Welcome back!

This week we break down why niche-focused products outperform broad platforms, walk through the path an ugly MVP took to reach 100,000 users, explore how an AI tools hit a $250M valuation by killing meeting bots, and show how another AI escaped a 3-year plateau to reach $10M ARR.

Have suggestions? Send them to us at [email protected].

Founders Intel

Is your product focused enough to win?

Data Intel

• Vertical SaaS companies report 35-60% higher customer retention than horizontal platforms.

• The vertical SaaS market hit $106 billion in 2024 and is growing at 16.3% CAGR toward $369 billion by 2033.

• Micro-SaaS startups are growing at ~25% annually, outpacing traditional SaaS.

• Niche-focused products see faster activation, lower churn, and higher lifetime value.

Why It Matters

Building broad feels safe. Serve everyone, capture more market. But “everyone” means competing with giants who have deeper pockets and longer runways.

The founders winning right now are doing the opposite: picking a specific audience, solving their exact problem, and owning that slice completely. Toast owns restaurants. Procore owns construction. SimplePractice owns therapists. These aren’t small markets; they’re focused ones.

When your product speaks directly to one group’s workflow, onboarding friction drops, support costs shrink, and users stay longer because switching means losing something built for them.

Quick Tip

Look at your best customers. What industry, role, or use case do they share? Rewrite your homepage and content to speak only to that group for 30 days and measure signup quality. If conversions improve, you’ve found your niche. If not, test the next segment.

Behind the Tool

The Ugly MVP That Turned Into 100,000 Users

The Spark

Nicolai spent 12 years as a hobby musician, rapping in German and running his own band. But his day job was physics, where he eventually completed a PhD. Throughout that time, one problem kept frustrating him: every song needed visual content, and creating music videos without a budget or production skills was nearly impossible. Back then, he tried animating stick figures himself.

When Stable Diffusion dropped in late 2022, Nicolai started experimenting. He hacked together an AI animation tool, initially without a clear use case. But as users trickled in, a pattern emerged. Musicians kept showing up. They wanted to visualize their tracks. So he narrowed the focus entirely: Neural Frames became an AI music video generator built from musicians, for musicians.

The Build

For the first year, Nicolai worked completely solo, living the indie hacker digital nomad life while grinding on the product. He wasn’t a software developer. The first version of Neural Frames looked, in his words, “really, really bad.” He didn’t even know what a landing page was.

But he kept shipping. He tried Reddit with little traction. Twitter didn’t work either since he had no audience. Revenue was minimal, but a handful of users kept writing in. Some complained the site didn’t work on mobile. Others said things like, “I just visualized my dream.” That feedback kept him going.

The bootstrapped approach forced discipline. No funding meant no runway to waste on broad positioning. Nicolai leaned into the niche. While competitors chased the general AI video market, Neural Frames stayed focused on one problem: helping musicians create professional music videos without big budgets or production teams.

The Breakthrough

On a random Sunday, Nicolai posted Neural Frames to Hacker News. Then he went to dinner with friends at a pizzeria. Midway through the meal, he checked his phone. Something was happening.

The post went viral. And the reason surprised him. The site’s rough, unpolished look actually resonated with the HN crowd. They could see someone had hacked something together out of passion, not corporate calculation. It felt real.

From there, the momentum built. TAAFT discovered Neural Frames and listed the tool. For months, it became one of the biggest traffic sources, sending a steady stream of musicians who were actively searching for exactly what Nicolai had built. The combination of viral exposure and targeted discovery turned attention into revenue.

As income grew, Nicolai started hiring. The team expanded, monthly active users crossed 100,000, and brand searches for Neural Frames started climbing in SEO tools.

The Next Chapter

Neural Frames now produces 4K music videos in minutes using models like Kling and Seedance. The platform extracts lyrics, maintains character consistency across scenes, and helps musicians tell visual stories that match their sound.

Nicolai is focused on staying ahead of the rapid improvements in AI video. The quality gap between last year and today is staggering, and he expects the next five years to reshape how all visual content gets made. For now, the goal is simple: keep building the best product for musicians and let the niche compound.

Key Lesson

Polish can wait. Authenticity scales. Nicolai’s ugly first version worked because it signaled realness to the right audience. Combine that with relentless niche focus, and a bootstrapped founder can outmaneuver larger players who spread themselves too thin.

Tool of the Week

Granola’s No-Bot Playbook

What’s Granola?

An AI notepad that transcribes meetings from your device and enhances your typed notes. No bot joins the call. No awkward “Granola is recording” announcements. Hit $250M valuation with $67M total funding and 10% weekly user growth since launch.

What Worked

• Killed the bot, kept the value: Every competitor (Otter, Fireflies, Read AI) sends a bot into meetings. Granola refused. The app runs locally, captures audio from your device, and transcribes without anyone knowing. This unlocked access to sensitive conversations (board meetings, investor calls, executive recruiting) that’d never allow a visible bot.

• Cut 50% of features before launch: Spent a year in stealth onboarding users one by one. Tested dozens of prototypes. When they finally shipped, they’d stripped the product to a single use case: open the app, take notes, let AI enhance them. The simplicity drove 70% weekly return rates.

• Targeted VCs first, then expanded: Started with investors who take 15+ meetings weekly and need perfect recall. VCs told founders. Founders told other founders. The product spread through the exact network that’d later fund them. Lightspeed, Spark Capital, plus angels from Vercel, Replit, and Shopify all invested.

• Built sharing as a growth loop: When you share Granola notes, recipients get an interactive link where they can ask the AI questions about the meeting. Non-users experience the product’s value without signing up. Every shared note becomes a product demo.

Founder Quote

“There’s no bot joining your meeting or special UI to open. It’s a notepad, like Apple Notes. You can open it when you want and close it when you don’t. It sounds simple, but it took a lot of work. We had to build many features and cut them out to determine what was truly core.” — Chris Pedregal, Granola CEO

Key Lesson

In a crowded market, find the constraint everyone accepts (meeting bots) and eliminate it entirely. The segment that can’t use competitors becomes your wedge.

Fresh Out of the Lab

Z-Image

What Is It?

An open-source 6B parameter image generator from Alibaba Tongyi Lab that runs on consumer GPUs (16GB VRAM). It produces photorealistic images in 8 steps with sub-second inference on high-end hardware.

What’s New

Z-Image uses a Scalable Single-Stream DiT (S3-DiT) architecture that processes text and images together. It supports bilingual text rendering (English and Chinese), excels at typography in images, and includes three model variants: Turbo for speed, Base for fine-tuning, and Edit for natural language image modifications.

Why It Matters

Founders building creative tools now own the full image generation stack without API costs. Z-Image runs locally on RTX 3060-tier hardware, generates images extremely fast, and ships under Apache 2.0 for commercial use. No cloud dependencies. No per-image fees. Full control over the pipeline.

Founder’s Edge

This Week’s Builder Toolkit

• Dev Tool: Dub.co is an open-source link management platform built for marketing teams that can create branded short links, generate QR codes, and track analytics.

• Free Dataset: Explore the 2025 MAD Landscape from FirstMark, an interactive market map of 1,150+ ML/AI/Data companies. Search by category, view data cards, and spot whitespace in the ecosystem.

• No-Code App: Glide turns spreadsheets into polished apps in minutes. Connect Google Sheets or Excel, drag and drop components, and deploy internal tools or client portals with built-in AI features.

• Productivity Hack: Edit videos and podcasts by editing text with Descript. The AI removes filler words, enhances audio quality, and generates captions automatically..

• Learning Resource: The LLM Course by mlabonne (69k GitHub stars) walks you through building and deploying LLMs. It covers fundamentals, fine-tuning, quantization, and RAG with hands-on Collab notebooks.

Note: If you’ve found a tool that’s sped up your build process, hit reply and share it. We’ll feature the best submissions in a future issue.

AI Founder’s Journal

How Jenni AI Escaped a 3-Year Plateau

For three years, Jenni AI was stuck. The product targeted SEO writers who needed AI to churn out articles faster. Despite using the same GPT-3 technology as competitors pulling in millions, David Park watched larger, better-funded rivals dominate while his growth flatlined.

The problem wasn’t the product. It was the customer.

SEO writers wanted AI to replace their work entirely. They’d generate content, publish it, and move on. No loyalty, no word-of-mouth, no reason to stay. The moment a cheaper tool appeared, they’d switch.

Park’s team had a difficult conversation. They decided to abandon the SEO market entirely and rebuild for academic researchers and students.

The psychology was different. Students didn’t want AI to write for them. They wanted AI to help them learn while writing. They needed proper citations, paraphrasing assistance, and plagiarism avoidance. These were features SEO writers never cared about, but students would pay for and tell their classmates about.

The pivot landed immediately. Word-of-mouth spread through universities. Within 15 months, Jenni AI went from $2,000 to $150,000 MRR. By early 2025, they hit $10M ARR with a 23-person team.

Park also noticed something odd in his analytics. Users kept searching for “Jenny AI” instead of “Jenni AI.” Rather than ignoring the typo traffic, he bought jenny.ai.

The decision felt risky. But word-of-mouth was their second-biggest growth channel, and misspellings were leaking potential customers. Over the next 10 months, that single domain captured 52,000 visitors, converted 20,000 new users, and generated 271 paid subscribers. At an $80 average lifetime value, the domain returned $21,680 in revenue, paying for itself and then some.

Here's why both moves worked:

The pivot succeeded because students had a fundamentally different relationship with the product. They used it repeatedly throughout their academic careers, recommended it to peers in study groups, and valued features that created stickiness. SEO writers had no such attachment.

The domain purchase worked because Park recognized that word-of-mouth has friction. When someone tells a friend about your product verbally, they might misspell it. Owning the misspelling removes that friction entirely.

Key Takeaway

When growth stalls, question your customer, not your product. Find users who need what you built, not users you wish would buy. And once word-of-mouth starts working, capture every path to your product, even the ones people get wrong.

— David Park, Jenni AI founder

Weekly Challenge

One Experiment. One Week. One Win.

The Goal

Reduce your release cycle by at least 50% for one feature or update.

How It Works

1. Pick ONE update you planned to ship next month.

2. Cut scope to the smallest version that delivers value.

3. Ship it within the next 7 days.

Why It Works

Speed exposes assumptions faster than planning. Ship small, measure results, then decide if it deserves more time.

Spotlight

We’ll showcase the fastest turnaround in an upcoming newsletter.

AI Market Watch

Deals, Discoveries, and Demand

Megadeals

• 7AI – $130M Series A (AI security agents)

• Eon – $300M Series D (cloud data backup)

• NVIDIA – $2B investment (engineering and design)

• Black Forest Labs – $300M Series B (image generation)

• Snowflake – $200M partnership (agentic AI for enterprise data)

New Research

• FINDER (deep research agent benchmark)

• DeepSeek-V3.2 (open model matching GPT-5)

• CodeVision (code-as-tool for image reasoning)

• Polarization by Design (AI persuasion dynamics)

• Is Vibe Coding Safe? (61% functional, 10.5% secure)

• Training LLMs for Honesty (OpenAI confession method)

• The Missing Layer of AGI (System-2 coordination for reasoning)

Search Trends

Top “best AI for…” searches:

1. Presentations

2. Research

3. Video editing

4. Image generation

5. Automation

Thanks for reading!

Got a project in motion? Hit reply and share what you’re working on with us.

We’ll be back next week with more founder stories, fresh tools, and data worth your time.

See you soon,
— AI Empires