With 'Workspace Agents,' OpenAI is Turning ChatGPT Into A Team Of Always-On Agents

The intense competition in the large language model space, often described as the LLM war, continues as the technology evolves.

Erupted in late 2022 when OpenAI launched ChatGPT, what began as a conversational demo quickly captured global attention, demonstrating how generative AI could produce coherent, useful responses at scale and sparking widespread adoption across industries.

Within months, rivals including Google, Anthropic, Meta, and Microsoft accelerated their own releases, turning the field into a high-stakes race for capabilities, user bases, and enterprise integration.

ChatGPT’s arrival marked a pivotal shift from experimental AI research to practical tools embedded in daily workflows, setting the stage for successive waves of innovation focused not just on smarter chat but on autonomous, collaborative systems.

Now, three-year-old ChatGPT gets a big update: 'workspace agents for ChatGPT.'

The feature is designed to, extend the platform deeper into organizational environments.

These shared agents are designed for teams and enterprises handling complex, multi-step tasks and long-running workflows that span tools, data sources, and people.

Unlike earlier custom GPTs that operated mainly within individual conversations, workspace agents emphasize persistence, collaboration, and integration across business systems. They run in the cloud, continue operating when users are offline, and can function through both the ChatGPT interface and Slack.Building an agent starts with a natural language description of its intended role or by uploading a relevant file.

ChatGPT then assists in outlining steps, connecting approved tools, incorporating company-specific practices, and testing initial performance.

Once deployed, the agent becomes a reusable asset that team members can share, reducing the need to repeatedly explain standard procedures.

It supports handoffs between people, maintains shared context, and executes approved actions without requiring constant human oversight.

Practical examples highlight the scope. An agent might intake software requests, verify compliance with internal policies, route them for approvals, and automatically create tickets in project management systems with recommended next steps.

Another could scan feedback from Slack channels, support tickets, and external sources, then prioritize issues and compile weekly summaries for product teams.

Sales agents can qualify leads by synthesizing call notes and account research, draft follow-up emails, and update CRM records.

Risk or compliance agents screen vendors and generate assessment reports. In Slack, agents join discussions, pull relevant context from connected systems, suggest solutions, and log outcomes elsewhere.

Central to the design is integration with workplace tools, including files, emails, chats, code repositories, calendars, and drives, all within the organization’s existing permission boundaries. Agents can update issues in tools like Linear, create documents, send messages, or schedule events. They support scheduled runs and background execution, allowing processes to advance independently.

To address governance needs, the system includes approval gates for sensitive actions, activity logs, and administrative controls for Enterprise and education users to manage permissions and tool access.

The feature is currently available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans.

Previous custom GPTs remain usable, with a planned migration path ahead.

OpenAI is offering free usage until May 6, 2026, after which a credit-based model applies.

The agents leverage the company’s Codex model family and incorporate safeguards such as protections against prompt injection.Reactions across platforms have mixed excitement about automating routine coordination with questions about real-world reliability, integration quality, and the balance between autonomy and human oversight.

Early feedback from testers points to time savings on repetitive data work and cross-team handoffs, though outcomes will likely vary based on implementation details and review processes.

This release continues OpenAI’s pattern of evolving ChatGPT from a standalone chatbot toward a more embedded infrastructure component in professional settings, reflecting broader industry moves to make AI systems active participants in organizational operations rather than passive responders.