Design AI-native interfaces that manage non-determinism, agency, and user trust.
npx skills add https://github.com/RefoundAI/lenny-skills --skill ai-native-ux
Please help me install this Agent Skill. Skill: ai-native-ux Source: https://github.com/RefoundAI/lenny-skills/tree/main/skills/ai-native-ux Command: npx skills add https://github.com/RefoundAI/lenny-skills --skill ai-native-ux
Designing user experiences for AI products is fundamentally different from traditional software because model outputs are probabilistic, user intent arrives as natural language, and autonomy must be balanced against human control. This Skill provides frameworks and verified insights from 14 product leaders to design interfaces that handle this fluidity without breaking user trust.
Help me design the user experience for an AI feature that suggests actions to users, including how much autonomy the AI should have and how users correct its mistakes.
High-intent search queries and answers about installing and using this skill.
Design fault-tolerant interfaces that account for the model's error rate, such as presenting multiple generated options instead of a single result. Add effortless correction mechanisms like feedback buttons so users can quickly pivot when outputs miss the mark.
Start with high-control, low-agency features where the AI suggests and humans act. Increase agency one version at a time only after performance is verified, since skipping to full autonomy removes visibility and destroys user trust when errors occur.
NLX (Natural Language Experience) is the design of invisible UI constructs for conversational interfaces, including prompts, editable agent plans, and proactive follow-ups. Unlike fixed buttons and forms, it requires explicitly designing the grammars that structure natural language interaction.
Replace blank prompt boxes with visual starting points, suggested prompts, and guided options so users understand what is possible. Pair this with post-generation tools that let users tweak and adjust AI output to their exact needs.
Inline AI suggestions should return in roughly 200 milliseconds to preserve the user's flow state, based on GitHub Copilot's experiments. Suggestions should appear as non-disruptive cues like italicized gray text within the user's existing tools.
Avoid high agency when the system has not been tested under high control, when errors carry high downside, or when you lack visibility into system behavior. Use low-downside patterns like drafts and recommendations instead of direct actions until trust is earned.
patternflyIdentify and apply AI UX design patterns to chat interfaces and generation workflows.
SpaceZephyrApply Karpathy's mental models to evaluate AI product decisions.
grasbergDesign and optimize prompts with structured schemas and evaluation frameworks.
imgamerOrchestrates an end-to-end game UI production workflow from UX specification to review.
