Skyfall AI Bets on World Models to Automate Corporate Operations
AI-generated image · Bay Street Wire
Co-founders of former Microsoft-acquired Maluuba aim to prove that simulating physical reality outperforms LLMs in high-stakes business decision-making.
Former Maluuba co-founders Sam Pasupalak, Sumit Pasupalak, and Kaheer Suleman have launched Skyfall AI, a Toronto- and San Francisco-based venture focused on developing AI world models. Unlike large language models (LLMs) that rely on text, world models are trained on spatial, physical, and movement data to create internal 3D representations of environments, which proponents argue enables the integration of cause and effect rather than simple pattern matching, BetaKit reports.
Skyfall aims to demonstrate "enterprise super intelligence" by automating a company's operational functions—including product, marketing, and finance—with minimal to no human intervention. To test this, Skyfall announced plans to acquire a "micro-B2B-SaaS" business for up to $1 million USD, intending to install an AI CEO and double the entity's revenue within six months. According to BetaKit, Sam Pasupalak believes world models are superior to LLMs for high-stakes decision-making and long-horizon planning.
Backed by Garage Capital and Inovia Capital, Skyfall operates as a for-profit, research-first organization. Inovia Capital partner Karamdeep Nijjar noted on LinkedIn that Pasupalak is focused on real-world applicability over static benchmarks. While the goal is to eliminate monotonous operational tasks, Pasupalak acknowledged to BetaKit that certain human skills, specifically managing business relationships, remain beyond the reach of world models.
The venture's approach to labor is stark: Pasupalak stated the first acquisition target will be owner-operated or employ one to five contractors/employees, who will receive support and "generous departure package[s]" as the AI takes over.