The AI industry’s biggest players are all chasing the same destination: the autonomous employee. OpenAI is transforming ChatGPT into a digital coworker, Anthropic wants Claude orchestrating enterprise work, and Google continues weaving Gemini into business software. Their shared bet is that companies will automate individual jobs piece by piece until humans oversee the work rather than perform it.
Skyfall AI’s founders believe the enterprise AI industry is chasing the wrong goal. Instead of validating its technology through controlled customer pilots, the startup plans to acquire a small B2B SaaS or e-commerce company for as much as $1 million and attempt to run it with AI as the CEO.
“We want to democratize the power of a CEO to all the small businesses around the world, rather than building another marketing or sales agent,” Sam Pasupalak, Skyfall’s cofounder and CEO, told me in an exclusive interview. “If you want a truly autonomous business, you have to buy a business and operate the whole thing end to end. Unless you run a business with minimal human intervention, you’ll never know whether an autonomous enterprise is actually possible.”
The system will oversee pricing, marketing, customer support, finance and operations while progressively reducing human involvement, with an ambitious objective: double the company’s revenue while publicly documenting both its successes and failures. It’s an unusually public and expensive way to validate a research hypothesis.
Rather than building what Pasupalak describes as the “next virtual software engineer,” Skyfall has set its sights on what it believes is a far more difficult challenge: a virtual CEO. Unlike software engineering, which he characterizes as largely logical and deterministic, running a business demands abstract reasoning, long-term planning and high-stakes decisions made with incomplete information.
“A lot of startups are building marketing agents or sales agents, which in our belief is very narrow AI implementation—rather than tackling the broader challenge of organizational decision-making,” Pasupalak says.
Skyfall Claims LLMs Have Reached Their Enterprise Limits
Pasupalak and cofounder and CTO Kaheer Suleman met nearly two decades ago while studying computer science at the University of Waterloo, bonding over an ambitious goal: building intelligent machines during what many now call the AI winter. After graduating in 2011, they cofounded Maluuba, one of the earliest deep learning startups, working alongside Turing Award winners Yoshua Bengio and Richard Sutton.
The company’s early conversational AI systems reached hundreds of millions of devices through partnerships with Samsung, LG and BlackBerry. “Around 2016, one of the demos we built could answer questions about an entire Harry Potter book. Today, post-LLM, models do that across the web. But ten years ago, that was cutting-edge research,” Pasupalak recalled.
Microsoft acquired Maluuba for approximately $160 million in 2017, turning the lab into the foundation of Microsoft AI in Canada. Nearly a decade later, the founders have reunited under Skyfall with a new ambition: building what they believe could become the first autonomous enterprise.
The founders believe the industry’s direction is driven as much by economics as by technology. With billions of dollars invested in AI infrastructure and frontier LLMs, companies have strong incentives to demonstrate that the current architecture continues to improve.
“If you think about the big players—whether it’s OpenAI, Anthropic or Gemini—their hook, line and sinker is basically LLMs,” he says. “All their resources are on LLMs. So they have to create benchmarks, or partner with companies that do benchmarks, where LLMs are shown to have high accuracy. The result is that the entire research community, since so much money has been poured into this, is just going for n-plus-one research improvements.”
Skyfall is emerging from stealth alongside Morpheus, a benchmark designed to test whether today’s frontier AI models can continually adapt as enterprise environments evolve. According to the company’s latest research, models such as GPT-5.5 and Gemini perform well in familiar business conditions, but deteriorate as competitors launch products, customer behavior shifts and market dynamics change—evidence, Skyfall argues, that current systems excel at applying pre-trained knowledge but struggle to learn continuously from experience, and that scaling today’s architecture alone may not be enough to run an enterprise.
Its answer is what the founders call Enterprise World Models—AI systems designed to understand how an organization evolves, simulating how decisions ripple across a business, from pricing and customer acquisition to hiring, operations and long-term revenue.
Ultimately, Pasupalak doesn’t envision AI replacing meaningful human work so much as eliminating the operational burden that prevents leaders from focusing on it. “We want all businesses to be run autonomously because we believe people should spend their time on things they’re passionate about—not operational tasks. We want to build the autonomous era.”
Why Buying A Real SaaS Company Matters
Skyfall wants to test its ideas in public—with real customers, real revenue and real operational consequences. The founders have spent years training their systems in increasingly sophisticated simulated business environments, but argue that no matter how realistic those become, they can’t capture the unpredictability of actual customers, competitors and markets.
“We found that you’ll always be limited by the gap between simulation and the real world,” Suleman says. “The only way to solve that is to actually have real data. There’s no way around that if you want to solve the sim-to-real problem.”
Before trusting AI with a real company, however, they first had to determine whether it could operate an entire business—even a simulated one. “Last year we started with RollerCoaster Tycoon,” Pasupalak explained. “You’re essentially operating an entire theme park and trying to maximize revenue. That gave us confidence that we could operate inside a simulated business. The next step is buying a real company. After that, perhaps next year, we’ll look at automating businesses worth tens of millions of dollars.”
The founders assert that explicitly modeling a company’s evolving state makes AI better suited to long-term planning and high-stakes decisions than today’s prompt-driven agents.
“We’re building what we call a latent world model,” Suleman says. “Instead of predicting future states in token space, we’re predicting them in a latent space. That allows us to focus only on the features that actually matter to the dynamics of the business.”
Skyfall Isn’t The First, But It May Be The Most Ambitious
Skyfall isn’t entering completely uncharted territory. The closest comparison is Anthropic’s Project Vend, developed with AI safety startup Andon Labs, which gave a Claude-powered agent control of a real office vending business.
The first version stumbled spectacularly—employees persuaded it to stock novelty tungsten cubes and sell them at a loss—but later iterations became considerably more competent. By mid-2026, the experiment had expanded to multiple locations across San Francisco, New York and London, with AI colleagues providing oversight, and Andon Labs said the operation had become “almost boring” because it was running so smoothly.
Earlier attempts at appointing an AI CEO—NetDragon’s virtual executive Tang Yu or Dictador’s humanoid CEO Mika—never approached real autonomy; both functioned largely as decision-support experiments. What makes Skyfall different is less the ambition than the willingness to make the hypothesis falsifiable: buy a real revenue-generating business, publicly measure whether AI can grow it while reducing human intervention, and document the failures alongside the successes.
Can AI Really Become The CEO?
Even Pasupalak doesn’t believe AI should replace every aspect of leadership. He draws the line, at least for now, at human relationships.
“More than half of my time goes into operations—meetings, budgeting, coordinating teams, campaigns. Imagine if AI handled all those operational tasks. Then I could spend far more time on strategy,” he says. “When it comes to motivating employees and building relationships with people, that’s still very difficult for AI to replicate.”
Likewise, even if AI eventually takes on much of a company’s operational workload, Suleman doesn’t believe humans will disappear from the equation. Instead, he envisions their role evolving from executing day-to-day tasks to providing oversight and accountability. “Trust is still important, and you probably don’t want to relinquish actual responsibility to an AI. Humans should always remain responsible in some capacity.”
Skyfall’s first experiment is intentionally modest compared with its long-term ambition: automate as much of business operations as possible and measure not only performance but also whether human intervention steadily declines over time. Whether it succeeds remains an open question. Anthropic has already demonstrated that AI can autonomously operate a relatively constrained retail business. Skyfall is testing a far more demanding proposition: whether AI can wholly manage an organization where strategy, competition, customer behavior and market conditions constantly evolve.
