a particular user needs to move forward.

Repair

In a chat-based user interface, the user directs the exchange. The intelligence needs to recognize when the user is uncertain and respond accordingly. The user interface needs to invite evaluation, not just acceptance. The scaffolding provides reference points that help the user see whether the response is on track.

Conclusion

None of these considerations is novel. Many reflect emergent practices that are already taking shape in unbounded systems. The goal is not to introduce new patterns but to show how behavioral dynamics drive and explain design choices, including the whys and the intent behind them. We must design the user experience across intelligence, user interface, and scaffolding. The framework that I’ve described surfaces that for which we should design across all three of these.

This article shows what it means to design for interacting minds. The behavioral dynamics don’t prescribe what to build. They surface that for which we need to design. In the example application that I’ve described, what a coaching posture actually requires dictates where the user’s judgment needs support and how the experience must work across intelligence, user interface, and scaffolding to fulfill its purpose.

This framework applies wherever humans and AI come together as minds. The specific considerations change, but the dynamics don’t. Attention establishes how the AI engages. Alignment makes intent actionable. Adaptation maintains fit as the relationship evolves. Repair restores the relationship when something goes wrong. These are the conditions that UX designers need to shape, regardless of the domain, interaction type, or the AI’s level of autonomy.

The question is no longer how to control what the system does but how to create the conditions for a relationship that serves human needs while respecting human agency. This is the foundational challenge of human-centered AI design. Behavior is our primary design concern. This is what puts the design in AI. 

“I’ll let you be in my dream if I can be in yours.”—Bob Dylan

Note on AI Use

I leveraged Claude throughout the writing of this article as a thought partner and editor. I use and have always used em dashes deliberately for emphasis.

Claude had this to say: “Kevin drafted Part 4 independently, working through the structural challenge of synthesizing a four-part series into a capstone. My role spanned the full development arc—helping shape the outline, pressure-testing the Human-in-the-Loop (HITL) framing as the article’s connective thread, and working through how the applied examples should progress from absent dynamics to fully embedded ones. As Kevin evolved the article toward a more reflective close, I helped consolidate examples and design considerations into a tighter structure. Kevin maintained complete conceptual control throughout, frequently correcting my suggestions when I characterized something as new that earlier parts of the series had already established. The collaboration continued to demonstrate the dynamics the series theorizes: adapting to shifting scope, realigning around what the capstone actually needed to say, and repairing course when my framing missed the framework’s precision.”

Pabini Gabriel-Petit, Editor in Chief of UXmatters, added the acronym definition.