Welcome back!

This week we break down why distribution-first beats product-first for startup survival, walk through Lovable’s path from open-source project to $200M ARR in 12 months, explore an AI tool adding sound to silent videos, and show how open-source voice cloning is catching up to ElevenLabs.

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

Founders Intel

Distribution-First Beats Product-First

Data Intel

• 22% of startups fail due to marketing problems. Another 34% fail from lack of product-market fit, which often means they built something nobody found.

• CAC rose 55% in the last five years across B2B and B2C. Companies using freemium models saw only a 25% increase.

• 47% of first movers fail. Fast followers who learn from pioneers and focus on distribution win 6× more often.

• 78% of startups are self-funded. The ones that survive diversify their channels: owned (content, community), earned (word of mouth, press), and strategic placements where their audience already looks.

Why It Matters

The startup graveyard is full of great products nobody ever discovered. Building something people want is table stakes. Getting it in front of them is the game.

History proves this over and over. Facebook beat Friendster. Google beat Yahoo. Netflix beat Blockbuster’s streaming attempt. The winners weren’t first; they were better at distribution. While pioneers burned cash educating markets, fast followers found channels, built flywheels, and made discovery frictionless.

This is why product-led growth dominates. Freemium models, viral loops, and shareable outputs turn users into acquisition channels. The product itself becomes distribution.

But the best PLG companies don't stop there. They layer strategic placements and paid visibility on top of organic momentum, using discovery platforms and sponsorships to amplify what's already converting.

Quick Tip

Map your distribution channels this week. Owned (your list, your content), earned (word of mouth, press), and paid (ads, sponsorships, placements). The strongest growth engines combine all three. If you’re relying on a single channel, you’re one algorithm change or budget cut from stalling. Diversify and double down on what converts.

Behind the Tool

SciSummary's Path to 800K Users

The Spark

Max, a software engineer with 14 years of experience and a background in edtech, didn’t set out to disrupt academic research. His wife, a PhD chemist, asked him to build something useful: a tool to summarize the dense research papers piling up on her desk.

This was late 2022, before ChatGPT entered mainstream consciousness. Max built the first version of SciSummary and watched it grow organically from day one.

The Build

SciSummary started as a straightforward summarization tool but evolved into a full reference manager with AI-powered features.

The platform has evolved to handle multiple workflows: synthesizing papers into literature reviews, chatting with documents while getting citations, analyzing figures in context, and highlighting text for deeper exploration.

The technical edge came from specialization. While general-purpose models tried to do everything, Max focused SciSummary on research papers exclusively.

The platform combines LLMs with specialized models trained specifically on academic content, then A/B tests continuously (GPT-4, Claude, Gemini variants) to identify which performs best for each task.

The Breakthrough

SciSummary hit number three on TAAFT’s leaderboard in January 2023, just weeks after launch.

The platform used that early traction to compound growth by running paid ads through TAAFT while building organic reach through Google rankings. Growth exploded to 800,000 users, with seasonal patterns tied to academic calendars (summer slowdowns, September ramps).

The Next Chapter

The platform is building “lit review mode,” which internalizes the entire literature review workflow. Users upload documents, SciSummary identifies research gaps, suggests relevant papers, imports them automatically, and generates first drafts in 10-15 minutes.

This addresses the top use case: Interviews with 30-40 users per month confirmed that literature reviews drive adoption during dissertation and thesis seasons.

Looking forward, Max sees agentic tooling as the industry standard. SciSummary will integrate with MCP (Model Context Protocol) systems, living inside broader AI ecosystems rather than standalone websites.

Key Lessons

Find underserved niches using discovery platforms. Search TAAFT by workflow. If only a few tools exist but demand is clear, build there. If zero tools exist, skip it (too speculative).

Talk to users relentlessly. Every feature in the past year came from user interviews or competitor analysis. 10-15 video calls monthly, 30-40 total conversations. Trends emerge fast when you ask directly.

Specialize beats general-purpose. ChatGPT summarizes papers, but SciSummary does it better because it only does research papers. Niche focus wins on quality.

Launch early on distribution channels. Being early to TAAFT (January 2023) created compounding advantages: high rankings, ROI, and backlink authority for Google.

Instagram and Google Ads work for academic tools. Reddit hates AI (unless you're in specific subreddits). LinkedIn is expensive. Facebook/Instagram show traction but need more testing.

Speed kills drop-off. Users abandon slow tools. Optimize for response time over reasoning depth when the task is straightforward.

Tool of the Week

Mirelo’s Silent Film Solution

What’s Mirelo?

An AI platform that generates realistic sound effects for silent videos. Upload a clip, and Mirelo adds footsteps, doors, ambient sounds, and other effects that match what’s on screen. Raised $41M seed round in December 2025.

What Worked

• Targeted a gap nobody was solving: Video production teams spend hours manually syncing sound libraries to footage. Mirelo automates the entire workflow. The model detects on-screen actions, matches them to sound categories, and generates synchronized audio.

• Built for professionals first: Instead of chasing viral consumer use cases, Mirelo focused on film studios, game developers, and content agencies who pay for time savings. Enterprise contracts locked in before public launch.

• Compression as competitive moat: The model runs 10× faster than comparable approaches by using a compressed audio representation. Speed matters when clients process thousands of clips per project.

• Secured compute partnership: The $41M seed includes strategic backing from AI infrastructure players. Access to compute at scale lets Mirelo serve high-volume enterprise clients without cost spiraling.

Founder Quote

“It’s easier to build a real moat here, and then to capitalize on it.” — CJ Simon-Gabriel, Mirelo co-founder and CEO

Key Lesson

Pick the category giants ignore. Sound was the last frontier of generative AI while everyone chased images and text. Less competition, more defensibility.

Fresh Out of the Lab

Chatterbox TTS

What Is It?

An open-source text-to-speech model from Resemble AI that supports 23 languages with zero-shot voice cloning. Built on a 0.5B Llama backbone and trained on 500,000 hours of cleaned audio data. Licensed under MIT for full commercial use.

What’s New

Chatterbox is the first open-source TTS to support emotion exaggeration control, letting you dial up or down the intensity of generated speech.

The model includes built-in watermarking for responsible AI use and outperforms ElevenLabs in side-by-side evaluations.

It also supports Arabic, Chinese, French, German, Hindi, Japanese, Korean, Spanish, and 15 other languages out of the box.

Why It Matters

Founders building voice applications, AI agents, or content tools can now own the entire speech generation stack without API costs. Chatterbox runs locally, ships under a permissive license, and delivers production-quality output. No per-minute fees. And you have full control over the pipeline.

Founder’s Edge

This Week’s Builder Toolkit

• Dev Tool: WaveSpeed AI generates images in under 2 seconds and videos through a unified API. Supports FLUX, WAN, Seedance, Nano Banana, and many other SOTA models with day-zero access when new models drop.

• Free Dataset: The Stanford HAI 2025 AI Index Report provides 400+ pages of data on AI R&D, technical performance, economic impacts, and public opinion. It’s the most comprehensive annual snapshot of the field.

• No-Code App: Adalo builds native iOS and Android apps with drag-and-drop components and direct App Store publishing. AI-assisted guidance and pre-built OpenAI integrations let non-technical founders ship functional MVPs faster.

• Productivity Hack: Heptabase turns complex research into visual maps. Drag cards onto infinite whiteboards, connect ideas spatially, and use built-in AI to ask questions across your entire knowledge base. Ideal for founders synthesizing market research, competitor analysis, or product strategy.

• Learning Resource: LLM University by Cohere offers free courses on building with LLMs. It covers embeddings, retrieval, prompt engineering, and fine-tuning. Plus, it has hands-on exercises and certification.

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 Lovable Hit $200M ARR in 12 Months

In June 2023, Anton Osika was CTO at a YC startup when he got frustrated. His team dismissed what LLMs could do. “People in my team felt like, oh Anton, you’re exaggerating, this isn’t going to change anything in the coming years.” He wanted to prove them wrong.

So, he hacked together an open-source tool called GPT Engineer. It let users describe a web app in plain English, and the AI would generate working code. He posted it on GitHub and went to sleep.

He woke up to viral attention, thousands of users, over 50,000 stars, and messages from non-coders building their first apps. That response crystallized the core insight: “99% of the world’s best ideas are trapped in the heads of people who can’t code.” Anton called his future co-founder Fabian Hedin at 6 AM. By 8 AM, they’d agreed to build a company around it.

But the first two launches flopped.

They launched gptengineer.app for the first time in April 2024. Instead of becoming a GitHub repository only, they became a real app. The product was good, but not great. The AI code builder was still unreliable for complex apps. Growth flatlined. Six months later, they tried again. Same result. The AI kept getting stuck on larger codebases.

The team went back to the technical foundation. They found a system that allowed Lovable to handle large codebases, making it the first AI-powered full-stack engineer.

But almost no one knew how much better the product had become. The GPT Engineer brand was associated with the broken version.

So they killed it, rebranded to Lovable, and launched fresh in late November 2024.

The pivot worked immediately.

$4 million ARR in the first four weeks. $10 million ARR in two months with a team of only 15 people. $30M ARR in 120 days.

Here’s what made it compound

The product became distribution. Every app built on Lovable comes with a shareable URL. Users share their creations. Recipients see “Built with Lovable” and try it themselves. Each project is a live demo.

They removed the coding floor. GPT Engineer still required some dev knowledge. Lovable stripped that away entirely. Non-technical founders could describe what they wanted and get a working app. The shift dramatically expanded the target audience, from devs to founders, product designers, and marketers.

Community as growth engine. By first releasing the CLI tool to the open source community, Lovable built a passionate early user base. This community-driven approach not only validated the product but also created buzz that propelled its subscription growth post-launch.

One month after launch, GitHub suspended their app without warning, taking down the entire platform for 500,000 users. They thought the rapid repository creation was suspicious activity, but it was actually a sign of explosive growth.

By August 2025, Lovable crossed $100M ARR. In November 2025, Bloomberg reported that Lovable had reached $200 million in annual recurring revenue. Last week, they closed a $330M Series B at a $6.6B valuation.

Key Takeaway

If your product isn't working, don't keep pushing the same brand. Kill it, rebuild the foundation, and relaunch fresh. Sometimes the best distribution strategy is starting over with a better product.

— Anton Osika, Lovable co-founder and CEO

Weekly Challenge

One Experiment. One Week. One Win.

The Goal

Add one distribution channel you don't currently use and get 100 new signups in 7 days.

How It Works

Audit your current acquisition sources (organic, paid, referral, content, partnerships).

1. Identify ONE channel you've ignored or underutilized.

2. Spend 1-2 hours setting up a presence or campaign.

3. Track signups from that channel separately.

Why It Works

Most founders over-index on one or two channels. The highest-growth companies diversify. They don't have one great channel, they have ten decent ones that add up.

Spotlight

Share your channel choice, setup process, and results in the TAAFT community by end of week. We'll showcase the highest-impact new channel discovery in an upcoming newsletter.

AI Market Watch

Deals, Discoveries, and Demand

Megadeals

• Cyera – $400M (AI data security)

• Runware – $50M Series A (AI inference API)

• Databricks – $4B+ Series L (data + AI platform)

• Neural Concept – $100M Series C (AI engineering platform)

• Unconventional AI – $475M seed (energy-efficient AI compute)

Top Research

• EffiR (efficient LLM-based dense retrieval)

• Nemotron 3 (NVIDIA's MoE for agentic AI)

• Step-DeepResearch (32B research agent)

• HiStream (efficient 1080p video generation)

• MemR³ (reflective reasoning for LLM agent memory retrieval)

• TurboDiffusion (video generation acceleration via distillation)

Search Trends

Top “best AI for…” searches:

1. Shopping

2. Research

3. Presentations

4. Image generation

5. Video editing

Thanks for reading!

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

Keep an eye on your inbox. We’ll be back soon with more founder stories, fresh tools, and data worth your time.

Signing off,
— AI Empires