Prototyping with Generative AI

Product managers are responsible for shaping new products and features, and prototyping is a core skill because it helps align stakeholders and communicate what is being built. In the past, prototyping often required engineers or designers and took significant time.

Today, AI tools allow product managers to prototype quickly on their own, and this is increasingly expected.

When interviewers assess prototyping skills through vibe coding, they mainly use two question types:

1. Design X for Y, which is a classic product design question but you are allowed to use AI for prototyping
2. Use AI to solve for X

Interviewers look for:

• Clear problem framing and the ability to handle ambiguity
• Use of real user insights
• Concrete, visual, and interactive prototypes
• Awareness of tradeoffs and constraints

Using AI prototyping to impress

AI prototyping is more than just faster design. Here are some ways you should use the AI tools at your disposal to give you an edge in product design interviews.

Instant replication of familiar UI patterns

Search for “Cash App transaction UI,” “Binance trading card,” or “Coinbase onboarding.”

Drop a screenshot or link into Lovable → it generates screens with similar patterns.

Interviewers instantly recognize the UX and can focus on product decisions instead of getting bogged down with UI intricacies.

Real-time LLM interaction via your own API key

Connecting your own LLM API key (e.g., OpenAI) to your prototype means your demo responds with actual AI reasoning, not static text.

The prototype feels alive and much closer to a real product than slides or wireframes.

Collaborative iteration with the interviewer

AI tools like Claude allow you to iterate on prototypes quickly. Use this to your advantage and treat the interview as a co-creation session rather than a presentation. For example:

• “Let’s simplify this onboarding step.”
• “Let’s try a different risk explanation.”
• “Let’s adjust messaging for a different segment.”

Making live changes demonstrates flexibility, good product judgment, and strong collaboration skills.

Vibe-coding tricks that make you stand out

Micro-skills interviewers remember:

• Auto-generate UI from screenshots or URLs
• Use LLM to generate flows from user stories
• Pull live data (e.g., crypto prices, balances)
• Generate UX copy on the fly

These tricks let you move faster and look extremely competent without heavy engineering.

AI vibe coding framework

A repeatable approach for design, system, and AI product questions.

Step 1: Clarify the prompt

Ask these 3 high-leverage questions:

1. What problem are we solving?
2. For whom?
3. Under what constraints?

Step 2: Define the motivation

Use AI to surface quick context. For example:

• “Why is Gen Z entering crypto?”
• “What causes drop-off in fintech onboarding?”

These insights support a grounded hypothesis.

Step 3: Segment users

You can use AI-assisted TAM / persona generation. Prompt example:

“Segment US crypto users by age, risk tolerance, investing frequency, and motivations. Rank segments by opportunity size.”

Pick one primary segment and build around it.

Step 4: Identify pain points & form a hypothesis

Example hypothesis:

“Gen Z values transparency and simplicity, but current crypto apps feel opaque and risky. We need a UX that explains risk clearly, builds trust, and reduces friction.”

This hypothesis guides your prototype decisions.

Step 5: Build the AI prototype live

Use Lovable + your LLM to:

• Build the core flow
• Focus on the single most important interaction
• Narrate tradeoffs and decisions while you build

Step 6: Co-create with the interviewer

Treat feedback as collaboration. You can take the interviewer’s feedback on:

• Changing user flows
• Adjusting copies
• Simplifying or deepening logic

And make changes to your prototype on the spot.

Step 7: Transition to production thinking

Even in a mock interview, close with real-world execution. This is a great seniority signal to interviewers.

• Instrumentation & observability: What would you log and track (funnel metrics, errors, model performance, feedback signals)?
• AI-specific risks and fundamentals:
  • When you might use RAG (external knowledge vs. pure LLM)
  • How you handle hallucinations and safety boundaries
  • How memory / context is managed for users
  • How you’d monitor and debug (logs, traces, evaluation)
• Validation & evaluation: How you’d test correctness, quality, and user satisfaction.
• Guardrails & compliance: Especially for fintech/healthcare or kids: limits, disclosures, approvals, data privacy.
• Rollout & monitoring: Alpha → beta → phased rollout, with clear rollback plans.

Instead of simply focusing on UI, you want to demonstrate that you understand AI fundamentals (RAG needs, observability, memory, safety, hallucination risks) while you’re vibe coding.

How to differentiate yourself

The top 1% of candidates do these:

• Prototype only the critical interaction, not the entire app
• Use LLMs to generate UI and explain their reasoning as it appears
• Focus on decision-making, not pixel perfection
• Introduce metrics early (“Here’s what we’d track on day one…”)
• Communicate constraints clearly (“We won’t build X in v1; too complex”)
• Stay structured and calm under ambiguity
• Bring AI fundamentals into the discussion (data, RAG, observability, safety)

You would want to avoid the following mistakes:

1. Overbuilding. It is unnecessary to build a full backend or entire app.
2. Focusing on framework over product. Interviewers expect to see prototypes, and not just you walking through popular PM frameworks like CIRCLES or AARM.
3. No narrative. A prototype without a story = chaos. A story without a prototype = generic.
4. Ignoring metrics. Interviewers won’t understand your design decisions if they don’t understand what constitutes success.
5. Being a perfectionist. When doing prototyping on the fly, you will definitely not be able to craft a perfect prototype, nor should you aim to. Agility and the ability to iterate live is what interviewers are looking for.

Recommended workflow

Use a simple, reusable tech stack:

• Notion / Google Doc: Structure your thinking, capture tradeoffs, and add roadmap/strategy.
• GPT-5 (or latest LLM): Ideation, flows, copy, segmentation, insight generation.
• Lovable + your LLM API key: Rapid UI, real-time AI responses, interactive flows.
• N8N (optional): Simulate simple backend logic (auth → API → callback).

Always finish with your “Close the Doors” mindset:

• If this went to production, here’s what I’d instrument.
• Here’s what I’d guardrail and monitor.
• Here’s how I’d validate and experiment.
• Here’s how I’d roll this out safely.

Practical templates

The following two templates cover the majority of product design questions you’ll be asked to prototype.

1. Standard product design (“Design X for Y”)

Example prompt:

Design a crypto app for Gen Z.

1. Clarify prompt
2. Why now? (insight from AI or domain knowledge)
3. Segment users (AI-assisted)
4. Pain points + hypothesis
5. Build prototype live (core flow)
6. Co-iterate with feedback
7. Add production considerations:
   • Metrics
   • AI risks/guardrails (if relevant)
   • Rollout plan

2. AI product question (“Use AI to solve X”)

Example prompt:

Use AI to reduce user friction or increase trust in an investing app.

1. Define data inputs & sources
2. Choose AI approach: retrieval, generation, classification, or hybrid
3. Define boundaries: what AI does vs. what needs human/guardrails
4. Prototype the chat or task flow with LLM + UI
5. Validate assumptions live with interviewer feedback
6. Add evaluation & observability:
   • What you measure for quality
   • How you detect hallucinations or bad outputs
   • How fallback flows work (e.g., human review, simpler logic)

There is a tendency to overbuild for take-home assignments. Follow this to ensure that you deliver a quality assignment without spending excessive amount of time on it: