“OpenAI's early agent strategy was flawed because it focused on cloud-based deployments orchestrated from ChatGPT, rather than local execution on a user's machine.”