OpenAI rolls out team agents, plugin
- OpenAI said on April 22 it added workspace agents to ChatGPT, letting teams build shared, cloud-run agents for multi-step work across apps. - The company also published a Codex plugin for Anthropic’s Claude Code, so developers can call Codex inside Claude Code for reviews or handoffs. - Both releases extend OpenAI’s push from solo chatbots toward governed team workflows and cross-tool automation. (openai.com)
OpenAI on April 22 added workspace agents to ChatGPT, a new shared agent system for teams, and it has also published a Codex plugin for Anthropic’s Claude Code. (openai.com) (community.openai.com) Workspace agents are Codex-powered agents that run in the cloud, can keep working when a user is offline, and are meant to be shared inside an organization. OpenAI said they operate within company permissions and controls. (openai.com) OpenAI said teams can use the agents in ChatGPT or Slack, and the company framed them as an evolution of GPTs rather than a replacement. The feature is in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. (openai.com) The company’s examples are office workflows, not one-off prompts: a software review agent that routes approvals, a weekly metrics agent that drafts reports, and a lead outreach agent that updates a customer relationship management system. (openai.com) OpenAI’s developer cookbook shows the model more concretely. One sample agent checks the next day’s sales meetings, pulls account context from Microsoft SharePoint, searches the web for attendee and company news, writes a briefing document, and sends a summary on a daily schedule. (developers.openai.com) That setup also shows the governance layer OpenAI is selling to companies. The cookbook says workspace owners can enable agents, assign access by role-based access control, and decide who can build or publish them. (developers.openai.com) The Claude Code plugin is a separate release aimed at developers who already use Anthropic’s coding agent. OpenAI said the plugin lets Claude Code users invoke Codex for a standard review, an adversarial review, or a delegated background task. (community.openai.com) (github.com) OpenAI said the plugin is not a separate runtime. It delegates through the local Codex command-line interface and app server, using the same authentication, configuration, environment, and Model Context Protocol setup a developer already uses with Codex. (community.openai.com) The plugin’s GitHub repository lists commands including `/codex:review`, `/codex:adversarial-review`, `/codex:rescue`, `/codex:status`, `/codex:result`, and `/codex:cancel`. The README says it requires either a ChatGPT subscription, including Free, or an OpenAI application programming interface key. (github.com) Those two launches land days after OpenAI expanded Codex itself on April 16. The company said more than 3 million developers use Codex weekly, and that update added computer use, memory, browser access, scheduling, and more than 90 additional plugins. (openai.com) Anthropic describes Claude Code as an agentic coding system that reads a codebase, makes changes across files, runs tests, and delivers committed code. OpenAI’s plugin effectively turns that rival environment into another surface for Codex. (anthropic.com) (github.com) Taken together, the releases show OpenAI trying to own two layers at once: shared workplace automation inside ChatGPT and developer workflows that can reach beyond OpenAI’s own interface. (openai.com 1) (openai.com 2)