Behavioral Divergence Evals
The art of making agents more different, and giving your human brain a more effective way at managing the subagent swarms.
Open a browser, and you are running the experiment, not watching one. You write a prompt, it goes to several agents at once, and a live score ranks how far apart their answers land, so the round goes to whoever pulls them furthest apart. The room grades every answer pass or fail until we stop agreeing with each other, and you leave with the software to run your own.
About the speaker:
Kai Ase Siren has spent more than ten years building developer infrastructure, from language tooling through multi-cloud platforms to government systems. She now works on the layer underneath agentic systems: what an agent is permitted to run, how its context gets composed, and how you prove afterward that it did what it claimed. Much of that work is open source under the coilyco GitHub org.
Speaker links