Workspace agents roll out

OpenAI began rolling out workspace agents in ChatGPT on April 22, giving teams shared AI helpers that can run multi-step jobs across workplace tools. (openai.com) OpenAI said the agents are powered by Codex, run in the cloud, and can keep working when a user is offline. The company said teams can use them inside ChatGPT or Slack and set them up to gather context, follow approval steps, and move work across systems. (openai.com) The product is in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. OpenAI said workspace agents are an evolution of GPTs, and that GPTs will stay available while customers test the new system and prepare to convert existing tools. (openai.com) A workplace agent is meant to act less like a chatbot and more like a shared team process: one setup, reused by multiple people, with the company’s permissions applied. OpenAI’s examples included agents for software review, product-feedback routing, weekly metrics reports, lead outreach, and third-party risk checks. (openai.com) Anthropic released a separate report on April 22 based on a survey of 81,000 Claude users about work, productivity, and fears about AI. The company said people in jobs more exposed to AI reported more concern about job displacement, especially early-career respondents. (anthropic.com) Anthropic said users in the highest-paid and lowest-paid occupations reported the biggest productivity gains, most often because AI let them take on new tasks rather than only finish old ones faster. The report also said respondents who saw the biggest speed gains were more likely to worry about losing work to AI. (anthropic.com) The two releases landed a month after Anthropic expanded its Economic Index, a public tracker showing how Claude is used across U.S. states, countries, industries, and occupations. Anthropic says that index is built to show whether people are using Claude to collaborate on work or delegate work to it. (anthropic.com) OpenAI’s launch also builds on its earlier Codex rollout in May 2025, when the company introduced a cloud coding agent that could work on tasks in parallel inside isolated environments. In that release, OpenAI said Codex could write features, fix bugs, run tests, and propose pull requests. (openai.com) Together, the announcements show the two companies pushing on different parts of the same workplace question: OpenAI is shipping tools that automate team workflows, while Anthropic is publishing data on how workers say those tools are changing their jobs. (openai.com) (anthropic.com)