Former Microsoft AI executive splashed $1 million to buy a real company, testing whether an “AI acting as CEO” could fully autonomously run the business

Can AI really replace human managers? According to a report by Forbes, Skyfall AI, an AI startup founded by a former core member of the Microsoft AI team, is planning to spend $1 million to acquire a real B2B SaaS or e-commerce company, boldly installing “AI as CEO” to run the business independently end to end. This high-profile and costly experiment aims to prove that AI has the ability to make autonomous decisions and learn continuously in dynamic markets, attempting to break the current limitation that mainstream AI can only serve as an “autonomous employee.”
(Background: Binance CZ on AI and Bitcoin: AI can’t beat inflation, and BTC is the “hardest asset” to preserve wealth)
(Additional context: Taking on Nvidia’s dominance! AMD launches the “Helios” rack-mounted AI system, and Microsoft Azure announces it will roll out first)

Table of contents

Toggle

• Led by former Microsoft AI veterans, blasts mainstream LLM direction as wrong
• Launches “enterprise world model,” using real markets to break the simulation bottleneck
• Takes on Anthropic’s vending machine experiment! Aims to double revenue and keep everything public

While Silicon Valley giants are still debating how AI can help employees improve efficiency, a group of top AI scientists has already decided to skip that stage and put AI in the driver’s seat. On July 20, 2026 (Taipei time), according to Forbes, Skyfall AI—newly founded by former Microsoft (Microsoft) AI team core members—announced a highly ambitious plan: the company expects to invest as much as $1 million to acquire a small B2B SaaS or e-commerce company, and have AI independently serve as that company’s Chief Executive Officer (CEO), conducting end-to-end comprehensive management.

Led by former Microsoft AI veterans, blasts mainstream LLM direction as wrong

The power behind this expensive experiment is quite prominent. Skyfall AI co-founders Sam Pasupalak (CEO) and Kaheer Suleman (CTO) met nearly 20 years ago while studying computer science at the University of Waterloo, and founded the conversational AI system Maluuba in 2011. The company was acquired by Microsoft in 2017 for as much as $160 million, becoming a core foundation of Microsoft Canada’s AI research lab.

Now, these two AI veterans have launched sharp criticism of the industry direction currently led by OpenAI, Anthropic, and Google. They believe that focusing on building an “autonomous employee” approach to automate a single job is wrong. Skyfall points out that, for now, agentic AI has reached its limit in enterprise environments—especially in “continual learning.” Once the model faces a competitor’s new product, changes in customer behavior, or shifts in market dynamics, its performance can drop dramatically. Their “Morpheus” benchmark test shows that even top-tier models like GPT-5.5 or Gemini cannot effectively adapt to dynamic enterprise environments.

Launches “enterprise world model,” using real markets to break the simulation bottleneck

To address this pain point, Skyfall abandoned the traditional prompt-driven AI agents approach and instead built “Enterprise World Models” and “latent world models.” This technology allows AI to simulate in its mind the chain-of-effects its decisions would produce across the entire organization.

Although they previously succeeded in testing AI running an entire virtual theme park in the classic game RollerCoaster Tycoon, the team emphasizes that even a simulation environment that is more realistic still cannot fully capture the uncertainty of real customers, competitors, and markets (the sim-to-real problem). Pasupalak is unequivocal: “If you want a truly autonomous enterprise, you must buy a company and run it end to end. Unless you run it with the minimum amount of human involvement, you can never know whether an autonomous enterprise is actually feasible.”

Takes on Anthropic’s vending machine experiment! Aims to double revenue and keep everything public

In the field of AI experiments, this is not the first time humans have tried to delegate power. Previously, Anthropic launched the “Project Vend” program, letting Claude agents autonomously run an office vending machine business. Although at first employees were convinced to buy strange items and it lost money, later it became stable. Meanwhile, NetDragon’s virtual executive “Tang Yu” or Dictador’s robotic CEO “Mika” in Poland still essentially lean toward supporting decision-making. Skyfall’s uniqueness is that they buy a real revenue-generating company and publicly track on the internet both the “degree to which human intervention is reduced” and “revenue growth,” aiming to double the revenue of the acquired companies.

However, the founder also clarified that this experiment does not mean AI will be able to fully replace human leaders in the future. Pasupalak explains that AI can handle most operational work—such as meetings, budgets, team coordination, and marketing activities—so humans can focus on big-picture strategy; but “motivating employees” and “building deep interpersonal relationships” are still difficult for AI to replicate. CTO Kaheer Suleman also emphasized that humans must still play a role in oversight and accountability within the system, and ultimately trust and responsibility cannot be fully handed over to machines. This bold social experiment moving from “tools” to “autonomous enterprises” will undoubtedly offer disruptive insights for future business models.