Aakash GuptaIf you can’t AI prototype after this, nothing will help you
CHAPTERS
- 0:00 – 1:40
Why most AI prototypes feel like “slop” (and why that’s fixable)
Aakash and Sachin open by addressing the common critique that AI-generated prototypes look impressive at first glance but aren’t shippable. They set the central goal of the episode: moving from generic, low-craft outputs to prototypes that actually help product teams make decisions.
- •Initial wow factor vs real product quality gap
- •Why “AI slop” is a useful warning sign, not a dead end
- •The episode’s promise: a repeatable path to higher-craft prototyping
- 1:40 – 5:17
How Anthropic (and Apple) “product shape” with prototypes before committing to a roadmap
Sachin contrasts the typical roadmap-first workflow with Anthropic’s prototype-first approach: build many prototypes, dogfood internally, then productionize the best. He frames this as "product shaping" and argues AI makes this approach affordable for far more companies.
- •Traditional flow: prioritize problem → then design solution
- •Anthropic flow: prototype many problem-solution pairs → test → productionize winners
- •Apple/iPhone example: prototyping in a lab to discover the right direction
- •AI reduces prototype cost enough to democratize product shaping
- 5:17 – 7:45
Diagnosing “AI slop”: generic styling, no differentiation, shallow scenarios
Sachin explains why a one-prompt CRM demo is still slop despite being functional. The issues are lack of craft, lack of differentiation, and lack of customer-specific workflow insight—and those are exactly what better prototyping skills address.
- •Generic visuals that resemble wireframes, not a real product
- •Me-too product patterns that can’t win against incumbents
- •Basic scenarios that may not match real user jobs-to-be-done
- •High-quality prototypes are possible, but require skill
- 7:45 – 11:37
The AI Prototyping Mastery Ladder: 15 skills from apprentice → journeyman → master
Sachin introduces a structured framework for becoming effective with AI prototyping tools. The ladder emphasizes fundamentals (prompting, editing, consistency), then higher-leverage practices (debugging, diverging, validation), and finally advanced outcomes (functional prototypes and product shaping).
- •15 distinct skills grouped into levels
- •Apprentice: prompting, editing, design consistency
- •Journeyman: versioning/debugging, diverging, customer validation
- •Master: functional prototyping + shaping roadmaps via tested prototypes
- 11:37 – 21:40
Design consistency via baselining: recreate your product first, then prototype on top
Sachin demonstrates a practical method to avoid generic outputs: start by recreating an existing product screen as a baseline, refine it with targeted edits, and reuse it as a template. This creates prototypes that inherit real visual style and component structure.
- •Start with “recreate this screenshot” for accuracy
- •Iterate with small edits to match real UI details
- •Batch related edits to reduce round-trip time
- •Save/duplicate the baseline as a reusable template for the team
- 21:40 – 26:00
From baseline to feature exploration: prototyping an “Ask AI” note chatbot in NoChoy
Using the NoChoy baseline, Sachin prompts an exploratory feature addition rather than writing a detailed spec. The tool produces multiple entry points and a coherent chat UI that matches the existing product’s design language, illustrating why baselines unlock higher-quality exploration.
- •Explore-style prompts for early-stage discovery
- •Multiple entry points (header + footer) generated automatically
- •Design inherits existing aesthetic because of the baseline
- •Prototype includes real interactions (chat flow, message alignment, stored conversations)
- 26:00 – 31:14
Diverging as a superpower: generating multiple solution directions fast (LinkedIn demo)
Sachin explains diverging—using AI to generate multiple competing design variants like a designer would. He demos Magic Patterns by recreating LinkedIn’s homepage and generating several “News in Your Network” layouts to compare and refine before customer testing.
- •Diverging = multiple variants, not one “best guess” output
- •Magic Patterns’ built-in variant generation
- •Different layout concepts (feed card, multi-card grid, sidebar module)
- •Best practice: narrow with designer/PM discussion, then test with users
- 31:14 – 34:39
Tool diversity = more creativity: same prompt in Bolt yields different designs
Sachin shows that even without explicit diverging features, tools can be prompted to generate multiple designs. He also highlights a subtle advantage: different tools’ system prompts produce meaningfully different outputs, so using multiple tools multiplies idea space quickly.
- •Prompting “explore multiple designs” can force divergence in generic builders
- •Different system prompts lead to different design directions
- •Using 2+ tools can yield 8+ variants quickly
- •More variants improves the odds of finding a winning solution
- 34:39 – 37:53
Making prototypes functional: deploying a real Ask AI experience with live model calls
At the master level, Sachin demonstrates a deployed, interactive prototype that queries real note content and calls OpenAI APIs. He adds a model selector to compare outputs across models—enabling product decisions that usually require engineering effort.
- •Publish prototype to a shareable URL
- •Prototype loads real content (articles/transcripts)
- •Live Q&A powered by OpenAI API
- •Model comparison UI (e.g., GPT-5 vs GPT-4.1) for product decision-making
- 37:53 – 40:14
API keys and secrets: how “vibe coding” tools handle security now
Aakash probes common concerns about embedding API keys in AI-built prototypes. Sachin explains that modern tools increasingly support secret storage patterns to prevent keys from leaking into client-side code, though implementations vary by platform.
- •Tools often request an API key and guide users to obtain/fund it
- •Secret storage workflows to avoid exposing keys in frontend code
- •Environment/secret vault approaches differ by tool
- •Earlier “keys in client code” pitfalls are being addressed
- 40:14 – 46:15
Customer validation at scale: surveys + PostHog analytics + session replays + heatmaps
Sachin shows how functional prototypes enable richer validation than static mockups: embed surveys, instrument analytics, and review session replays. He illustrates how heatmaps reveal which entry points users actually use, helping simplify the UI based on evidence.
- •In-prototype survey prompts to collect structured feedback
- •Integrate analytics (PostHog/Mixpanel/Amplitude) via prompts
- •Custom event tracking for key actions (buttons, closes, etc.)
- •Session replay + heatmaps to identify friction and dead UI elements
- 46:15 – 56:50
Discovery vs delivery: who prototypes, and why prototypes don’t replace PRDs
They discuss three operating models (PM-led, design-led, collaborative) and the danger of treating prototype code as production. Sachin argues prototypes are primarily for discovery, and PRDs still matter for strategy, hypotheses, metrics, and open questions.
- •Three team paradigms: PM-led, design-led, PM+design collaboration
- •Prototypes accelerate discovery; generated code isn’t usually production-grade
- •Claude Code/Cursor can align with existing codebases better than app builders
- •PRDs remain essential for strategy, differentiation, acquisition/monetization, hypotheses, and metrics
- 56:50 – 1:10:25
Choosing tools: lowest-friction adoption, market categories, and opinionated picks
Sachin maps the tool landscape into AI app builders, purpose-built prototyping tools, and AI coding tools for engineers. He advises teams to start with whatever is easiest to get approved, then upgrade as needs grow—ending with candid recommendations for what he’d buy.
- •Start with tools you already have access to (Figma/Google/Microsoft ecosystems)
- •3 categories: app builders vs prototyping-first tools vs AI coding tools
- •Tool tradeoffs: speed (Bolt), UI focus (v0), robustness (Replit), ease (Lovable)
- •Purpose-built prototyping tools: Reforge Build, Magic Patterns, Figma Make, Alloy
- •Hot take picks: Cursor (if technical), Magic Patterns as a fast on-ramp
- 1:10:25 – 1:12:39
Wrap-up: where to learn more and closing requests
Sachin shares where to follow his work and how to go deeper via his Reforge course. Aakash closes by encouraging subscriptions, ratings, and pointing viewers to his tools bundle.
- •Follow Sachin on LinkedIn for ongoing AI/PM insights
- •Reforge course: AI Productivity for PM workflows beyond prototyping
- •Episode recap emphasis: better discovery through prototypes
- •Closing calls-to-action (subscribe, review, bundle)
