The standard chatbox UI has become a habit but fails to expose the complex internal state of AI agents, creating a low ceiling for interaction.
AI systems violate long-held user expectations of predictability and permission, requiring a new design grammar for collaboration instead of command.
An overview of 12 design patterns, grouped into four clusters, that can help make AI legible and build user trust beyond the chat window.
Use confidence lenses and evidence trails like citations or tool traces to visually communicate the AI's certainty and the source of its information.
Implement patterns like verification, branch and merge, autonomy dials, and budgets to give users control over where the AI can proceed without intervention.
Use correction and editable memory patterns to align the AI with user preferences and context, ensuring it learns from interactions over time.
Employ replay summaries, streaming updates, temporal handoffs, and detailed receipts to provide a clear audit trail of the AI's work.
Real-world AI workflows combine multiple patterns, and it is crucial to design these interactions to be accessible to all users, not just visually.
A look at emerging patterns like spatial layouts for memory, plurality for showing disagreement, and step-through debugging for complex workflows.
The core job of an AI designer is to shape the human-machine relationship by exposing the model's hidden states and building a foundation of trust.
Jobs that call for the skills explored in this talk.