How to Change Your Design Approach in the Age of Generative Technology
1. From HCI to HMIx
Generative AI changes the two-way relationship between humans and computers into a triadic relationship in which humans, models, and interfaces move together.
- In HMIx, it is difficult to design the interface separately from the model’s behavior.
- The model’s variability, where results can differ even with the same input
- Invisible interactions created as multiple models and agents collaborate
- The designer’s focus also expands from static screens to experiences that coordinate user intent and model behavior.
2. What Designers Need to Design Together
To create good generative experiences, you need to understand the conditions of both the interface and the model at the same time.
- Identify the technical possibilities and limitations of the model and agents.
- The criteria for success and failure of results, and unexpected behavior
- How the model or orchestrator exchanges information with the UI
- Connect the required inputs with the outputs users expect.
- How to supplement missing context with metadata and prior actions
- The flow for revising and trying again when results fall short of expectations
- Prototype early by placing the ideal experience, product requirements, and expected model behavior in parallel.
3. Prompt–Plan–Show–Next for Building Trust
Breaking complex generative interactions into four stages makes it possible to concretely check the user’s understanding and sense of control.
- Prompt: Collect the user’s intent and the context needed.
- Use choices, recommendations, and metadata to create a good first result with minimal input.
- Plan: Show the process and assumptions before execution.
- Let users confirm the approach, make adjustments, and then start multi-step work.
- Show: Reveal the process the model carries out and the basis for the result.
- Reduce uncertainty by showing why they need to wait and what information and references were applied.
- Next: Suggest actions that improve the result or lead to the next task.
- Provide natural iteration paths such as edits, regeneration, and setting changes.
4. Shifting to a Result-Centered Experience
Generative products are judged for quality and trust not by the polish of the screen, but by the entire process of reaching the result.
- The Adobe GenStudio team found that the existing design and handoff process using Figma alone made it difficult to handle rapidly changing model experiences.
- They designed an experience that more objectively expresses brand guidelines and lets users confirm the basis applied to generated results.
- Adobe Firefly supports iterative work by providing generation history and next actions such as style adjustments.
- The product experience includes not only good results, but also a flow for safely revising insufficient results.
- It becomes important to connect user intent, model confidence, recovery after failure, and next actions into a single end-to-end workflow.
