OpenAI shifts to workspace agents

OpenAI has replaced Custom GPTs for team workspaces with Codex-powered “workspace agents” that can run shared tasks in ChatGPT and Slack. (openai.com) OpenAI announced the change on April 22, 2026, saying workspace agents are shared agents for organizations that can handle “complex tasks and long-running workflows” while operating under company permissions and controls. The company said they run in the cloud, so they can keep working after a user closes the app. (openai.com) On OpenAI’s business product pages, the agents are pitched for multi-step work across Slack, Google Drive, Microsoft SharePoint and other enterprise tools, with governance and admin controls layered on top. OpenAI’s ChatGPT Business help pages also describe Business as a shared workspace with centralized billing, admin controls, usage visibility and separate ChatGPT and Codex seat types. (openai.com, help.openai.com) The shift lands as OpenAI keeps changing which models people in ChatGPT can actually use, and what happens when they hit limits. On April 9, 2026, OpenAI said GPT-5.3 Instant Mini became the fallback model after users hit rate limits for GPT-5.3 Instant, and that fallback does not appear in the model picker. (help.openai.com) That same fallback change appears in OpenAI’s Enterprise and Education release notes, showing the routing logic is not just a consumer issue. OpenAI’s model notes also say GPT-5.4 mini is now used as a fallback for GPT-5.4 Thinking for paid users, while enterprise customers can choose GPT-5.4 mini as the default for Auto routing. (help.openai.com, help.openai.com) OpenAI has also been pruning older options. Its GPT-5.3 and GPT-5.4 help article says GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and Thinking were retired from ChatGPT on February 13, 2026, with Business, Enterprise and Education customers keeping GPT-4o inside Custom GPTs only until April 3, 2026. (help.openai.com) That timeline helps explain why workspace agents matter now: OpenAI is moving teams away from static, user-built GPTs and toward managed agents tied to seats, permissions and connectors. In the same April 2, 2026 Business release notes, OpenAI was still shipping admin controls for enterprise connectors such as Outlook actions, including warnings that workspace owners should review permissions before enabling them. (help.openai.com) The product stack around those agents is still uneven across plans. OpenAI’s help center says ChatGPT Business is self-serve, while some workspace-agent features are sold through “Contact sales,” and enterprise-focused connectors and controls continue to roll out through release notes rather than as a single finished package. (help.openai.com, openai.com, help.openai.com) For customers, the practical change is that “which model am I using” is no longer a simple menu choice. In OpenAI’s current ChatGPT products, agents, fallbacks, auto-routing, seat types and connector permissions are now part of the same workplace software problem: keeping the system available, affordable and inside company rules. (help.openai.com, help.openai.com, openai.com)