Aakash GuptaAakash Gupta

Give me 60 minutes, I'll make your AI Designing 81% Better

Aakash Gupta and Xinran Ma on design with AI workflows: prompt engineering, prototyping tools, divergent ideation.

Aakash GuptahostXinran Maguest
Feb 20, 20261h 1mWatch on YouTube ↗
AI design mind map: prompting, ideation, prototyping, conscious designClarifying the ask vs context engineeringLightweight PRD/spec for AI prototypingCustom GPT as a structured prompt generatorTool comparison: Lovable vs V0 vs Bolt vs Google AI StudioGoogle Stitch for divergent UI exploration (YOLO/variations)Risk mitigation: hallucinations, bias, empathy gaps, human-in-the-loop
AI-generated summary based on the episode transcript.

In this episode of Aakash Gupta, featuring Aakash Gupta and Xinran Ma, Give me 60 minutes, I'll make your AI Designing 81% Better explores design with AI workflows: prompt engineering, prototyping tools, divergent ideation Designing with AI is framed as an end-to-end workflow problem—constraints, system behavior, and iteration—not just writing better prompts.

At a glance

WHAT IT’S REALLY ABOUT

Design with AI workflows: prompt engineering, prototyping tools, divergent ideation

  1. Designing with AI is framed as an end-to-end workflow problem—constraints, system behavior, and iteration—not just writing better prompts.
  2. Xinran’s “AI Design Mind Map” breaks the space into prompting, ideation, design/prototyping workflows, and a less-discussed layer: conscious, risk-aware design.
  3. Workflow 1 demonstrates using a Custom GPT to interview you for key inputs and output a lightweight, front-end-focused “PRD for prototyping” that can be pasted into tools like Claude, Lovable, or V0.
  4. Workflow 2 demonstrates using Google Stitch for divergent UI exploration (including “YOLO mode”) and exporting selected designs into Google AI Studio to add interaction and iterate.
  5. The episode compares popular prototyping tools (Lovable, V0, Bolt, Claude Artifacts, Cursor, Google AI Studio) and explains tradeoffs in design quality, speed, learning curve, and cost.

IDEAS WORTH REMEMBERING

5 ideas

Design with AI is a workflow and systems problem, not a prompting trick.

Xinran emphasizes understanding the system, constraints, and desired behaviors end-to-end; prompting is only one interface layer in a broader design process.

Start by clarifying the ask before adding context.

High-quality outputs come from stating what you want, what to include/avoid, and then adding only the context necessary to reach that outcome—more context is not automatically better.

Use a Custom GPT to eliminate the “blank canvas” loop.

Instead of vibe-coding aimlessly, the Custom GPT asks structured questions (users, needs, platform, key flows) to produce a ready-to-paste spec that anchors the first prompt.

Write “PRDs for prototyping,” not full PRDs.

The spec should prioritize front-end screens, components, and interactions while excluding backend architecture and success-metric filler that can dilute model attention and cause “context rot.”

Run a cheap ‘mock pass’ in Claude before spending on heavier prototyping tools.

Xinran uses Claude as a quick visualization sanity check to catch prompt issues early, then moves to tools like Lovable/V0 for more polished results.

WORDS WORTH SAVING

5 quotes

Designing with AI isn't about prompting. It's about understanding the entire workflow, the system, the constraints, and the behaviors.

— Xinran Ma

Not more context means better, but the necessary context that is related to the goal.

— Xinran Ma

AI is not human, right? It does not really have enough empathy.

— Xinran Ma

I like to shift it to ChatGPT in order to save tokens for Claude.

— Xinran Ma

You can make it very refined or YOLO, which is going crazy.

— Xinran Ma

QUESTIONS ANSWERED IN THIS EPISODE

5 questions

Can you share the exact system instructions/components of your Custom GPT that prevent ‘sign up/log in first’ from appearing as a recommended flow?

Designing with AI is framed as an end-to-end workflow problem—constraints, system behavior, and iteration—not just writing better prompts.

What’s your rule-of-thumb for how many screens and how much interaction detail a “PRD for prototyping” should include before it becomes too heavy?

Xinran’s “AI Design Mind Map” breaks the space into prompting, ideation, design/prototyping workflows, and a less-discussed layer: conscious, risk-aware design.

When you say Claude is a ‘mock run’ tool, what specific failure signals tell you the prompt/spec needs revision (vs the tool/model being the problem)?

Workflow 1 demonstrates using a Custom GPT to interview you for key inputs and output a lightweight, front-end-focused “PRD for prototyping” that can be pasted into tools like Claude, Lovable, or V0.

In Stitch, what prompt patterns reliably increase divergence without blowing up the design system (e.g., YOLO mode causing irrelevant page-wide changes)?

Workflow 2 demonstrates using Google Stitch for divergent UI exploration (including “YOLO mode”) and exporting selected designs into Google AI Studio to add interaction and iterate.

How would you adapt the Custom GPT workflow for redesigning an existing product (starting from screenshots/Figma) instead of a blue-sky idea?

The episode compares popular prototyping tools (Lovable, V0, Bolt, Claude Artifacts, Cursor, Google AI Studio) and explains tradeoffs in design quality, speed, learning curve, and cost.

EVERY SPOKEN WORD

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome