What is the ai-native-ux skill?
The ai-native-ux skill is a SKILL.md in the RefoundAI lenny-skills repository. It turns insights from 14 guests on Lenny's Podcast and Newsletter into four guided moves: map agency against control, design for non-determinism, structure conversational grammars and iterate on feedback loops. It is advisory guidance, not a code generator.
The ai-native-ux skill sits in the skills folder of RefoundAI's lenny-skills repository. Its description reads: help users interact with probabilistic models by designing interfaces that manage fluidity, intent and agency while keeping trust and control.
Most principles are anchored to a named guest from Lenny's Podcast or Newsletter, such as Aparna Chennapragada on natural language interfaces and Gustav Söderström on fault-tolerant interfaces. A references folder holds the full guest insights and a list of templates.
How it shapes a design conversation
- The agent asks how the interface behaves when the model answers wrongly or with low confidence.
- It asks whether visual starting points help people past a blank prompt box.
- It checks whether response latency is fast enough to protect flow state.
- It asks how AI-generated suggestions are told apart from content the user wrote.
The seven principles
- Design for fluid intent: weigh the flexibility of language against the precision of buttons and forms.
- Define decision boundaries: decide what the algorithm owns and what people own.
- Build structured grammars so natural language is not left entirely to the model.
- Make correction effortless with simple, high-frequency feedback.
- Optimize for instruction following, including multi-step and multi-modal requests.
- Climb an agency-control ladder: start with high control and low agency, then raise autonomy once behavior is verified.
- Preserve flow state with non-disruptive cues such as italic gray inline suggestions.
Mistakes the skill tells the agent to flag
- The Empty Prompt Trap: expecting users to know what to type without visual starting points.
- Over-building rigid UI that cannot keep pace with model improvements.
- Black Box Autonomy: raising agency before human-in-the-loop controls exist.
- Ignoring non-determinism and treating the model as a fixed decision engine.
Who the ai-native-ux skill is for
Best for
- Product designers and managers planning chat, suggestion or agent features.
- Teams deciding how much autonomy an AI feature gets in its first version.
- Reviewing a prototype for dead ends when the model returns something unexpected. Pair it with interface-review for screen-level checks.
Not for
- Producing finished UI code or visual styling. For a build-first design skill, see frontend-design.
- Measuring model quality. The skill lists AI Evals as a related skill rather than covering it.
- Wording the interface itself. ux-copy sits closer to that job.
Use it with UiChemy
This skill shapes the thinking behind an AI interface, not the screens. Once your agent has built them, route the layout through Figma or UiChemy Composer and convert it to an editable WordPress page. Click Get Started on uichemy.com.
How to install AI-Native UX Skill: Design Interfaces Around AI Output
git clone https://github.com/RefoundAI/lenny-skills.gitClone the repository
Run
git clone https://github.com/RefoundAI/lenny-skills.git. The skill sits in the skills/ai-native-ux folder.Copy the folder into your project
The README shows the pattern
cp -R lenny-skills/skills/writing-prds .claude/skills/. Swap in this skill:cp -R lenny-skills/skills/ai-native-ux .claude/skills/.Or grab it individually
The README also points to per-skill download links on the vendor site at refoundai.com/lenny-skills, if you want one folder without cloning.
Ask a matching question
Describe an AI feature you are designing, for example how much autonomy an assistant in your product should get.
Check that it loaded
Ask the agent how to balance agency and control in an AI feature. The answer should draw on the skill, for instance by starting with high control and low agency.
Key features
- Four guided movesThe How to Help section covers mapping the agency-control balance, designing for non-determinism, structuring conversational grammars and iterating on feedback loops.
- Guest-sourced principlesSeven core principles, most tied to a named speaker such as Kevin Weil, Ryan J. Salva or Adriel Frederick, drawn from 14 guests and posts.
- Agency-control ladderStart with small, testable features that have high control and low agency, then raise autonomy one version at a time after performance is verified.
- Named templates and frameworksIncludes NLX design elements, fault-tolerant user interfaces, low downside design patterns for agents and a 200 millisecond latency sweet spot for inline suggestions.
- Review questionsSix prompts the agent can put to you, from the human and AI control split to how suggestions are distinguished from authored content.
- Common mistakes listFlags the Empty Prompt Trap, over-built rigid UI, black box autonomy and ignored non-determinism.
- References folderA references subfolder holds artifacts.md with the full template list and guest-insights.md with all 14 sourced insights.
- Related skills namedThe skill points to Ai Product Strategy and Ai Evals as neighbours in the same library.
Frequently asked questions
What is the ai-native-ux skill?
It is a skill in the RefoundAI lenny-skills repository, built from 14 guests and posts across Lenny's Podcast and Newsletter. It helps an agent reason about interfaces for probabilistic models: where control sits, how to handle variable output and how to keep trust while giving users room to steer.
How do I install the ai-native-ux skill?
Clone RefoundAI/lenny-skills, then copy the skills/ai-native-ux folder into your project's .claude/skills directory. The README documents that cp -R pattern using another skill as the example. Afterwards, ask about an AI feature you are designing, such as an assistant, a suggestion panel or an agent workflow.
Which AI coding agents can use the ai-native-ux skill?
The README says Claude Code, or any agent that reads SKILL.md files, applies the skills when you work on a matching task. The repo topics also list claude-code. Beyond that, it does not name individual agents, so test it in yours.
Is this a design skill or a product management skill?
It belongs to a product management library. The README files it under Playbook: Building AI Products, next to AI Product Strategy and AI Evaluation Strategy. Expect product and interaction guidance, such as decision boundaries and feedback loops, rather than visual styling advice or component code.
What is the agency-control ladder?
It is a scaling idea quoted from a post on AI product development. Begin with features that are high control and low agency, small enough to test and observe. Then increase autonomy gradually, one version at a time, only after performance is verified in low-stakes settings.
Does the skill generate interface code?
No. The SKILL.md holds principles, templates, review questions and common mistakes, and it tells the agent how to help you think. If you want screens built, use a build-oriented skill such as frontend-design and let this one inform the decisions.
What license does the ai-native-ux skill use?
The repository is MIT licensed, both in the README badge and in GitHub's license field. The README adds an attribution note: the insights, quotes, frameworks and templates remain the intellectual property of Lenny Rachitsky and his guests, so keep that credit when you share material.