The One-Person Unicorn Sounds Crazy – Until You Narrow the Scope

01 /The Concept

Rocketable is building a holding company made up of software businesses – but not just any kind. Its founder calls it an "AI-maximalist holding company."

The company went through Y Combinator earlier this year and quietly closed a $6.5M seed round in the spring, announcing it only at the start of this month.

The general logic of a software holding company is familiar enough. What makes Rocketable different requires a bit more explanation.

02 /Why It Matters

In early 2024, Sam Altman predicted we'd soon see billion-dollar companies run by 10 people – and eventually by just one, the founder, with AI handling everything else. Rocketable's founder, Alan Wells, published a manifesto in January of this year taking that idea to its extreme: billion-dollar software companies run entirely by AI, with zero employees.

Alan's plan is to turn Rocketable into a multi-billion-dollar holding company made up of billion-dollar AI-run software companies – with himself as the only human in the entire structure. His job: identify and acquire existing software businesses, remove the people, and hand operations to AI.

The acquisition targets are specific: cloud services that have been running successfully for at least two years, with annual recurring revenue between $500K and $2M, already profitable or on a clear path to it. Rocketable promises to evaluate a deal in two weeks and close in under two months. Within one quarter of closing, it says it will transition the acquired company to full AI operations – freeing the original founder to move on to whatever comes next.

The idea came from Alan's own experience. In 2023, he bought a small, profitable cloud service and spent time figuring out how far he could push AI automation. The results were encouraging enough to try it at scale.

In January, the full vision sounded fantastical. By March, Alan had started filling in the details – and the picture got more grounded.

The acquired service had 100,000 lines of code and no documentation. Rather than work through it manually, Alan handed the whole thing to AI. Claude analyzed every line and generated comprehensive documentation. He then ran cross-reviews using OpenAI and Google models – each checking the other's work, catching gaps, and adding improvements. Total cost: 48 hours and $100.

After that, Claude could take feature requests and implement them. But the more interesting finding was that Claude was even better at writing its own prompts than executing instructions directly. Give it a feature request, and it would first draft the implementation prompt, then write the code against that prompt, then test the result for correctness and bugs – running the full improvement loop autonomously.

Want to add a feature users have been asking for? Feed Claude the user feedback with a short summary, give it repository access, and it figures out how to implement it within the existing architecture. Want to improve the UI? Let Claude take screenshots as it navigates the product pages, and it audits and updates them on its own. Want a code review? Have Claude generate the prompt a world-class CTO would use, then run the same code through Codex and let the two models debate each other to surface the best version.

The limit is real: using AI to introduce fundamentally new product directions or redesign architecture is still science fiction. But getting AI to continuously improve and extend what it – or a predecessor model – already built? That's already happening.

So Rocketable refined its thesis. The key shift happening now isn't that AI can write code better than humans. It's that AI can improve what it has already built or reverse-engineered. That's the specific capability Rocketable is designed to exploit: buy working, profitable software products and turn them into self-improving systems under AI management. The company itself becomes a permanent laboratory for testing how far that capability can go – given whatever AI technology happens to exist at any given moment.

If the initial vision was "AI-run companies," the operating version is narrower and more honest: take something a human invented and proved out, then let AI handle the ongoing improvement cycle. Nothing beyond that – but also nothing less.

03 /Opportunities

Every product and business cycles through phases. Periods of sharp, qualitative leaps – new architectures, new markets, new capabilities – alternate with longer stretches of incremental improvement and expansion on whatever the last leap produced. Rocketable is explicitly positioning itself to own the incremental phase, which is historically the longest and often the most profitable.

The business model is coherent. Acquiring proven, cash-flowing software products and running them more efficiently with AI is a real arbitrage opportunity – especially right now, while most of the market still treats "AI-managed company" as a thought experiment rather than an operating strategy.

The open question is what happens when an acquired product needs a genuine step-change to stay relevant and Rocketable's AI can't provide it. The honest answer is a shrug and a pragmatic bet: either the product has already returned the investment before that moment arrives, or AI capabilities will have advanced far enough to handle it. As they say – we'll cross that bridge when we get there.

The scale of the opportunity is significant. There are enormous numbers of software products in the world, the vast majority of them in "steady improvement" mode. And a meaningful share of their founders are tired – tired of maintaining something that works but doesn't excite them anymore, tired enough to leave but unable to simply walk away from revenue and users who depend on them.

Rocketable's model solves that problem for both parties. For buyers: a defined acquisition path with a realistic integration plan. For sellers: a clean exit without abandoning users.

The first way to move: build something similar before the model becomes crowded. The window where this looks like an eccentric bet rather than obvious strategy probably isn't that long. A narrower version of the same play – apply the model to one specific software vertical rather than the full market – could be an easier starting point with lower capital requirements.

And it doesn't have to be limited to software. The Anthropic team recently published results from a month-long experiment in which Claude autonomously managed a snack and beverage vending machine in their office – controlling inventory purchasing and pricing. The results were poor: revenue declined steadily throughout the month. But the team's response wasn't to stop. They've improved the algorithms and moved to the next phase of experiments.

They're working within the same conceptual frame as Rocketable, applied to a physical business. The fact that they're doing it at all is its own signal. There's something real in this direction – worth digging into.