Fun experiment with AI Agents - lets see how they do when we ask them to collaborate with each other 🤣. The answer in this somewhat technical paper is ‘not very well’, which turns out to be a primer to questions we humans might find useful - how to train (both humans and AI) for collaborative intelligence. Extrapolating these agents to human collaboration and it would appear that they succeed and fail for similar reasons - not because of an inherent deficiency, but through problems in communication and context awareness. Lots to ponder, especially as I suspect we’re going to see AI Agents appear quite frequently in our workflows in 2026…