Lesson Content

Design variant generation.

In the design phase, AI is most useful as an option-generator. Given a design problem – "I need a user onboarding flow for a B2B SaaS product" – AI can rapidly generate five to ten alternative approaches with different interaction models, scope levels, and user assumptions. This expands the design consideration space before the team anchors on a single approach.

Effective use: generate five to ten variants, review for relevance to your actual user and constraints, then select or combine the most promising elements. AI expands the option set; the designer applies judgment to select and refine.

Architecture Decision Records (ADRs) with AI.

An ADR documents a significant architecture decision – the decision made, the context that drove it, the alternatives considered, and the trade-offs accepted. AI can draft ADRs from a conversation about the decision:

1. Describe the decision and the context
2. Ask Claude to draft an ADR covering: context, decision, alternatives considered, trade-offs, consequences
3. Review the draft for accuracy and completeness
4. The builder writes the judgment section: why this was the right call

ADRs with AI assistance are more likely to actually be written – reducing the documentation debt that accumulates when decisions are made faster than they can be documented manually.

Rapid prototyping with AI.

For web-facing products, AI code generation produces low-fidelity prototypes from a description significantly faster than manual coding. Use case: produce a clickable prototype of a proposed interface to test with users before building the real thing. The code is throwaway – the learning is the point.

Ask Claude to generate prototype code with explicit throwaway framing: "This is a prototype to test with users – not production code. Optimize for speed and clarity of the concept, not for scalability or security."

Stress-testing design decisions.

Before committing to a design decision, run a stress-test conversation: "I'm planning to [describe design decision]. What are the failure modes of this approach at scale? What assumptions does this design make that might not hold? What would a user who is not the typical case experience?" These questions reveal fragilities before they cost time to fix.