The Contract Clerk for an AI Steering Committee

Case: How to get reliable answers about contracts

Building Trustworthy AI_ A Design Blueprint

14:28

Your take away: Many AI projects fail because we ask language models to calculate, find data, and apply logic. Use one AI for the human side of things, and another with tools enabled to handle data and logic.

Last week, I gave feedback to an AI steering group working on their first lighthouse project. And we hit a classic problem. They were asking a language model to solve a task that requires rules, logic, and precision.

Put differently, they were asking someone with a language degree from uni to do complex math.

Language models are amazing at conversation. But they’re not great at counting days. They guess. And when you’re dealing with contracts, guessing is not an option.

So I built a smarter setup. One model talks to the user. A second model, with Python activated, does the math behind the scenes. The result is a precise, safe system that finds the correct termination date every time.