OpenAI Launches Workspace Agents in ChatGPT — Persistent AI Workers for the Whole Team
OpenAI introduced Workspace Agents in ChatGPT on April 22, 2026 — shared, team-level agents powered by Codex that run in the cloud and execute long-running tasks even when users aren't active. Teams build an agent once and deploy it across ChatGPT or Slack on a schedule. Free in research preview until May 6 for Business, Enterprise, Edu, and Teachers plans.
Original sourceOpenAI made its most significant move into enterprise agentic AI yet on April 22, 2026, launching Workspace Agents in ChatGPT — persistent, team-level AI workers that run in the cloud and can execute tasks autonomously even when no human is actively in the chat window.
Workspace Agents are powered by Codex, OpenAI's AI software engineering system, and are designed for long-horizon tasks that previously required either constant user supervision or custom API integrations: drafting and filing weekly reports, writing and reviewing code, responding to routine Slack messages, aggregating data from multiple sources, and running scheduled analysis pipelines. Teams configure an agent once and it becomes a shared resource accessible via ChatGPT's interface or directly embedded into Slack workflows.
The key differentiators from existing ChatGPT are the *shared* nature (the whole team works with the same configured agent, not individual sessions), the *scheduled execution* (agents run at specified times without a human kicking them off), and the *persistent state* (the agent can be instructed to remember context across runs). The Slack integration means Workspace Agents can surface their work in existing team communication channels rather than requiring everyone to visit ChatGPT separately.
OpenAI is offering the feature in research preview at no cost until May 6, 2026 for Business, Enterprise, Education, and Teachers plan subscribers — after which it will move to a credit-based consumption model. The tight preview window is almost certainly designed to build adoption before the pricing switch. Organizations wanting to test team-level agentic workflows have two weeks to do so at no additional cost.
The timing is notable: Zed Editor launched its own Parallel Agents feature on the same day, and Google's Cloud Next announcements earlier this month introduced enterprise agent orchestration capabilities. The week of April 22 is shaping up as the moment the industry collectively shifted from single-instance AI assistants to orchestrated, multi-agent execution environments.
Panel Takes
The Builder
Developer Perspective
“Scheduled execution + Slack integration means Workspace Agents slot into existing engineering workflows without requiring teams to change behavior. The May 6 credit cutover is tight — if you want to assess real usage patterns before committing, start your evaluation this week. The Codex backbone raises the ceiling for what these agents can autonomously execute on the technical side.”
The Skeptic
Reality Check
“'Research preview' with a hard credit-switch deadline is a classic enterprise adoption tactic — generous access builds dependency before pricing kicks in. The credit model for scheduled agents running autonomously could result in unpredictable costs at scale if agents are triggered frequently or run longer than expected. Organizations need governance policies around agent scheduling before this goes into production.”
The Futurist
Big Picture
“Persistent, scheduled, shared agents in the cloud represent the organizational primitive that makes AI feel less like a tool and more like a team member. The moment an AI can run on a schedule and report back into Slack without human initiation, the organizational model for knowledge work starts to fundamentally shift. OpenAI is positioning itself as the infrastructure layer for that shift.”