OpenAI brings 'workspace agents' to ChatGPT: shared AI that does work on your behalf

Workspace agents are shared, cloud-based AI that OpenAI introduced in ChatGPT on April 22, 2026. The agents do work on your behalf, handling tasks like preparing reports, writing code, and responding to messages. The feature is the successor to custom GPTs and is powered by Codex. Because they run in the cloud, they keep working even after you close your laptop. ASAP summarizes what this launch changes, based on the original announcement.

Taking over the work people used to do

Workspace agents are AI agents that take on tasks people already do at work. OpenAI said in its April 22, 2026 announcement that agents handle jobs such as preparing reports, writing code, and responding to messages. But they do not roam free. They operate only within the permissions and controls an organization sets, and some actions are designed to require human sign-off before they proceed. It is a structure that tries to hold onto autonomy and control at the same time.

Codex is the engine, the cloud is the stage

Codex is the model that powers every workspace agent, OpenAI confirmed in its launch post. The agents run in the cloud, so they keep going on long-running tasks even after a user closes the laptop. Here is where it diverges from a personal tool. A team can build an agent once, share it across the organization, use it together in ChatGPT or Slack, and improve it over time. That makes it closer to an asset a team inherits than a tool one person keeps switched on.

Describe it in plain English, and external apps snap in

Workspace agents are designed to connect directly to external apps and data sources including Slack, Google Drive, Microsoft apps, Salesforce, Notion, and Atlassian Rovo. A user opens Agents in the ChatGPT sidebar and describes the desired workflow in plain English. ChatGPT then walks through every step, from defining the process to connecting tools, setting the schedule, and choosing which actions need human sign-off. In other words, the setup is written through conversation, not code.

Four plans, and a deadline named July 6

The research preview is available on 4 plans: ChatGPT Business, Enterprise, Edu, and Teachers. The free preview period was extended to July 6, 2026, with credit-based pricing starting on that date. The feature stands as the successor to custom GPTs for enterprise workflow automation. Two signals belong together here: every plan is a business tier, and the free window has a hard nail in it.

Why it matters: from chatbot to work delegate

Workspace agents signal ChatGPT shifting from a chatbot that answers to a delegate that works. Until now AI answered questions, but it now carries real tasks through to completion within set permissions. Because a company builds an agent once and a team shares and improves it, the unit of automation moves from the prompt to the agent. A prompt evaporates, but an agent stays and accumulates, which is a different kind of thing entirely.

What Korean enterprises should read into it

For Korean organizations, the notable detail is that the connection list includes Slack, Google Drive, Microsoft, Salesforce, and Notion. Many of those names overlap with the collaboration stack teams actually run domestically. But the free window ends on July 6, 2026 and shifts to credit-based pricing, so anyone planning a pilot should first do the math on the cost switchover. The human sign-off configuration is also worth mapping against local security and compliance requirements.

Still worth caution: the preview label

This is a research preview, not a general release. The plans are limited to Business, Enterprise, Edu, and Teachers, so individual users cannot touch it right now. Codex power and always-on cloud execution are the design OpenAI described, but how stable long-running automation proves to be in real work is too early to judge from the announcement alone. How far the no-approval zone extends, and how credits are consumed, are the real variables in any adoption decision.

Source: ASAP summary of OpenAI's "Introducing workspace agents in ChatGPT" announcement (April 22, 2026; powered by Codex, always-on cloud execution, successor to custom GPTs, connections to Slack, Google Drive, Microsoft, Salesforce, Notion, and Atlassian Rovo, research preview on ChatGPT Business, Enterprise, Edu, and Teachers, free period extended to July 6, 2026 before credit-based pricing), with VentureBeat, Decrypt, and Reworked reporting.