OpenAI is trying to bring long-running AI agents beyond software development with ChatGPT Work, a product released last month on its $20 subscription tier. It is designed to complete multistep office projects by connecting to tools such as email, Slack, Notion and Figma rather than only answering questions.
The wider access makes the agent more useful, but also gives it more opportunities to expose private information or take an unwanted action. An OpenAI desktop engineer told TechCrunch that his test setup can control much of his digital life and acknowledged the possibility that it could pull information from a private message into a document.
Adoption outside coding remains the core challenge. An OpenAI-backed study found that 98% of company employees used Codex in June, compared with 17% of organizational subscribers and less than 1% of individual subscribers. ChatGPT Work adds guidance and a friendlier interface for non-engineers, but convenience does not solve permission design. Organizations still need limited connectors, visible action previews and records showing which data an agent used.