Building GenAI Systems That Make Business Decisions with Thomas Wiecki (PyMC Labs)

​More teams are using LLMs as decision tools, not just content generators.

​In this episode, we look at two real-world systems that go beyond prompting: one used by Colgate to simulate consumer reactions to new product ideas (like mango-flavored toothpaste), and another agent that automates complex modeling workflows to guide media spend decisions. We talk with Thomas Wiecki (Founder of PyMC Labs and co-author to PyMC) about what it takes to build GenAI systems that hold up in production… and what these examples tell us about where generation ends and real structure begins.

​This isn’t a demo reel: we’ll talk about how these systems actually perform, where they break, and what it takes to combine LLMs with probabilistic modeling in production.

​We’ll discuss:

​🧪 How LLM-generated survey responses performed with 90% accuracy, even when sliced across demographics
🔁 What it takes to build a closed-loop system where GenAI generates and critiques product and ad ideas
🤖 How PyMC Labs built a structured agent to automate the analytics behind how companies decide where to advertise
🧭 What these systems reveal about where GenAI excels—and where statistical structure and validation still matter

​Whether you're experimenting with agent frameworks, evaluating LLM outputs, or building GenAI systems beyond the interface layer, this conversation will sharpen your sense of where GenAI is actually useful—and how to make it trustworthy.