Free Lesson
Build Your First Simulation Environment for Evals & RL
60 min
Sep 4, 2026 12:00 PM
What you'll learn
Understand What an AI Simulation Environment Is
See how tasks, tools, state, and scoring become a reusable world for evals and reinforcement learning.
Learn from Bubbles, Tal’s Claude-Powered Robot
Use its spatial failures and self-written memories to see why agents need explicit tasks, traces, and grading.
Build a Simulation Environment Live
Turn a real task into tools, test cases, and a grader you can reuse for evals or RL.
Why this topic matters
Better models alone don’t produce reliable agents—the environment determines what you can observe, evaluate, train, and trust. Aman and Tal will unpack tasks, tools, state, and rewards through Bubbles, the PiCar-X robot Tal gave to Claude. They’ll use its spatial-reasoning failures and self-written memories to show why evals need repeatable simulations, then build one live.
