Vibe Experimentation: An AI PM's Guide

You’ve heard about vibe coding, but have you heard about vibe experimentation?

There’s a lot of people talking about AI PM. But one revolution I have seen no one talk about is the impact of AI on experimentation.

That’s insane. This is one of AI’s biggest impacts to product management.

AI vibe coding tools have collided with experimentation to create an entirely new phenomenon: vibe experimentation.

One startup I work with went from 24 to 196 experiments in a quarter, smashing their OKRs, just by implementing a vibe experimentation program.

This is the edge that will separate tomorrow’s great PMs from everyone else.

---

A New Era

Most product teams have 100s of ideas for quick experiments, but they never get to them due to the development costs involved.

That’s why it’s so common to hear a team that just runs 2-3 experiments / quarter. It takes 8 weeks to do a single experiment.

This has led to the steady and slow death of quick growth experiments in core product teams. Those are all outsourced to the growth teams.

But it shouldn’t have to be this way.

In 2025, with the latest AI tools, all core product teams can vibe experiment a multitude of growth experiments every quarter.

It’s as simple as this:

---

The Unlock for Vibe Experimentation

Here's what makes vibe experimentation possible in 2025: AI tools can now connect directly to your existing codebase and design systems.

This isn't about building random prototypes that look nothing like your actual product. It's about creating experiments that inherit your brand, your design patterns, and your technical constraints from day one.

There are three stacks you can use to make this work:

Stack 1: Prototype + Code + Experiment

The first stack has a few more steps, but is also easiest for the PM/ experiment designer:

1. Prototype: You can use an AI prototyping tool like Lovable or Bolt + your brand guidelines imported as context.

2. Code: Then you’ll either go to a coding tool like Cursor to integrate to your code base or have your engineering integrate.

3. Experiment: Then you launch with your experimentation platform.

It still is vibe experimentation - you’re skipping a lot of the back and forth on the design - but it’s not immediate either.

Stack 2: Vibe Code + Experiment

This second stack removes the AI prototyping tool and jumps straight to a vibe coding platform:

1. Vibe Code: You use a Cursor, Claude Code, or Windsurf connected to your GitHub repo + Figma design tokens to build the variant using your actual components, colors, typography, and interaction patterns

2. Experiment: Then, you can use LaunchDarkly or Amplitude to release your feature to users and experiment with it

It’s best for teams with established design systems who need experiments to feel native to their product. And it’s fast.

Stack 3: All in One

This final stack simplifies everything. You head to an advanced experimentation platform.

For instance, Kameleoon’s prompt-based experimentation can do it all in one place: you can create variations of live pages using your design system, and then host the experiment to check for statistically significant results.

Other experimentation platforms will get there one day - but are currently behind. For example:

• Optimizely allows you to tweak copy or images, but you cannot build new flows

• Amplitude’s Agents one day promise to launch experiments, but right now it suggests them.

This is often the fastest method to really get into a vibe experimentation workflow.

You can choose whatever stack that works for your company.

The key? Make sure you’re achieving the goals of a vibe experimentation program: learning more, faster!

"We went from running a few experiments per quarter to launching ideas in real time."

—Marcella Gutierrez, Dir of Digital Experience at Fossil ($1B /yr rev watchmaker) that went from 10 experiments to 100+ per quarter

---

Here’s the playbook I wish every PM had in 2025: the exact stacks, 100 experiment ideas, the 5-step workflow, and how to get engineering bought in. It’s what separates the teams stuck at 2 experiments/quarter from those running 20+.

Inside, you’ll get the exact prompts, stacks, and review scripts teams use to go from idea → live experiment in under a week - without blowing up prod or brand trust.

That’s all in the deep dive 👇