OpenAI workspace agents

OpenAI has started rolling out workspace agents in ChatGPT, letting companies build shared bots that can keep working across Slack and other business apps after employees log off. (openai.com) OpenAI said on April 22 that the agents are powered by Codex, run in the cloud, and are aimed at “complex tasks and long-running workflows” such as reports, code, inbox replies, lead qualification, and risk reviews. The company said teams can use one agent together in ChatGPT or Slack and improve it over time. (openai.com) The launch is a research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. OpenAI’s help docs say admins can keep agents off by default, test them before publishing, share them privately or across a workspace, and schedule recurring runs; Enterprise workspaces with enterprise key management were excluded at launch. (openai.com) (help.openai.com 1) (help.openai.com 2)) The product shifts OpenAI’s workplace pitch from one-off chat assistants to software that can pull context from company systems, follow a process, ask for approval, and move work between tools. OpenAI called workspace agents “an evolution of GPTs” and said it will add a way to convert existing custom GPTs into agents. (openai.com) That change lands after OpenAI spent months tightening controls around shared workplace AI. The same release notes say eligible workspaces can connect agents to Google Drive, Google Calendar, Slack, and SharePoint, add files and custom Model Context Protocol servers, and view version history and analytics. (help.openai.com 1) (help.openai.com 2)) OpenAI paired the agent launch with a new image model inside ChatGPT. ChatGPT Images 2.0, announced April 21, can generate multiple images at once, render non-Latin text more accurately, use web information when paired with thinking models, and keep characters and layouts more consistent across sets of images. (openai.com) (pcmag.com)) PCMag reported that Images 2.0 can produce up to eight images in one request and is available to ChatGPT and Codex users, while advanced “thinking” outputs are limited to Plus, Pro, and Business plans. OpenAI described the model as improving text rendering, multilingual support, and visual reasoning. (pcmag.com) (openai.com)) The broader backdrop is that OpenAI’s product lines have been moving closer together: tools that first showed up for developers are increasingly appearing inside ChatGPT for office workers. OpenAI’s Business release notes said last week that GPT-4.1 had been added directly to ChatGPT, even as separate help pages now say GPT-4.1 was retired from ChatGPT on February 13, 2026 and remains available through the application programming interface, leaving the company’s model lineup in flux. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3)) For customers, the immediate change is simpler than the product map: ChatGPT is becoming a place to assign work, not just ask questions. OpenAI’s latest releases all point to the same use case inside companies — agents that can act, and models that can produce text, code, and images in the same workspace. (openai.com) (openai.com) (help.openai.com))