Introducing Workspace Agents in ChatGPT

Introduction

Artificial Intelligence is no longer just a personal productivity tool. Furthermore, on April 22, 2026, OpenAI announced a shift that changes how entire teams work together. The launch of Workspace Agents in ChatGPT marks the most significant update to the platform since custom GPTs were first introduced. Consequently, organizations can now deploy AI that runs in the cloud, handles complex multi-step workflows, integrates with the tools teams already use, and keeps working even when no one is actively watching. This guide explains what Workspace Agents are, how they work, what they can do, and why every professional from beginners to enterprise teams needs to understand this development right now.

What Are Workspace Agents in ChatGPT?

Workspace Agents in ChatGPT are shared AI agents that teams can create, publish, and use together to handle complex tasks and long-running workflows. They are powered by Codex OpenAI’s cloud-based coding and reasoning engine which gives them a persistent workspace for files, code, tools, and memory. Additionally, they run in the cloud, meaning they continue executing tasks even after the user who initiated them has signed off.

These agents represent a structural evolution from custom GPTs, which were primarily session-based and designed for individual use. Unlike their predecessors, Workspace Agents pull context from multiple systems simultaneously. They follow team-specific processes, request human approval at sensitive decision points, and hand off completed work across connected tools. As a result, they function less like chatbots and more like active team members operating within defined boundaries.

From Custom GPTs to Workspace Agents: What Changed

The transition from custom GPTs to Workspace Agents in ChatGPT reflects a deliberate strategic shift. Custom GPTs responded to individual conversational requests. Moreover, they lacked persistent memory, cross-team sharing, and the ability to take actions autonomously across enterprise systems.

Workspace Agents replace that model entirely. Specifically, the key differences are:

• Persistent operation: Agents run continuously in the cloud without requiring active user sessions.
• Team-level sharing: A single agent can be built once, shared across a workspace, and reused by multiple teams.
• Cross-platform action: Agents connect to and operate within tools like Slack, Google Drive, Salesforce, Notion, and Atlassian Rovo.
• Persistent memory: Agents maintain context across sessions, retaining notes, outputs, and prior run histories.
• Approval workflows: Agents pause and request human input at defined checkpoints before taking sensitive actions.

OpenAI has confirmed that existing custom GPTs will remain available during a transition period. Furthermore, a conversion tool will allow teams to migrate their GPTs directly into workspace agents without rebuilding from scratch.

Key Features of Workspace Agents in ChatGPT

1. Codex-Powered Cloud Execution

Every Workspace Agent in ChatGPT runs on Codex, OpenAI’s cloud-based AI execution engine. Consequently, agents are capable of writing and running code, using connected apps, remembering what they have learned, and continuing work across multiple steps all without manual intervention. They can also schedule future work and wake up on their own to continue tasks across days or weeks.

2. Multi-Tool Integrations

At launch, Workspace Agents ship with native integrations across a broad set of enterprise platforms. These include Slack, Google Workspace (Gmail, Drive, Calendar, Docs, Sheets), Microsoft apps (SharePoint, Outlook), Salesforce, Notion, and Atlassian Rovo. Additionally, teams can connect custom MCP servers for deeper integrations with proprietary systems.

The Slack integration is particularly notable. An agent can live inside a Slack channel, respond to @mentions, follow threads, and take ownership of tasks without requiring team members to switch to a separate interface. Therefore, work continues in the same environment where it already happens.

3. Workflow Builder with Natural Language

Creating a new agent requires no coding knowledge. Teams access the agent builder through the Agents tab in the ChatGPT sidebar. From there, they describe the workflow in plain language. The builder translates that description into a structured workflow with defined steps, selects appropriate tools, and prepares a testable draft agent. Moreover, teams can refine the agent through conversation or by editing workflow instructions directly.

4. Shared Workspace Directory

The Agents tab functions as a team directory, a centralized repository where agents built by anyone in the organization can be discovered and reused. As a result, workflows become organizational assets rather than individual knowledge. Teams avoid duplicating effort and can instead build on each other’s work over time.

5. Scheduling and Triggers

Agents can run on recurring schedules or respond to incoming triggers such as messages, system updates, or data changes. For instance, a metrics reporting agent can automatically pull data every Friday, generate charts, draft a narrative summary, and share the report all without any manual input. Therefore, high-frequency operational tasks become fully automated without requiring developer involvement.

6. Enterprise-Grade Controls and Governance

Security and oversight are built into the core of Workspace Agents in ChatGPT. Admins on Enterprise and Edu plans control access through role-based permissions (RBAC). They can determine who builds agents, who publishes them, and which tools and actions agents can access. Furthermore, the platform includes built-in prompt injection safeguards. The Compliance API allows admins to monitor all agent configurations, track runs, and suspend any agent immediately if needed.

Pre-Built Agent Examples Teams Can Use Today

OpenAI has published several ready-to-use agent templates that teams can deploy immediately. These include:

Software Reviewer: Reviews employee software requests, checks them against approved tools and policies, recommends next steps, and files IT tickets when needed.

Product Feedback Router: Monitors Slack, support channels, and public forums, then turns feedback into prioritized tickets and weekly product summaries.

Weekly Metrics Reporter: Pulls data every Friday, creates charts, writes the summary, and shares a report with the team automatically.

Lead Outreach Agent: Researches inbound leads, scores them against qualification criteria, drafts personalized follow-up emails, and updates the CRM.

Sales Meeting Prep Agent: Researches accounts, summarizes recent call notes from connected tools, and posts deal briefs directly into a team’s Slack room.

Each template is customizable. Additionally, teams can add their own files, skills, and memory configurations to tailor behavior for their specific workflows.

How to Build a Workspace Agent: Step-by-Step

Building a Workspace Agent in ChatGPT follows a clear, repeatable process.

Step 1 Enable Agents: The workspace admin must enable Workspace Agents for the organization and assign building and publishing permissions through role-based controls.

Step 2 Open the Agent Builder: Click Agents in the ChatGPT sidebar and select “New Agent.” The conversational builder interface opens immediately.

Step 3 Describe the Workflow: Write a plain-language description of the task. Specify what success looks like, any constraints the agent must follow, and which tools it will need.

Step 4 Connect Tools: Select approved apps and connectors. Authenticate the required integrations. The builder guides the configuration process step by step.

Step 5 Add Skills and Memory: Upload or create skills that define how the agent formats outputs. Enable memory so the agent retains context across runs.

Step 6 Test in Preview: Run the agent with realistic examples, including complex or ambiguous scenarios. Refine instructions based on observed behavior.

Step 7 Publish and Schedule: Share the agent with the team via a private link or the workspace directory. Set a recurring schedule if the workflow repeats on a predictable cadence.

Pricing and Availability

Workspace Agents in ChatGPT are currently available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. Consumer plans including Plus, Pro, and Free are excluded from the initial rollout. Workspace Agents will be free until May 6, 2026, with credit-based pricing starting on that date. Under the credit model, each agent run consumes credits proportional to task complexity, the number of tools called, and execution time.

Why This Matters for Professionals and Marketers

The arrival of Workspace Agents in ChatGPT reshapes how teams think about productivity, automation, and collaboration. For marketing professionals and freelancers, these agents handle high-frequency operational tasks from lead research to report generation that previously consumed hours of manual effort. Consequently, teams redirect that time toward higher-value creative and strategic work.

For entrepreneurs and small business owners, the ability to build shared agents without engineering support is especially significant. As a result, smaller teams can now operate with the automation capabilities that previously required dedicated development resources.

Building the Right Skill Set for the Agentic Era

Understanding agentic AI is no longer optional for professionals who want to remain competitive. Those who hold an AI Expert certification develop a structured foundation in AI systems, model behavior, and prompt design — all of which directly support effective use of tools like Workspace Agents in ChatGPT.

Moreover, as these agents become central to enterprise workflows, understanding how to design, deploy, and manage multi-step agentic pipelines becomes a core professional competency. An Agentic AI certification builds exactly that expertise — helping professionals move from passive AI users to active AI architects within their organizations.

For those operating at the intersection of AI infrastructure and enterprise strategy, a Deep tech certification provides the technical grounding needed to evaluate platforms like Codex, assess integration architectures, and contribute to governance decisions with confidence.

Finally, marketers seeking to put these tools to measurable commercial use will benefit significantly from an AI powered digital marketing expert certification. It equips professionals to connect AI-driven automation with campaign performance, audience targeting, and scalable content operations in modern digital environments.

The Broader Context: Enterprise AI in April 2026

The launch of Workspace Agents in ChatGPT did not happen in isolation. April 22, 2026 saw multiple major enterprise AI announcements land within hours of each other. This simultaneous activity signals that the enterprise AI agent market has reached a competitive inflection point one where every major technology platform is racing to become the default automation layer for teams.

What distinguishes Workspace Agents is their integration of persistent cloud execution, cross-platform connectivity, and a shared team directory into a single, accessible interface. Furthermore, their foundation in Codex, a model designed for reasoning, code execution, and multi-step task completion gives them a technical depth that simple chatbot-based automation tools cannot match.

What Comes Next

OpenAI has confirmed several upcoming additions to Workspace Agents. These include new triggers that start work automatically in response to events, better dashboards to track and optimize agent performance, expanded action support across additional business tools, and support for workspace agents within the Codex app. Consequently, the platform will grow more capable and more embedded in daily organizational workflows over the coming months.

FAQs

1. What are Workspace Agents in ChatGPT?

   They are Codex-powered, cloud-based AI agents that teams can build, share, and use together to automate complex, multi-step workflows across enterprise tools.

2. When did Workspace Agents launch?

   They launched on April 22, 2026, as part of OpenAI’s most concentrated enterprise product week to date.

3. How are Workspace Agents different from custom GPTs?

   Custom GPTs were session-based, single-user tools. Workspace Agents run continuously in the cloud, maintain memory, operate across multiple tools, and are designed for shared team use.

4. Which plans support Workspace Agents in ChatGPT?

   They are currently available for ChatGPT Business, Enterprise, Edu, and Teachers plans. Consumer plans are excluded from the initial rollout.

5. Are Workspace Agents free to use?

   Yes, during the research preview period. Free access ends on May 6, 2026, after which credit-based pricing applies.

6. What tools do Workspace Agents integrate with at launch?

   They integrate natively with Slack, Google Workspace (Gmail, Drive, Calendar, Docs, Sheets), Microsoft apps (SharePoint, Outlook), Salesforce, Notion, and Atlassian Rovo.

7. What model powers Workspace Agents?

   They are powered by Codex, OpenAI’s cloud-based coding and reasoning engine, combined with GPT-5.5 as the underlying intelligence layer.

8. Can agents run without a user being online?

   Yes. Agents run in the cloud and continue executing tasks even after the user who initiated them has signed off or is inactive.

9. How does the agent builder work?

   Users describe a workflow in natural language. The builder translates that description into structured steps, connects tools, generates skills, and prepares a testable draft agent.

10. Can agents be shared with an entire organization?

    Yes. Agents can be shared privately, via a link, or published to the workspace directory where any team member can discover and use them.

11. Do Workspace Agents have persistent memory?

    Yes. Memory is an optional feature that allows agents to retain notes, outputs, and context from previous runs across sessions.

12. Can agents run on a schedule?

    Yes. Teams can configure agents to run at recurring intervals, such as pulling and reporting weekly performance data every Friday automatically.

13. How does approval workflow work within agents?

    Agents include configurable checkpoints where they pause and request human approval before taking sensitive or irreversible actions.

14. What security controls exist for Workspace Agents?

    Enterprise admins manage access through role-based controls, the Compliance API monitors all agent runs, and the platform includes built-in prompt injection safeguards.

15. Can agents be used inside Slack?

    Yes. The Slack integration is a first-class feature. Agents can live in channels, respond to @mentions, follow threads, and take ownership of tasks without leaving Slack.

16. Will custom GPTs still work after Workspace Agents launch?

    Yes. Custom GPTs remain available during the transition period. OpenAI will also release a conversion tool to migrate GPTs into workspace agents.

17. Can non-technical users build agents?

    Yes. The conversational agent builder is designed for plain-language input, requiring no coding knowledge to create and deploy agents.

18. What pre-built agent templates are available at launch?

    Templates include a Software Reviewer, Product Feedback Router, Weekly Metrics Reporter, Lead Outreach Agent, and Sales Meeting Prep Agent.

19. How does credit-based pricing work after May 6, 2026?

    Each agent run consumes credits based on task complexity, the number of tools called, and total execution time.

20. What new capabilities are coming to Workspace Agents?

    Upcoming additions include automatic event-based triggers, improved performance dashboards, expanded tool integrations, and Workspace Agent support within the Codex app.