I was at my desk when an AI optimized a theme-park queue in RollerCoaster Tycoon and revenue numbers blinked upward. You felt the tiny chill that arrives when software starts making money without a human hand on the wheel. The founders then asked a blunt question: what happens if we give it a real company?
This Startup Wants to Buy a Company Just to See What an AI CEO Does With It
Skyfall AI quietly left stealth this week with an offer that reads like a tech thought experiment: the team will buy a small SaaS or e-commerce business for up to $1 million (€920,000) and hand operational control to an AI designed to make long-term executive decisions. I want you to track what they test and why it matters—because this is less about replacing tasks and more about testing whether software can steer a business toward growth.
In a demo, a simulated theme park’s daily revenue curve changed after an algorithm tweaked prices.
Skyfall’s founders—Sam Pasupalak, Kaheer Suleman, and Sumit Pasupalak—say they started by training their systems inside RollerCoaster Tycoon. That experiment let them chase a clear signal: adjust admission, concessions, and ride mix, and watch cash flow move. The success there convinced them to take the risk with a real customer base and real invoices.
Can an AI actually replace a CEO?
Short answer: not today, but the experiment is asking smarter, narrower questions than headline-grabbing claims. Skyfall’s pitch is surgical: have an AI oversee pricing, marketing, customer support, finance, and operations, then steadily reduce human intervention while targeting a doubling of revenue in six months. They’re not promising miracles—just measurable decisions and continuous learning.
On a whiteboard I counted the list of tasks a CEO delegates each week and it ran longer than I expected.
The team argues that large language models such as ChatGPT, Claude, and Gemini are limited by a single paradigm—big models, more data. Skyfall calls for “enterprise world models,” systems built to learn continuously from changing business conditions. You should note the name-dropping: Microsoft acquired Maluuba, the founders’ previous startup, and big labs such as Anthropic have already tested AI with physical kiosks and vending machines.
Skyfall published a benchmark that shows today’s models do well on short tasks but struggle with decisions whose consequences arrive months later. Their fix: keep the model aware of history, tradeoffs, and delayed outcomes instead of treating each prompt as a fresh problem.
How will an AI handle pricing, marketing and support?
Skyfall claims the AI will run pricing strategies, iterate marketing funnels, route and respond to support tickets, and manage cash flow while learning from results. In research, the model rebalanced ad spend and pricing in simulations, then measured downstream churn and lifetime value. You should think of the AI as a continuous optimizer that revises plans based on measured outcomes, rather than a script that executes a single campaign.
I watched a founder argue with their code like it had opinions and a memory.
Before buying an actual firm, Skyfall tested inside simulations. The founders told Forbes the RollerCoaster Tycoon demo gave them confidence to move into real businesses; they’ve hinted that if the experiment succeeds, they will scale to firms worth tens of millions next year. That’s ambitious fundraising choreography—raising expectations while trying to keep control of real customer relationships.
They’re not the first to toy with letting AI run things. Anthropic ran a vending machine, Andon Labs experimented with a coffee shop in Stockholm, and reports suggest Mark Zuckerberg is building agents to assist him at Meta. But Skyfall says it will be the first to buy an operating business specifically to prove an AI can act as CEO.
A customer called to complain and the system rewrote the reply in under a minute.
Those micro-interactions will be a major test. Customer support is messy: tone, escalation, refunds, and public relations all matter. The AI can draft responses and handle routine refunds, but the team plans to phase humans out gradually so edge cases remain human-supervised early on. If the system truly learns from each resolution, it may shrink response time and error rates—but you should watch for new failure modes, like brittle policies or skewed incentives.
One metaphor: the AI behaves like a thermostat that learns your schedule—small adjustments, then a new baseline. That’s efficient in stable conditions but risky when the system misreads a sudden spike.
The boardroom I visited had one question: who takes liability when the algorithm is wrong?
Legal and governance questions are inescapable. If pricing algorithms trigger price gouging or an automated support reply violates consumer law, accountability still rests with humans. Skyfall’s plan to reduce human involvement “gradually” is a hedge, but regulators and customers won’t accept automation as an excuse for harm.
There’s also the talent argument: founders at Skyfall say they envision a future where people trade operations for creative work. That’s an appealing image, and it’s easy to sell hope. I warned the founders that business operations are social systems—culture, trust, and incentives can’t be fully encoded in a model overnight.
I read their benchmark and felt both impressed and skeptical at once.
Technically, Skyfall’s research is disciplined: they measure models against shifting business states and longer horizons. Practically, success will depend on the quality of data, the granularity of feedback, and whether human stakeholders accept machine decisions. You’ll want to watch for cherry-picked results and test cases that favor incrementally measurable wins over messy, real-world complexity.
Second metaphor: their approach is a small sailboat testing a new current—you can learn speed and drift, but storms reveal the real limits.
Skyfall’s founders have a story about freeing people from mundane tasks. That’s persuasive rhetoric, but the test they’ve chosen is blunt and measurable: buy a company for up to $1 million (€920,000), run it, and try to double revenue in six months. If they succeed, you’ll see other firms copy the blueprint; if they fail, you’ll learn where automated leadership breaks down.
I’ll be watching their first acquisition, the KPI dashboards, and the point at which humans step back. Will you hand your business to an AI CEO?