04 / EXPERIMENTS2026

Divergent Agents

I compare how agents solve the same problem.

Multiplesolution paths

Differentialtesting across outputs

Pythonbenchmark harnesses and agent tools

Different approaches

I built Divergent Agents to explore how agents approach the same problem. The Python tools and agent skills route problems through different reasoning approaches and produce several solutions to compare.

Comparing the outputs

I use differential testing to compare independently produced outputs. When the outputs disagree, I inspect the difference and use it to decide what to check next. The repository includes benchmark harnesses and recorded demos of this process.

Code and research notes

The repository contains the implementation, experiments, and research notes. This project focuses on agent tooling. My current research also looks at how predictive models keep learning as their representations change.

NEXT PROJECTTiny Cache