OpenAI introduces Dots agent for asynchronous task delegation

Three key challenges emerge in real-world use. First, vague instructions can lead to wasted effort—e.g., an agent spending hours on misaligned research. Second, defining clear completion criteria (e.g., ‘a table comparing five companies’) is critical to avoid aimless work. Third, securing access to internal tools demands careful permission management to prevent data leaks or prompt injection risks from external websites. OpenAI plans phased rollout, prioritizing higher-tier plans like ChatGPT Pro or Business, with features subject to organizational policies.