I love experimenting with new AI tools, especially ones experiencing explosive growth. When Emergent hit $50M ARR in just 7 months, I had to understand what was driving that momentum.
So I spent some time building the same self-destructing journal app four different ways on their platform. Think Mission Impossible meets therapy session: write your thoughts, watch them burn, never look back.
The goal? Understand how different prompting strategies affect what AI generates. Does a simple founder story produce different results than detailed designer specs? What happens when you give AI full creative freedom?
What I found was fascinating, and something worth considering as AI vibe-coding scales.
P.S. Emergent is currently offering 75% off monthly plans. If you want to experiment with AI vibe-coding, it’s genuinely fast and fun to build with. Worth checking out: https://app.emergent.sh/?via=rutu
The Experiment: Four Attempts, Four Strategies
Attempt 1: Founder Mode I told a story. “I want to build an app that recreates the therapeutic ritual of journaling and shredding. Write freely, then destroy the pages so you’re not tempted to re-read and overthink.”
Emergent asked 4 clarifying questions: authentication method, visual theme, animation style, additional features. I picked: no auth, calming/minimal, paper shredding, keep it simple.
Time: 8 minutes
Credits: 5.71
Result: An app called “Shred It” with warm beige tones, Playfair Display headings, Lora body text, and a paper shredding animation.
Attempt 2: Designer Mode Full specifications. User stories, accessibility requirements (WCAG AA contrast: 4.5:1), edge cases, tone guidelines (“peaceful and calming like meditation apps”), and mobile-first responsive design. The works.
Emergent asked 4 questions: color theme preference, animation style, authentication, and fonts. I said “you decide” for everything.
It chose: Sage green + cream palette, Cormorant Garamond + Manrope fonts, dissolving particles animation.
Time: 10 minutes
Credits: 7
Result: “Fleeting Notes” with onboarding flow, streak counter, and the same aesthetic DNA.
Here’s where it gets interesting: The AI noticed the textarea was too narrow. It said “the writing area should be wider” and attempted to fix it twice. The final product? Still had a narrow textarea.
Attempt 3: Creative Freedom Same story as Attempt 1, but I added: “For design decisions, use your best judgment. Just build it.”
Emergent asked 3 questions. I said “rest you decide” after picking no accounts.
Time: 15 minutes
Credits: 10.21
Result: “FleetNote” with landing page, structured navigation, circular countdown timers, and the same sage green + cream palette. Cormorant Garamond serif fonts again. The “Paper & Peace” theme.
Attempt 4: The Corrected Concept By now I’d realized I’d been building the wrong thing. The first three attempts had a 15-minute waiting period after submission, creating the exact re-reading temptation the app was meant to prevent.
I corrected the concept: Write for 7 minutes (timer counts down while writing), then immediate 10-second destruction countdown.
Fresh account. Claude Opus model. Ultra thinking mode. I said “you decide” for all design choices and “add features based on what you think will make this app the best.”
Emergent asked 4 questions about color theme, animation style, writing experience, and additional features. I gave it full creative freedom.
Initial result: “Release & Let Go” with sage green palette, Cormorant Garamond fonts, and a breathing exercise I never asked for.
Then I iterated three times:
- “Can you update/change the design/color theme?” → AI recommended Midnight Zen (dark charcoal + lavender) and explained why: “Dark themes feel more private. Lavender = relaxation. The destruction animation will pop beautifully.” (+6 min, ~8 credits)
- “Change the icon and remove em-dashes” → Switched to flame icon (+6 min, ~8 credits)
- “Are you going to do a burn animation?” → AI implemented burn effect with embers
Time: 29 minutes total.
Credits: 34.07 (6x Attempt 1).
Result: Sophisticated dark theme with breathing exercises, but the burn animation felt rushed rather than cathartic.
Wait. They All Look… The Same?
Different accounts. Different prompts. Different strategies.
Same sage green palette. Same serif typography (Cormorant Garamond). Same feather icons. Same minimal card-based layouts.
Even when I explicitly said “you decide,” Emergent returned to the same aesthetic.
This isn’t a bug. It’s how AI works.
Plot Twist: I Tried the Same Prompt in Claude
Just to see if this was an Emergent thing or an AI thing.
Result: Blue-to-purple gradient buttons. Modern sans-serif. Cool tones.
If you have worked with Claude you KNOW how much it loves the blue to purple gradients!
Completely different vibe. But equally predictable.
The insight: Every AI tool has learned design preferences.
Emergent’s training data says “therapeutic journaling = sage green + serif fonts + nature icons.”
Claude’s training data says “modern web app = blue-purple gradients + sans-serif.”
Both are executing patterns they’ve seen thousands of times. Neither asked what I wanted.
Why This Matters (And Why Emergent Should Care)
Mukund Jha (Founder and CEO of Emergent) recently wrote about why Emergent isn’t a “GPT wrapper.” They built their own coding agent, container tech, memory system. Infrastructure depth for production-grade reliability.
He’s right. That matters.
But there’s another layer of production-readiness that matters at scale: design systems that differentiate.
Here’s the challenge:
When 5 million apps are built on the same platform, they start to look like they’re from the same platform.
Remember “the Bootstrap look”? Or how you can spot a Squarespace site? That’s what’s coming for AI-generated apps.
Speed and consistency are valuable. But they come at the cost of differentiation.
The Three Things I Learned
1. Simple Prompts Often Work Better
Attempt 1 (founder story, 8 minutes) felt more usable than Attempt 3 (detailed specs, 15 minutes).
Why? Less overthinking. The AI filled in sensible defaults. Simple is often better than complex.
2. Iteration Costs Add Up Fast
Attempt 4 started at ~15-18 credits. Each “small change” (color theme, icon update, burn animation) added 6 minutes and ~7-8 credits.
By the end: 34 credits, 29 minutes, and I still wouldn’t ship it. The burn animation felt rushed, not cathartic. The whole point of the app is emotional release, and the animation didn’t deliver that.
Here’s the thing: a quick Google search for “CSS burn animation” surfaced dozens of examples, some incredibly realistic with char-by-char burning effects and ember particles. Designers know to look for existing solutions first, then customize. The AI generated something from scratch that looked technically correct but felt emotionally flat.
This is the gap: AI builds fast, but doesn’t know what already exists or how to evaluate “does this feel right?”
One more iteration to fix it? That’s 40+ credits. At some point, “just one more change” becomes a sunk cost trap.
3. I’d Been Building the Wrong Thing
Midway through, I realized my original concept was flawed.
I’d asked for: “Entry sits for 15 minutes, then deletes.” What users actually need: “Write for 7 minutes, then immediate destruction.”
The difference? A 15-minute waiting period creates the exact re-reading temptation the app is meant to prevent.
All three initial attempts faithfully built what I specified. None questioned whether it was the right product.
This is the design leadership gap: knowing WHAT to build, not just HOW to build it.
The Real Bottleneck Isn’t AI Capability
Emergent delivers on speed. All four versions shipped fast and looked polished.
The bottleneck is product clarity.
AI can’t tell you:
- Is this solving the right problem?
- Does this feel right emotionally?
- Will this stand out in a crowded market?
- Should we build this feature at all?
Those are human judgment calls.
What Emergent Could Do (My Unsolicited Design Take)
As Emergent’s agents learn from every app built on the platform, there’s a risk that compounding intelligence becomes compounding homogeneity.
Every therapeutic journaling app teaches the agent: sage green works. Every productivity app teaches the agent: blue gradients work.
The patterns reinforce themselves.
Here’s the opportunity: Build design intelligence that compounds differentiation, not just consistency.
What if Emergent asked:
- “Your app is in the wellness category. 847 other apps use sage green. Want to stand out with a different palette?”
- “This interaction pattern appears in 34% of apps on our platform. Want to try something unexpected?”
Not forcing uniqueness. Just surfacing the sameness so builders can choose.
My Honest Take
Would I use Emergent for client work? Absolutely, for rapid prototyping and investor demos.
Would I ship an Emergent app to production without designer oversight? Not yet.
The gap isn’t technical capability. It’s the difference between “works correctly” and “feels right.”
At hypergrowth, that gap compounds into millions of user interactions that either delight or disappoint.
Finally Getting My Reps In
One unexpected perk of AI vibe-coding:
I finally used those weights sitting next to my desk. Prompted, did a set of curls while Emergent built, checked the preview, repeat.
Shipped an app and got a workout. Very efficient.
Emergent made building fun. But it also revealed something important about what happens when speed scales without design systems to match.
If you’re scaling a product at AI velocity and thinking about this challenge, let’s talk.
Book some time with me: