From 0 to $50M ARR to Acquired

Here's how a 17-year-old took on MyFitnessPal and won.

March 09, 2026

Welcome back!

This week we break down the activation data behind why most signups never see your product’s value, walk through how a solo engineer with no design skills built SuperCraft into a profitable platform with zero ad spend, explore how Wispr Flow made typing feel broken, and show how a 17-year-old took on MyFitnessPal and got acquired in 18 months.

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

Founder’s Intel

Don’t track signups. Track activation.

Data Intel

  • Only 40-60% of users who sign up for a products ever complete the core action that defines value.

  • Products with a defined activation milestone see 2-3× better 30-day retention than those without one.

  • Reducing time-to-activation by 50% has been shown to increase trial-to-paid conversion by up to 20%.

Why It Matters

Signups are a vanity metric if nobody reaches the moment your product proves itself.

Every AI founder has a mental model of what their product does.

Most have never mapped the exact sequence of actions a new user takes to get there, or measured how many drop off at each step.

That gap is where growth dies quietly.

Activation isn’t onboarding.

  • Onboarding is the welcome email and the setup wizard.

  • Activation is the specific moment a user experiences the value they signed up for.

For a writing tool, it’s finishing the first draft. For a coding tool, it’s shipping the first working function. For a research tool, it’s the first answer that saves real time.

Once you define that moment precisely, you measure it. Then you do one thing: shorten the path to it.

The structural version of this question is more useful than the tactical one. Not “how do we guide users to activation” but “why does activation require this many steps at all.”

The ones with strong retention almost always have a short, obvious path to the first win. The ones struggling with churn often have the right product but a buried activation event that most users never reach.

Quick Tip

Write down the single action that makes a new user say “this works.” Map every step between signup and that action.

Count the steps and cut two of them.

Behind the Tool

The Engineer Who Couldn't Sketch

The Spark

Before SuperCraft, Sarang Zambare was building defect detection systems for factory floors.

He needed to show manufacturers what his industrial cameras might look like. No design skills. No budget for a designer.

So he started prompting Stable Diffusion and handing the outputs to engineers.

He expected them to laugh. They didn’t. They pushed back on dimensions, flagged proportions, asked for different angles.

The images were rough. The conversation was real.

That was the insight. Non-designers could communicate product ideas through AI-generated visuals faster than anyone had thought. And nobody had built software specifically for that gap.

The Build

SuperCraft is a collaborative canvas where teams design physical products using natural language.

Describe what you want. Get photorealistic renders and 3D models. Iterate without ever opening SolidWorks.

The wedge is the gap between “I’ve a product idea” and “I’ve a CAD file.”

Traditional design tools assume you arrive with a sketch. SuperCraft starts before that. A sentence is enough.

Sarang built it alone. Frontend, backend, the image models. No co-founder. No team.

The Breakthrough

SuperCraft spent $0 on paid marketing.

A Korean YouTuber discovered the product and made a tutorial showing how she used it to design handbags. The video hit 14,000 views inside an hour. Korean signups flooded in fast enough that Sarang added Korean language support the same week.

Then the product spread on Red Note in China.

Then something stranger happened. Users started writing in: “ChatGPT told me about SuperCraft.” No ad, no placement. The model was recommending it unprompted.

TAAFT showed up the same way. Sarang was manually emailing new users asking how they had found the product. TAAFT kept appearing in the replies.

The Next Chapter

Sarang is building toward AI that reasons inside the design space, not just generates.

The difference: a model that produces an output, evaluates it against your intent, and corrects it before you ever see the mistake.

He thinks that loop, applied to physical product design, hasn’t been built properly yet.

Key Lesson

Sarang’s advice, shaped by YC and proven through SuperCraft is don’t build for a thousand people who might use your product.

Find ten who genuinely need it, solve their exact problem, and expand from there.

A tight niche isn’t a ceiling. It’s the foundation everything else gets built on.

Tool of the Week

Wispr Flow’s Word-of-Mouth Machine

What’s Wispr Flow?

An AI voice dictation tool that works system-wide on Mac. Speak into any app, any text field, anywhere, and Wispr transcribes and cleans your words into polished prose in real time. No copy-paste. No switching windows. Raised a $50M Series B in 2025 led by Sequoia.

What Worked

  • Wispr Flow solved a problem people didn’t know they had until they tried it. Nobody was sitting around wishing they’d dictate emails. Then they tried it for one week and typing felt slow. That irreversibility is the product’s strongest growth mechanic. Users who cross that threshold don’t churn because going back feels like a downgrade.

  • No integration required set it apart from every competitor. Every other dictation or AI writing tool asks you to use their editor, their interface, their workflow. Wispr Flow works inside whatever you already use. Gmail, Notion, Slack, Linear, Claude. That zero-friction install meant the product spread through teams naturally. One person starts using it, others notice the speed, ask what it’s, and install it the same day.

  • Early traction came from a high-signal audience. Wispr Flow found its first users among founders, operators, and investors, people who write constantly, recommend tools loudly, and have followers who trust their opinions. When a well-known founder posts that Wispr Flow changed how they work, that post reaches exactly the audience most likely to try it and pass it on.

  • Personalization made it stickier over time. Wispr Flow learns your vocabulary, your style, and the context of what you’re writing. The longer you use it, the more accurate it gets. That compounding accuracy creates switching costs that grow without the product having to manufacture them.

Founder Quote

“For the first time in human history, people have stopped using the keyboard altogether. Voice isn’t a feature. It’s becoming the new way humans interact with technology.” — Tanay Kothari, Wispr Flow CEO

Key Lesson

The strongest retention mechanic isn’t a loyalty program or a feature gate. It’s a product that makes the previous way of doing something feel broken.

Wispr Flow didn’t need to explain its value in a landing page headline. Anyone who used it for 20 minutes felt it.

That felt experience is what drives the word-of-mouth. People don’t share Wispr Flow because they’re prompted to. They share it because they want the people around them to stop being slow.

If your product has that quality, even one person in a team or community is enough. The people who try it become the distribution.

Fresh Out of the Lab

nanochat

What Is It?

A minimal, from-scratch implementation of a chat-capable language model by Andrej Karpathy. It’s built on the same philosophy as nanoGPT. Strip everything down to its essential parts so builders read every line, understand every decision, and run the whole thing on a single machine.

What’s New

nanochat extends the nanoGPT foundation with the structures that define how modern chat models work in practice.

Conversation formatting, turn-handling, and the mechanics that separate a base language model from something you interact with.

The implementation stays deliberately small, making it a readable foundation for any builder who wants to understand how instruction-tuned models handle dialogue without wading through production codebases.

Why It Matters

Most founders building on top of LLMs treat the model as a black box. That’s fine until something breaks, a prompt behaves unexpectedly, or you need to fine-tune on your own data.

Karpathy’s nano series has become the fastest path to genuine understanding of how these systems work at the layer below the API.

If you’ve wanted to understand why certain conversation formats outperform others, or how training shapes the chat behavior you see in production, this is where you start.

Founder’s Edge

This Week’s Builder Toolkit

  • Dev Tool: Plausible is a lightweight, open-source analytics platform that gives you clean traffic and event data without cookies, GDPR headaches, or sending your users’ data to Google. Self-host it or use the cloud version.

  • Free Dataset: Sensor Tower’s State of Mobile 2025 report covers download trends, revenue benchmarks, and category performance across iOS and Android. Useful for validating market size before you build or benchmarking your app’s growth against category averages.

  • No-Code App: Glide turns a Google Sheet or Airtable into a polished mobile app in minutes. If you need an internal tool, client portal, or lightweight MVP that works on iOS and Android without writing Swift or Kotlin, this is the fastest path to a working prototype.

  • Productivity Hack: Screen Studio makes screen recordings look professional without any editing. Zooms follow your cursor automatically, background blur is built in, and exports are clean enough to post directly.

  • Learning Resource: Anthropic’s free course library covers prompt engineering, tool use, multi-step reasoning, and building with Claude in production. Practical and up to date. Find it here.

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

From Zero to Acquisition

By 2023, calorie tracking was a solved problem. Or so everyone thought.

MyFitnessPal had tens of millions of users. Cronometer had a loyal base. Every major app shared the same core mechanic. Search for what you ate, scroll through a database, log it manually.

Accurate enough. Tedious enough that most users quit within a week.

Zach Yadegari was 17 years old and already thinking about that drop-off.

Not the nutrition science, not the UI, not the monetization model. The moment people stopped showing up.

His answer was simpler than anyone expected. What if you took a photo?

Cal AI was a calorie tracker that used AI to identify food from a photo and log it automatically. No searching. No scrolling. Point the camera at the plate and it’s done.

The idea wasn’t technically new.

The execution was clean, the design was polished from day one, and the team knew exactly how they were going to get users before they ever submitted to the App Store.

The first channel was UGC in the fitness space.

They reached out to fitness creators on TikTok and Instagram who already had audiences of people actively tracking their health.

Not celebrity athletes. Everyday fitness creators posting morning routines and meal prep content to audiences of 50K-500K.

The arrangement was straightforward. Post about the app, show it working, get paid.

The content felt native because it was native. These creators used calorie trackers themselves. The integration was subtle enough that viewers didn’t feel sold to.

That strategy got Cal AI to roughly $1-2M per month.

Then they hit the ceiling.

The fitness creator pool in the US is finite. The biggest names had been worked. Audiences overlapped. Cost per new user was rising.

So they expanded the creative definition of who promotes a calorie app.

Lifestyle creators. Wellness influencers. College content.

They partnered with Mr. Beast for a video centered on a fitness transformation challenge, a $500K deal that the team acknowledges was slightly unprofitable on direct attribution.

But the brand authority it generated opened doors with future creator partners and gave Cal AI a credibility signal that most bootstrapped apps never get.

The real shift came when they layered in a performance channel alongside influencer.

The influencer creative that’d worked organically didn’t translate directly to paid placements. A 3-second product integration woven into a morning routine video outperformed on organic reach but underdelivered as a paid unit.

Users clicking a paid placement want direct information fast.

So they built a separate content operation specifically for that channel. Not creators showing their day. Creators showing the app, directly, in the first five seconds.

Here’s the problem, here’s the solution, here’s what it looks like in 30 seconds.

Attribution across multiple simultaneous channels was its own problem.

The App Store lets developers create custom product pages with distinct URLs. Each campaign linked to its own page.

Revenue generated from that page was trackable directly. When influencer traffic and performance spend were running simultaneously, the custom pages kept the signal clean enough to make real decisions.

An affiliate program, run through Tribe, added a third layer.

Creators made videos independently, earned a percentage of the revenue their content drove, and had financial incentive to produce the kind of direct-response creative that converted. The team wasn’t directing every shoot. The economics were.

By January 2025, Cal AI was doing $5.7M in monthly revenue. ARR had crossed $50M.

Eighteen months after launch, MyFitnessPal, the dominant player in the category they’d been quietly eating into, acquired 100% of the company.

Here’s what made it compound:

The product removed the activation barrier. Cal AI wasn’t dramatically better than competitors at calorie tracking. It was significantly easier to try for the first time. The photo-first mechanic got users to their first logged meal in seconds rather than minutes. That single structural advantage shaped every retention metric downstream.

Distribution matched the stage. UGC worked at zero budget. Influencer deals worked with cash flow. The performance channel worked once they’d enough creative volume to generate signal. Each channel had a moment. None of them were permanent.

Speed was a core value, not a tagline. Zach ran the company with an explicit directive. No one on the team should ever be a bottleneck. If something was blocked, it got addressed immediately, not in the next sprint.

The brand was built into the product. Cal AI shipped two versions of the app, one for users and one for creators. The creator version displayed the app name prominently on every screen and showed an idealized usage streak to make the content aspirational. Every creator video became an implicit ad, optimized without the creator having to think about it.

They hired behind the growth, not ahead of it. For most of the run, the founders were the marketing team. When the company was ready to scale, they brought in operators who’d done it before. The 30-person team that reached the finish line looked nothing like the team that started.

Key Takeaways

The fastest path from zero to acquisition isn’t the best product in the category. It’s the easiest product to try, distributed through the channel your competitors aren’t using yet, moving faster than anyone expects.

— Zach Yadegari, Cal AI co-founder

Weekly Challenge

One Experiment. One Week. One Win.

The Goal

Define your activation event and measure what percentage of this week’s new signups reach it.

How It Works

  1. Write down the single action that makes a new user say “this works.” Be specific. Not “they explore the product.” The exact action.

  2. Pull the data. Of everyone who signed up in the last 7 days, how many completed that action?

  3. If that number is below 60%, identify one step in the path you can remove or simplify this week.

Why It Works

Most teams optimize for acquisition when their actual constraint is activation. One week of measurement tells you more than three months of guessing.

Spotlight

Share your activation rate and the one change you made in the TAAFT community by end of week. We'll showcase the highest-converting content strategy in an upcoming newsletter.

AI Market Watch

Deals, Discoveries, and Demand

Megadeals
Top Research
Search Trends

Top “best AI for…” searches:

  1. Coding

  2. Research

  3. Presentations

  4. Studying

  5. Video Generation

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