02/16/2026
Some of you may be wondering what this IGLOO thing we are making is so I wanted to try summarize it in a way that is understandable.
Most AI systems today are powerful, but unpredictable. (Aka They hallucinate.) A Large Language Model (LLMs-like ChatGPT or Claude or Gemini) is basically a very advanced word-prediction machine. It was trained by reading huge amounts of text so it can learn patterns in language. When you type something, it doesn’t “think” or “know” facts the way a person does…it predicts the most likely next word, then the next one, very quickly, until it forms a full answer. It sounds smart because it’s extremely good at recognizing patterns, not because it truly understands what it’s saying. It’s probabilistic.
If you ask the same question twice you might get different answers. (Like the prompt that has been going around LinkedIn about walking or driving to the car wash that’s 100 feet away.)
IGLOO is built differently. It’s a deterministic AI runtime, which means:
• Same input → same result
• Every decision can be replayed
• No hidden self-modifying behavior
• Built for regulated, high-stakes environments
Think of it like this:
Most AI is very smart…but improvises.
IGLOO is smart, and keeps receipts. And runs on edge so it doesn’t use tons of compute that destroy the environment like LLMs do.
This isn’t another LLM wrapper. We have built the infrastructure that makes AI safe, auditable, and deployable where accountability matters-like fintech, aviation, pharma. It has been built from first principles as an operating system. We have a working v4 prototype. You can stop by the office if you would like to see it in action! Alex is the technical architect but Sam also contributed ideas in her perspective.
