After Google and Microsoft unveiled their enterprise-grade AI agent platforms, OpenAI also revealed its trump card. OpenAI officially launched its platform on ChatGPT.The all-new "Workspace Agents"This new feature, powerfully driven by the Codex model, breaks the past limitation that AI could only "passively answer questions," giving AI an independent sandbox in the cloud, persistent memory, and the ability to perform tasks across tools 24 hours a day.
With the official launch of these "digital workers" capable of automating sales, IT audits, and data reporting, the once-popular...Customizable assistants "GPTs"This also signals the start of the countdown to retirement.
Introducing workspace agents in ChatGPT—shared agents that can handle complex tasks and long-running workflows across tools and teams. pic.twitter.com/eHplfXCWlk
- OpenAI (@OpenAI) April 22, 2026
A paradigm shift from "passive chatting" to "active execution"
Since its launch at the end of 2023, GPTs has lowered the threshold for customized AI, but its core essence is still an "advanced chat box" - as long as the window is closed, all context and progress will be reset to zero, and the task cannot be actively advanced.
The newly launched Workspace Agents completely rewrite this logic. Through the underlying empowerment of the Codex model, each AI agent obtains a dedicated cloud sandbox environment, meaning it can store files, run code, and call external APIs and applications.
More importantly, these AI agents possess the ability to "remember" and "perform scheduled tasks." Even when employees leave work and shut down their devices, the AI agents can continue to monitor Slack channels or receive new requests in the cloud. At the same time, through human correction during daily use by the team, these AI agents can continuously evolve, turning the "best practices" of the most senior employees within the enterprise into standardized workflows that can be used by the entire team.
Five practical scenarios: Directly taking over tedious business operations
To demonstrate the disruptive power of Workspace Agents, OpenAI released five real-world internal application scenarios covering different departments:
• Spark (Following up on clues):For sales teams, it can automatically conduct online background checks on new potential customers, score them according to standards, and if it determines that they are worth following up, it will automatically draft and send out outreach emails via Gmail and create reminders on Google Calendar.
• Scout (Product Feedback Routing):It can monitor Slack, customer support, and public forums simultaneously. It automatically categorizes and prioritizes scattered user feedback and creates structured tickets in the Linear system, attaching a complete chain of evidence.
• Slate (Software Review):It automatically reviews employees' software usage applications for the IT department, verifies the company's security policies, and recommends the next steps in the workflow, significantly shortening the approval process.
• Tally (Indicator Report):Specializing in data analysis, this user can read complex business data from Google Sheets (such as annual recurring revenue (ARR) and monthly active users (MAU), automatically calculate week-on-week changes, generate visual charts, and write executive summaries.
• Trove (Third-Party Risk Management):It replaces the tedious due diligence of traditional compliance consulting firms. Based on internal standards, it automatically conducts online investigations into the financial and reputational risks of suppliers and generates a complete Google Doc assessment report.
Zero-code barrier and enterprise-level permission governance
Despite their powerful features, these AI agents are incredibly easy to set up. Users simply describe their needs in natural language within ChatGPT (e.g., "Build an AI agent for supplier due diligence"), and the AI will automatically handle the rest of the setup process: creating a checklist, attaching the necessary tools, writing role instructions and output formats—all without requiring any code writing.
Regarding the most critical aspects for enterprises—data security and access control—administrators can finely configure access permissions for each group through RBAC (Role-Based Access Control). For sensitive actions such as sending external emails or modifying important forms, a mandatory "pop-up manual approval" request can be set. The system also provides comprehensive log tracking and a Compliance API to ensure that all AI agent actions are under the enterprise's supervision.
This feature is currently available through subscription plans such as ChatGPT Business, Enterprise, Edu, and Teachers. OpenAI has adopted a rather aggressive pricing strategy: offering a completely free trial until May 6, 2026, after which it will switch to a points-based billing system, attempting to deeply integrate the AI agent into the workflows of various enterprises during the free period.
Analysis of viewpoints
With the release of OpenAI Workspace Agents, the competitive landscape of "enterprise AI" in the second half of 2026 has been fully established.
In this battle, Microsoft's Copilot Agents have an absolute home-field advantage due to their deep integration with Microsoft 365 (Word, Excel, Teams); while Google's earlier announcement...Gemini Enterprise Agent PlatformThe main focus is on the semantic integration of the full-end control tower capable of scheduling more than 200 models with Workspace.
Faced with these two giants who possess enterprise infrastructure, OpenAI's approach to breaking the deadlock is quite direct: drive usage scale with 9 million weekly active users.
The biggest hidden cost for companies implementing a new system is often "employee education and training." OpenAI is betting that since the vast majority of white-collar workers are already used to using ChatGPT every day, integrating the functionality directly into the ChatGPT sidebar's Workspace Agents will be the solution with the lowest learning cost and the highest penetration rate.
This also represents a fundamental shift in work philosophy—AI's role has officially transitioned from "assisting in work" to "replacing work." When this tireless, non-resigning digital worker begins to take over reports and due diligence, the human resource structure and management model of modern enterprises will inevitably undergo a violent upheaval that cannot be avoided.
