Why Your Company Should Own Its AI Model | E2278

Key Concepts

• Company World Model: An AI architecture where a company’s proprietary data is integrated into the weights of a custom model, allowing for autonomous, context-aware decision-making.
• Small Language Models (SLMs): Compact, specialized AI models that can run on local hardware (e.g., workstations) and are trained on specific company data.
• Reinforcement Learning (RL): A machine learning methodology used to update model weights based on interactions and feedback, enabling continuous learning.
• Agentic Computing: AI systems capable of executing multi-step tasks (e.g., paying invoices, drafting contracts) by interacting with external tools and APIs.
• Part 108 (Aviation): A regulatory framework for drone operations that enables "Beyond Visual Line of Sight" (BVLOS) flights, allowing for remote, autonomous logistics at scale.
• Post-training: The process of refining a base model (like Kimmy) with proprietary data to create a specialized, company-owned intelligence asset.

1. Aragon: Enterprise AI Operating System

Josh Cerot, CEO of Aragon, presented a vision for companies to "own their intelligence."

• Core Methodology: Instead of relying solely on expensive, closed-source frontier models, Aragon uses an RL-based approach to embed a company’s historical data (Slack, email, CRM) directly into the model's weights.
• Efficiency: By moving data into the weights, the system reduces token usage by up to 100x and increases execution speed by 10x compared to traditional RAG (Retrieval-Augmented Generation) methods.
• Continuous Learning: The model updates itself overnight based on daily interactions, ensuring the "Company World Model" remains current without manual retraining.
• Business Model: Aragon charges $5 per million tokens, positioning itself as a cost-effective alternative to frontier models like Anthropic’s Opus.

2. Iona: Autonomous Logistics

Etien Lu, CEO of Iona, discussed the "physical internet"—a drone-based logistics network designed for light cargo.

• Technical Specs: Iona drones are hybrid tilt-rotors capable of carrying up to 44 lbs over 60–80 miles.
• Regulatory Shift: The transition from Part 107 (line-of-sight) to Part 108 (BVLOS) is described as the "Level 4 autonomy" equivalent for drones, allowing for remote supervision of multiple aircraft from a single control center.
• Real-World Application: Iona is currently partnering with logistics giants like CMA CGM and local operators in Ireland to solve the "last-mile" problem in rural and remote areas where traditional van-based logistics are inefficient.

3. AI in Creative Industries

The discussion highlighted a shift in film production, specifically regarding the $70 million Bitcoin-themed film starring Gal Gadot.

• Production Methodology: The film is being shot entirely on a green-screen soundstage, with AI generating all backgrounds and lighting.
• Economic Impact: This approach significantly lowers production costs, allowing for more creative projects to be greenlit. The consensus is that AI acts as a tool to augment human performance rather than replace it, similar to how word processing replaced the typing pool.

4. Notable Quotes

• Josh Cerot (Aragon): "The only way to actually [own your intelligence] is to train a model on your proprietary data and then give you the weights... It’s yours."
• Jason Calacanis: "We’re on a bit of a collision course... Small language models can be produced and run on local hardware, and if they have all your data in it, well, what becomes the difference?"
• Etien Lu (Iona): "A drone is like a vehicle with wheels between a motorcycle and a 44-ton truck... We’re like the new delivery van in the sky."

5. Synthesis and Conclusion

The episode underscores a major shift toward sovereign AI. Whether it is Aragon’s "Company World Model" or Iona’s autonomous logistics, the common theme is the move away from generic, centralized AI toward specialized, locally-controlled, and agentic systems. By leveraging RL for continuous learning and taking advantage of new regulatory frameworks like Part 108, startups are successfully building "secret sauce" infrastructure that is faster, cheaper, and more proprietary than the current generation of general-purpose AI tools. The future of enterprise and logistics lies in these recursive, self-improving loops that operate on local hardware, effectively turning data into a permanent, actionable asset.