I'll be honest about where I started with the mobile app idea.
I started skeptical. "Build an app" has been the entrepreneurial advice equivalent of "just add bacon" — it sounds good, it's usually offered by people who haven't tried to do it recently, and the actual execution is harder than it sounds. I've met exactly two people who built apps that generated sustainable income. I've heard from many more who built apps that went nowhere.
That prior made me cautious about including mobile apps in the portfolio. But then I started stress-testing my skepticism, and I found that most of my resistance was based on what app development looked like before AI — and that context has changed enough to revisit the thesis.
Why the old app development math was discouraging.
The traditional model for an indie app looked something like this: you have an idea, you hire a developer (cost: anywhere from $5,000 to $50,000+ depending on complexity and geography), you ship the app, you discover that distribution is the hard problem and technical development was the easy one, you spend more money on marketing or user acquisition that may or may not work, and you find out six months later whether the economics make sense.
The failure rate in that model is high. Not because the ideas are bad, but because the capital requirement and the complexity of execution meant that any single mistake — wrong feature set, wrong pricing model, wrong target market — was costly enough to kill the project before you learned enough to get it right.
That model selects strongly for people with either deep technical skills, significant capital, or both. It doesn't leave much room for someone like me: a writer with legal training who has good ideas and good judgment about what users need, but who would have been completely dependent on an outside developer to build anything.
What's changed.
Two things have shifted the calculus enough to make app development worth testing.
The first is AI-assisted development. The ability to take a clear product specification and work iteratively with AI tools to build functional software — without being a trained software engineer — has moved from theoretical to practically achievable for a certain class of applications. Not complex, enterprise-grade software. But functional utility apps, productivity tools, niche information products, simple games. The kinds of things that can genuinely solve a problem for a defined user base.
I want to be measured about this. "AI helps you build apps" does not mean "anyone can build a great app with zero technical background." You still need to understand what you're building, why it works, and how to debug it when it doesn't. The bar is lower than it was. It is not zero.
The second thing that's changed is the app store distribution reality. Both Apple and Google have settled into a model where searchable, well-rated utility apps with clear value propositions can find organic users without enormous marketing budgets — especially in underserved niches. The gold rush days of "build any app and get paid" are long over. But the niche utility space, where you're not competing against well-funded categories, continues to produce viable small businesses.
What we're testing and why.
We're in early-stage evaluation of a small number of app concepts — utility applications in niches where we believe there's demand and limited existing supply, priced for small recurring subscriptions rather than one-time purchases, built to run on both iOS and Android from day one.
The thesis we're testing: can a small operation produce a functional, well-reviewed niche utility app at a cost structure that makes the economics viable? And can that app generate subscription revenue at a scale that justifies the build and maintenance investment without requiring ongoing active marketing?
We don't know yet. This is exactly the kind of question that can only be answered by doing it.
What I'm watching: the build time and cost, the initial App Store traction, the review sentiment (which tells you whether the app is actually delivering value), and the retention rate after the first month. Retention is the number that matters in subscription software — it tells you whether people are getting the value they paid for or just forgetting to cancel.
The honest risk assessment.
Apps take longer to build than you think they will. Distribution is harder than technical development. The niche utility space has gotten more crowded, even if it's not as crowded as the mainstream app categories. And subscription fatigue is real — users are more resistant to adding another monthly charge than they were three years ago.
These risks are real. I'm including this stream because the risk-adjusted upside is still interesting given what AI has done to the cost side of the equation. But I am not including it because I'm confident it will work.
Confident experiments are the worst kind. They stop paying attention too early.
This Week in AI: Apple and Google have both been updating their developer policies around AI-generated app content and AI-assisted development, trying to get ahead of a wave of low-quality apps built entirely by automation. The practical guidance for developers is to ensure genuine utility and originality — which, again, is just good product development advice regardless of how the app was built.
Follow along as the app experiments develop — The Upstream community gets updates before they hit the newsletter. Join at start.tenstreamslab.com.