agentsproduct-launchenterprise ChatGPT Work promises hours-long autonomy. Here's what breaks first. OpenAI's ChatGPT Work promises hours-long task completion, but the real failure mode of long-horizon agents is error accumulation and irreversible actions.