Aakash GuptaI Should Be Charging $999 for This AI Prototyping Masterclass
CHAPTERS
AI won’t replace developers—unless you already understand how to build
Nadav frames AI as a powerful tool rather than a replacement for engineering. The core idea: AI amplifies existing capability, so people who already understand how software is built will get the most value from AI-assisted building.
- •AI is not an “anyone can build anything” magic wand; it’s bounded by user understanding
- •If you can build without AI, you can build faster with AI; if you can’t, you’ll struggle to ship production-quality work
- •The knowledge gap (technical fluency) is shrinking, but still matters today
- •PMs are positioned to benefit heavily because they constantly iterate on ideas and communicate specs
When PMs should use AI prototyping: ideation, validation, and speed-to-learning
They lay out why AI prototyping tools are becoming a default part of PM workflow. The value is functional, testable experiences that dramatically reduce the cost of experimentation compared to traditional prototype-building.
- •AI prototyping is useful across many contexts, from tiny internal tools to product features
- •For PMs, AI becomes a “virtual developer” to unblock experimentation
- •Functional prototypes let users actually interact, not just look at screens
- •AI makes functional prototyping cheap enough to do for nearly every feature during ideation
- •Prototyping can also happen after Figma to reach a more production-like feel before building
Avoiding the trap: problem space vs solution space (research still matters)
Aakash raises the concern that rapid prototyping can push teams prematurely into solutions. Nadav agrees: strong research, user understanding, and defining the feature shape should come before generating prototypes.
- •Prototyping tools (AI, Figma, vibe coding) inherently bias toward solution space
- •PMs should not skip research, user conversations, and competitive/market scanning
- •By the time you focus on visuals, you should already know the problem, user story, and feature intent
- •Use AI for research too—before using AI for building
- •Treat AI prototyping as a tool with a place in the lifecycle, not a hammer for everything
Live demo kickoff: recreate an existing product screen as your starting template
Nadav demonstrates a key workflow: start from a screenshot of the existing product (LinkedIn) to generate a high-fidelity baseline. This creates a reusable foundation that keeps prototypes visually aligned with the real product.
- •Drop a screenshot, prompt the tool (Dazzle) to recreate the page, then build from there
- •Dazzle spins up a server and generates real app code, not just static mock screens
- •Starting from “your product” prevents blank-page prototyping and improves context
- •Reusable templates can become an org-wide accelerator (design-system-like starting point)
- •Keeping the template updated over time helps prototypes stay close to production UI/UX
Under the hood: design system CSS, code generation, and validation steps
They pause to interpret what Dazzle is doing technically (theme CSS, components, validation). The emphasis is that PMs benefit from understanding the basics of what the tool produces and how it verifies correctness.
- •Tool generates theme CSS (effectively a design system layer) and UI components
- •Validation includes type checking, build checks, and broken asset detection
- •PMs should build some technical intuition—even if they aren’t writing code daily
- •Discussion of Tailwind tradeoffs vs custom design systems for fidelity and flexibility
- •The output is positioned as “standard web app code” developers can recognize
Visual fidelity improvements: eyeballing colors, images, and layout via visual editing
Nadav fixes mismatched colors by using an image reference and an eyedropper-like workflow. This segment highlights the speed of direct manipulation (visual editing) to dial in UI accuracy without re-prompting the AI repeatedly.
- •Add the reference screenshot into the canvas for side-by-side comparison
- •Use visual tools to quickly correct colors and swap assets
- •Edits are reversible and don’t “break the page” the way people often fear
- •Visual editing is framed as a Wix-inspired strength: fast, concrete, low-latency iteration
- •Designers can help perfect org templates when fidelity matters
Add the first feature: sentiment analysis on posts (fast functional enhancement)
They prompt Dazzle to add a lightweight sentiment analysis feature to LinkedIn posts. The broader point is that PMs can now prototype “small but meaningful” features without waiting on engineering bandwidth.
- •Feature idea: show sentiment (positive/negative) on comments per post
- •AI prototyping unlocks many small tasks that previously wouldn’t justify dev investment
- •PMs can build prototypes even before an engineering team exists, enabling faster onboarding and delivery
- •Dazzle prototypes can be extended toward production, but prototyping remains the primary use case
- •Previewing and interacting with the prototype is central to evaluating the feature
Divergent solutions: rapidly exploring multiple UX implementations
Aakash emphasizes that the real leverage is generating multiple distinct solutions, not perfecting the first. They create a second variation: a separate sentiment summary module with an overview and drill-down concept.
- •Ask AI to create an alternate UI pattern (not just minor tweaks)
- •Second concept: sentiment analysis as a separate section with summaries/graphs
- •Use branching/toggles or separate sections to compare options quickly
- •Iterate beyond 2 solutions—3–4+ variations to discover better designs
- •After divergence, pick one direction and then polish it
Prompting mastery: clarity beats prompt-engineering theatrics
Nadav argues PMs don’t need elaborate system prompts; they need unambiguous intent. He explains misinterpretation risk, recommends “discuss mode” for major changes, and suggests using an LLM to detect contradictions before running prompts.
- •Most AI mistakes stem from miscommunication, not lack of technical detail
- •AI won’t push back when requirements don’t make sense (unlike a developer)
- •Use plan/discuss mode for big changes to confirm shared understanding
- •Large monolithic prompts can degrade results due to context switching
- •Pre-flight prompts: ask an LLM what’s unclear/contradictory to reduce costly failures
Editing workflow: visual edits vs prompting vs code (and what Cursor gets wrong)
They compare three editing modes: direct visual manipulation, natural-language prompting, and code edits. Nadav stresses immediacy: for many changes, visual edits are faster and more deterministic than prompting, and he contrasts this with Cursor’s “visual prompting” loop.
- •Dazzle exposes component vs instance vs underlying HTML element (div) distinctions
- •Inspect bound data and debug data flow through components when needed
- •Direct visual edits persist into the codebase (not transient like browser devtools)
- •Cursor’s visual editor often routes changes back through an LLM, losing immediacy and precision
- •PMs generally shouldn’t edit code unless something breaks or requires deep debugging
Building a multi-page, high-fidelity prototype to reduce usability and alignment risk
They extend the prototype into a dynamic post detail page with sentiment explanations and comment breakdowns. Then they discuss when high-fidelity prototypes are worth it: selling internally, fundraising, and conducting realistic usability tests.
- •Add navigation: click a post card → open a dynamic detail page
- •High fidelity helps stakeholders focus on value, not roughness or ‘ugly UI’ distractions
- •High fidelity is a selling tool (internal buy-in, leadership persuasion, even investors)
- •Usability testing is much stronger with realistic prototypes; avoid ‘Frankenstein’ mixed-fidelity flows
- •Best practice: test with real users (especially power users) via quick calls + published prototype links
Engineer handoff: publish link, share code, and keep specs inside the project
Nadav explains that a working prototype answers most engineering questions by demonstrating behavior. Beyond sharing links, teams can export standard code, use adjacent-code copying with developer AI tools, and embed spec context (PRD snippets) directly into project files.
- •A prototype provides behavioral clarity—often more valuable than raw code alone
- •Developers can download the project and use AI tools to replicate interactions in production code
- •Template + change history becomes powerful context for developer-side agents
- •Git integration (coming) simplifies collaboration and handoff workflows
- •Store spec/context in files inside the project so AI (and devs) can reliably reference it later
PRD + prototype becomes the new standard: 90% flows in prototype, edge cases in PRD
They argue PRDs aren’t dead, but their role changes: the prototype carries the main experience, while the PRD captures edge cases and completeness. Done well, PRD plus prototype should leave engineering with zero unanswered questions.
- •PRDs are often skimmed and can’t feasibly capture every UI nuance in text
- •Prototypes communicate the ‘picture’ efficiently and reduce ambiguity
- •PRDs remain essential for edge cases, constraints, and scenarios not represented in the prototype
- •Text is cheaper to maintain than code—use PRD for the long tail of requirements
- •Quality bar: after reading PRD + using prototype, developers should have no follow-up questions
The future: PMs must level up technical understanding as lines blur
Nadav forecasts that AI will blur the line between developers and tech-savvy builders, more than fully replacing engineers. He advises PMs to build technical fluency by interrogating code with AI and gradually contributing small changes, improving collaboration and speed.
- •AI replaces some simple dev tasks, but developers become code quality gatekeepers
- •PMs/designers may increasingly contribute small code changes via AI-assisted workflows
- •Organizations may need cultural/political adjustment to accept non-dev code contributions
- •Train the muscle: ask AI to explain your company’s architecture, diagrams, and system behavior
- •More technical shared language improves PM–engineering communication and execution speed
