How to use AI for Prototyping as a PM

Dec 03, 2024

In this post, I’m going to share some practical advice on:

Why PMs should be prototyping with AI - and in fact why this skill will become necessary for PMs
What are your options for tools (pros and cons of each)
How to tactically get started here

Why PMs should be prototyping with AI

There’s a ton of hype around AI coding agents right now - and for good reason. The barrier to entry for getting started with software is much lower than it was even a year ago. “AI is the new plastic”, whereas before software used to be more like “silverware” (harder to produce, took more time) - which means we need to treat prototypes the same way. On top of that, I could now use these “prototypes” to make lightweight changes to our existing product! What used to take weeks of back-and-forth with engineering could now be prototyped in hours, or minutes (sometimes, real time in the meeting).

I shared this on Lenny’s Podcast, how to use Replit for prototoypes or even building an app on my phone.

A week later, the CEO of Replit came on the podcast and built a prototype in real time!

We’re observing in real time a collapsing of the skill stack for AI PMs.

Claire Vo

framed the question as (paraphrasing): What if companies cared less about titles and more about skills?

A core component of this shift revolves around being able to communicate ideas and “prototype” with AI. This post focuses on code-based prototypes, but remember you can use code to communicate visual elements, user experiences, designs, mock-ups and ideas as well!

Replit isn’t the only agent out there - there are a number of tools constantly emerging (literally every week).

Understanding the Landscape

Having spent years navigating the AI product landscape, I've learned that choosing the right prototyping tools can make or break your development process.

The AI coding agent ecosystem broadly divides into four main categories, each with its own strengths and ideal use cases. Let's break them down based on my hands-on experience and how I would pick the right tool for the job:

Replit Agent: The Backend Powerhouse

I've personally found Replit Agent particularly powerful for backend-heavy applications.

Strengths:

Robust container and backend deployment capabilities (i.e. perfect for Twitter bots, scheduled jobs)
Excellent multi-modal capabilities, Replit Agent can use UI rendering for debugging errors (i.e. page not rendering correctly)
Comprehensive development environment - spin up a container and host the app directly from Replit

Challenges:

Requires frequent instruction clarification (copy pasting API docs and reiterating instructions)
Limited IDE experience compared to traditional tools
Docker limitations can be frustrating for complex deployments (Replit doesn’t use traditional Docker environments)

Vercel v0: Fast Frontend Prototypes

Vercel has an agent based framework called v0, which has emerged as a go-to tool for clean, modern UI development. I made my personal website (https://amank.ai/) using a combination of v0 and Cursor.

Strengths:

Produces exceptionally clean UI code using shadcn components
Seamless integration with modern web frameworks
Excellent for rapid frontend prototyping

Limitations:

Block-based workflow can be convenient but also a little confusing (essentially pulling 1 page at a time)
Iterations between v0 and IDE could be smoother (carrying over requirements)

Cursor: Familiar IDE Experience

Working with Cursor brings the familiarity of VS code (all your extensions and environment).

Strengths:

Powerful web and codebase reference capabilities using @ commands (@web, @codebase)
Intuitive interface for developers, familiar with VS Code
Strong integration with existing workflows and plugins

Limitations:

Still developing true "agentic" planning capabilities (taking user input to iterate)
Can sometimes miss context in complex projects

Windsurf: The New Agent on the Block

Windsurf is the newest entrant here. Windsurf claims to be an “Agentic IDE”, which takes human feedback in account when producing flows and generating + running code.

Strengths:

Strong planning agent capabilities, with human feedback
Strong terminal command execution
Cheaper than Cursor (for now) - $10/mo for unlimited queries

Limitations:

Limited web and codebase reference capabilities (not available yet)
Workflows are still a bit early relative to Cursor

Choosing the Right Tool

From my experience, the selection process comes down to 2 key factors:

Project Type:
- Frontend-heavy projects → Vercel v0
- Backend services → Replit Agent
- Full-stack applications → Combination of tools
Team Experience:
- Developer-heavy teams often prefer IDE based tools (i.e. Cursor, Windsurf)
- Designer collaboration works well with Vercel v0
- Teams that need a backend system and want to deploy quickly would use Replit

How to Get Started: Implementation Strategy

Based on my successes (and failures), here's my recommended approach:

Start Small: Begin with a simple prototype using one tool. I typically start with Vercel v0 for visual components or Replit Agent for apps that need backend services (i.e. databases). Have a project in mind (a twitter bot, or personal website)
Iterate Rapidly: Use the tool's strengths to quickly validate your concept. For instance, Vercel v0's clean UI generation can help validate design ideas with stakeholders faster.
Scale Thoughtfully: As your prototype grows, don't hesitate to combine tools. I often use Replit Agent for backend services while leveraging Vercel v0 for the frontend.

Example Prototypes

Here are a few extra ideas in case you’re trying to get started but need some prompts

Quick Wins (Afternoon Projects)

Restaurant picker for team lunches (great first Vercel v0 project!)
Meeting notes summarizer pulling from your calendar
Random ice breaker generator for standups

Internal Tools That Drive Impact

Simple product usage dashboard with Pendo/product analytics data
Customer feedback aggregator from multiple sources (Gong) and prioritizer
Customer health score dashboard

Customer-Facing Prototypes

Customer ROI calculator with visualizations (your sales team will love this one)
Pricing tier comparison tool
Product tour generator
Mobile site optimization

Looking Ahead

This space is evolving rapidly - I’m sure a week after I post this, it will already be out of date in some way. That being said, here are a few trends to keep an eye on when trying out AI prototyping tools:

Integration of multiple AI agents in development workflows (API based tools)
Improved context awareness and planning capabilities (human input guiding agent planning)
Better handling of complex, multi-service architectures

Remember, your goal isn't to find the perfect tool, but to leverage each tool's strengths to build better products faster. Your choices should align with your team's workflow and project requirements while maintaining the flexibility to adapt as both the technology and your needs evolve. Don’t get too stuck here - prototypes are meant to be built and shipped fast!

👇 If you liked this, consider subscribing, sharing and leaving any feedback!

Want to see more? I’ll be covering: