Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Code > Prompts: Why Your Agent Should Think in Code, Not English
See how agents can emit typed Python execution plans instead of English reasoning. This talk demonstrates a framework for building reliable, testable agents with live coding.
What if your agent’s “reasoning” was code, not English?
Live code walkthrough: I’ll build two agents side-by-side, a standard ReAct tool-caller and one using OpenSymbolicAI, where the LLM’s job isn’t to “reason” in English, it’s to emit typed execution plans as Python. Same model, same tools, same question. You’ll watch the ReAct agent burn tokens re-reading its own verbose reasoning while the behavioral agent plans once and executes deterministically. Then I’ll add 17 lines of code live and turn a failing agent into one that wins the race. No slides.