WIP
© 2026 Akshar Patel. All rights reserved.
Jump to Slide
RDSW 2026

Prototype Like a Pro:
Real-Time Vibe Coding

Came for the Vibes, Stayed for the Code

Akshar Patel (AP)
Product Manager, Airtable

Akshar Patel (AP)

Product Manager at Airtable
Enterprise AI features

No coding background
Started using AI coding tools in early 2025

40+ pull requests
27 features and fixes shipped to production

Prototypes → production code
Built and shipped using plain English

Who's Here?

PM or product role

A founder building something

Tried an AI coding tool
ChatGPT, Claude, Cursor

Built a working prototype
and showed it to someone

Before we start:

1. Open claude.ai
2. Create a free account
3. That’s it.

Free tier is all you need. Takes two minutes.

By the end of this session, everyone will have built something.

What if you could show people the thing
instead of describing it?

In the next 30 minutes, you’ll build a working prototype
using plain English. No coding experience needed.

85%
of devs use AI tools
46%
of code is AI-generated
#1
Collins Word of the Year 2025
Step 1 of 7

2 minutes

Define Your Idea

I want to build a type of tool that helps who do what.

Need inspiration? Pick one and make it yours.

Personal Dashboard
Habits, goals, daily metrics
Decision Helper
Weighted pros and cons
Pitch Timer
Timed presentation sections
Meeting Cost Calculator
Real-time meeting cost
Step 2 of 7

3 minutes

Write Your First Prompt

Too vague:

“Make me a dashboard.”

For who? What data? What actions?
AI will guess everything wrong.

“Build me a landing page. Make it look good.”

“Good” means nothing without context.
You’ll get something generic.

Specific:

“Build me a personal dashboard
that helps a founder track daily priorities.
Task list, pomodoro timer, notes section.
Dark mode, clean design.”

What it is, who it’s for, what problem it solves.
Features listed. Design direction given.

Say what it should do, not how to build it.

Ask for one thing at a time. Not five.

Step 3 of 7

2 minutes

Look at What You Got

Don’t accept it. React to it. Talk to it like a colleague, not a search engine.

Vague feedback:

“Make it better”
AI doesn’t know what “better” means to you

“Fix the layout”
Which part? What’s wrong with it?

Specific feedback:

“The buttons are too small to tap”
Describe what you see and why it’s a problem

“The header feels too heavy for the page”
Talk about the experience, not the code

Don’t like it? Say “undo that” or “go back to the previous version.” Nothing is permanent.

Step 4 of 7

3 minutes

Send One Piece of Feedback

Pick the one thing that bothers you most. Fix that first.

Layout

“Make it two columns”

Change how things are arranged

Feature

“Add a dark mode toggle”

Add something that wasn’t there

Bug

“The timer doesn’t reset”

Describe the behavior, not the fix

This is the loop.
Describe → Test → React → Repeat.

Talk to it like a colleague.
Not a search engine. Not a command line.

Step 5 of 7

5 minutes

Keep Going

Try:

“Make it more polished”
“Add smooth animations”
“Add a feature where...”
“Make it work on mobile”

See a red error? Copy it.
Paste it back with no comment.
That’s the fix 80% of the time.

Stuck?

Gone back and forth 3 times?

Start a new conversation.

Pro tip: Keep a planning doc
outside the chat. Paste it into
fresh sessions for clean context.

Raise your hand if you get stuck.

Step 6 of 7

5 minutes

Add one feature
you didn’t plan for.

Something that would make someone else want to use this.

“Add a share button”

“Add confetti on completion”

“Save my past sessions”

Power move: give it a reference.
“Style it like Stripe’s pricing page” beats describing every detail.

Step 7 of 7

3 minutes

Show Your Neighbor
What You Built

Everyone in this room just built working software.
Most of you don't write code for a living.

Turn to your neighbor. Show them what you built.
If you want to share with the room, raise your hand.

Reflection

The Process You Just Used

Before you build

Know the problem. Not the solution. What are you trying to make easier?

Find a reference. “Make it work like X” beats describing from scratch every time.

The build loop

Describe what you want in plain English.
Test it. Click around. Break it.
React with short, specific feedback.
Repeat until it’s good enough to show someone.

Your PM skills are your edge.
Describing problems clearly, giving specific feedback,
knowing what good looks like — that’s the whole job.
Context rots. Your notes don’t. Keep a planning doc outside the chat.

Honest Assessment

Where This Works. Where It Doesn't.

Great For

Prototyping & MVPs — idea to testable in hours
Internal tools — dashboards, calculators, admin panels
Stakeholder communication — show, don’t tell
Learning — great way to understand tech

Falls Apart For

Security-critical code — real vulnerabilities
Complex backend — databases, permissions, auth
Production at scale — performance, edge cases
Anything you can’t verify — don’t ship it

Prototype freely. Ship carefully.

Your Playbook

Three Rules, Three Horizons

Start With What You Know

Your domain expertise is the input. AI is the builder.

Iterate, Don't Accept

First output is a draft. Push it until it's right.

Choose Your Fidelity

Napkin sketch or production code. You decide how far to go.

This Week

Iterate on what you built today

This Month

Prototype a real idea at work

This Quarter

Show it to stakeholders

Questions?

Akshar Patel (AP)

Product Manager, Airtable

LinkedIn: linkedin.com/in/apatel13

Thank you for building with me today.