AI product design

Shipping AI features requires more than a prompt. Most AI products fail because they lack clear capability boundaries, fail to handle probabilistic errors, or strip users of control. This skill provides a rigorous framework for designing the human-system relationship in AI assistants, copilots, and autonomous agents. It moves beyond "chat" to define specific interaction patterns for transformation, recommendation, and execution.

What it does

• AI feasibility testing determines if a task requires probabilistic AI or if a deterministic rule-based system is faster and safer.
• Risk mapping identifies failure modes, consequences, and required human intervention levels for every AI role.
• Interaction selection maps tasks to the correct UI pattern, from inline assistance to batch workflows or agentic execution.
• Agentic control specification defines the permission, preview, and audit trail requirements for multi-step system actions.
• Trust and evidence design builds UI treatments for citations, uncertainty, and source verification to prevent overtrust.

How it works

1. Analyze the use case to justify AI value against latency, cost, and the burden of human review.
2. Map the capability contract to separate suggestions and drafts from high-impact system executions.
3. Design the lifecycle using HAX principles to ensure users can correct, undo, or escalate when the system fails.
4. Build the evaluation plan to measure task outcomes and recovery rates rather than just model benchmarks.

Frameworks & tools

This skill utilizes HAX (Human-AI Interaction) principles, lifecycle mapping, and claim-control matrices. It is model-agnostic and applies to any LLM-powered application or autonomous agent architecture.

Why this beats prompting it yourself

Generic prompts often result in vague UI suggestions or chat-centric designs that ignore risk. This skill enforces a structured specification process that covers edge cases, permission expansion, and verifiable evidence—details usually forgotten until after a failure.

Use cases

• Designing an autonomous agent that handles financial transactions or sensitive data.
• Building a coding copilot that needs to balance automation with user review.
• Creating a recommendation engine where users need to understand the "why" behind an output.
• Developing a batch processing system for high-volume content classification.

Known limitations

This skill focuses on product design and interaction logic. It does not provide model training code, backend architecture implementation, or prompt-only creative writing.