What happens when you push AI coaching systems past their comfortable middle ground? This talk documents a series of reproducible experiments stress-testing AI reliability across persona bias, attention blind spots, and context-compression trade-offs. Running the same coaching scenarios through Claude, OpenAI, DeepSeek, and Gemini surfaces how each model diverges under pressure — and what that means for athletes who depend on these tools.
A link to the talk recording will appear here once it's available.