The surprising patterns behind viral AI products
A deep dive into Bolt, Cursor, Granola, PhotoRoom, Replit and more
👋 Hi, it’s Kyle Poyar and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.
I’ve had a blast testing out new AI products for note taking (Granola), photo editing (PhotoRoom, Aftershoot) and even app building (Replit, Bolt). What’s been fascinating is seeing new 🔥 UX patterns emerge that I haven’t seen before. Turns out I’m not alone: these products are reaching millions of users and tens of millions in ARR in a matter of weeks. Repeat contributor Yaakov Carno — founder of Valubyl and author of the Product Led Growers newsletter — unpacks the subtle UX choices that are leading to breakout growth.
PS: I just teamed up with Maven to curate interactive, cohort-based growth courses from my favorite experts — most of whom you might notice from this newsletter. Explore the full Growth Unhinged x Maven collection here and get $100 off with the code KYLExMAVEN.
AI products are shattering growth records, with some hitting $100M ARR faster than any software in history.
Just look at Bolt—within two months, the product reportedly skyrocketed to $20M ARR with over 2 million users. Its secret? Not just powerful AI, but a ridiculously smooth, intuitive UX that makes complex development workflows feel effortless.
Bolt isn’t an isolated instance; I’ve noticed similar hockey-stick growth at companies like Cursor, Replit, Lovable and PhotoRoom.
Meanwhile, many AI products end up as a flash in the pan. They attract AI ‘tourists’—users who sign up, get confused, and churn almost instantly. Why? Because they can’t figure out how to use it, don’t trust its decisions, or feel like they’re wrestling with AI rather than collaborating with it.
The difference between AI-native products that explode in growth and those that fade away often comes down to one thing: user experience.
Unlike traditional software, AI-driven products are dynamic and unpredictable. They generate unique outputs, adapt to user input, and, when poorly designed, can feel frustrating, mysterious, or outright unreliable.
The best AI-native products don’t just deliver powerful automation. They guide users through a seamless, intuitive, and trustworthy experience—one where AI feels like an assistant rather than a guessing game.
I’ll break down five key UX challenges in AI-native products and how the best companies are solving them.
Challenge 1: AI feels like a black box
🧠 "AI feels like magic, but that makes it hard to trust."
Users hesitate to rely on AI when they don’t understand how it works. If an AI system produces results without explanation, people second-guess the accuracy. This is especially problematic in industries where transparency matters—think finance, healthcare, or developer automation.
How companies are solving this
Bolt breaks down the AI process step-by-step in real time, so users see exactly how their code is being generated or automated.
Cursor doesn’t just “fix” your code—it explains why it made each suggestion, reinforcing trust.
PhotoRoom adds explanations behind AI edits, helping users understand the “why” behind its decisions.
Pro-tips
✅ Show step-by-step visibility into AI processes.
✅ Let users ask, “Why did AI do that?”
✅ Use visual explanations to build trust.
Challenge 2: AI is only as good as the input — but most users don’t know what to say
📝 "Users don’t need better AI—they need better ways to talk to it."
AI is only as effective as the prompts it receives. The problem? Most users aren’t prompt engineers—they struggle to phrase requests in a way that gets useful results. Bad input = bad output = frustration.
How companies are solving this
Bolt & Replit make it easy for users to refine prompts with one-click enhancements, helping users improve their input before execution and get better results.
PhotoRoom introduces three AI editing paths:
Assisted mode: Guides users step-by-step through structured editing.
Image mode: Suggests similar images to spark inspiration.
Manual mode: Gives advanced users full control over edits.
Pro-tips
✅ Offer pre-built templates to guide users.
✅ Provide multiple interaction modes (guided, manual, hybrid).
✅ Let AI suggest better inputs before executing an action.
Challenge 3: AI can feel passive and one-dimensional
💡 "A great AI assistant should work with you, not just for you."
Many AI tools feel transactional—you give an input, it spits out an answer. No sense of collaboration or iteration. The best AI experiences feel interactive.
How companies are solving this
Replit uses a dual-mode AI assistant: agent mode (automates full builds) and assistant mode (helps with smaller refinements).
Cursor combines AI chat with execution, allowing users to switch between exploratory conversation and direct AI-powered coding assistance.
Fathom’s Ask Fathom feature turns AI meeting summaries into an interactive experience, letting users engage with transcript results instead of just receiving static output.
Pro-tips
✅ Design AI tools to be interactive, not just output-driven.
✅ Provide different modes for different types of collaboration.
✅ Let users refine and iterate on AI results easily.
Challenge 4: Users need to see what will happen before they can commit
🤔 "People don’t trust what they can’t test."
Users hesitate to use AI features if they can’t predict the outcome. The fear of irreversible actions makes them cautious, slowing adoption.
How companies are solving this
Bolt uses predefined AI prompts that allow users to test it before committing — even before signing up.
Replit adds confirmation & rollback checkpoints so users can preview AI-generated code before executing it, reducing risk and fear.
Fathom provides interactive onboarding, where users can test AI insights in a sandbox environment (a two minute test call) before having the recording bot join them in real meetings.
Pro-tips
✅ Allow users to test AI features before full commitment.
✅ Provide preview or undo options before executing AI changes.
✅ Offer exploratory onboarding experiences to build trust.
Challenge 5: AI can feel disruptive
⚙️ "AI should weave into the workflow, not interrupt it."
Poorly implemented AI feels like an extra step rather than an enhancement. AI should reduce friction, not create it.
How companies are solving this
Cursor lets users accept/reject AI suggestions instantly for a seamless workflow.
Granola seamlessly blends your rough notes with a comprehensive, contextual summary, allowing you to capture loose thoughts effortlessly - without ever losing focus on the call.
Grammarly adapts to each scenario, offering context-aware edits and replies - eliminating the need for repetitive “Write me a reply that…” prompts.
Bolt allows users to seamlessly switch between AI-generated code and a live preview.
Pro-tips
✅ Provide simple accept/reject mechanisms for AI suggestions.
✅ Design seamless transitions between AI interactions.
✅ Prioritize the user’s context to avoid workflow disruptions.
Designing AI that works for people
AI isn’t the differentiator anymore—great UX is. If you want your AI product to succeed, make sure it’s clear, trustworthy, and seamless—or watch users disappear.
The patterns we’ve explored—transparency, guided input, interactivity, predictability, and seamless integration—are key to creating AI experiences that drive real adoption and retention.
 | A guest post by
|
Epic to collab with you again Kyle, thank you! 🤓
10/10 subject line (will read the rest soon)