OpenAI has launched “workspace agents,” a new class of semi-autonomous agents that can perform a wide array of tasks for ChatGPT users in enterprise and team environments.
Currently in research preview, workspace agents allow organizations to create AI agents that execute multi-step workflows, interact with enterprise software, and continue operating without direct user input. Unlike earlier iterations of customized ChatGPT instances, these agents are built to persist and collaborate across teams.
The release changes how OpenAI positions ChatGPT for work. Rather than serving as a conversational assistant, the platform is being reoriented toward ongoing task execution.
The feature is currently available to ChatGPT Business, Enterprise, education, and teaching users, with free access during the preview period. A usage-based pricing model is expected to follow.
Using Codex
Codex, OpenAI’s code-oriented model, enables agents to perform actions including running code, retrieving and transforming data, and interacting with third-party applications like Slack and other enterprise suites. The agents can also retain context across tasks, allowing them to build continuity over time rather than resetting with each prompt.
Users create agents by describing a workflow within ChatGPT, using a text prompt. The system then maps the steps, connects relevant tools, and tests execution before deployment. Once active, agents can run on schedules or respond to triggers, including incoming messages or updates in connected systems.
OpenAI has provided early examples of internal use. These include agents that compile sales intelligence, generate weekly performance reports, route product feedback, and assist with accounting workflows. In each case, the agent is set up for oversight through approval checkpoints.
Shared access is a key feature. Agents can be published within an organization and reused by multiple teams, creating a repository of automated workflows. Over time, these agents can be refined through use, effectively capturing company knowledge in executable form.
Security and governance remain central concerns. OpenAI says administrators can define what data and systems agents can access, as well as require human approval for sensitive actions such as sending communications or modifying records. Monitoring tools provide visibility into how agents are configured and used, and safeguards are intended to reduce risks such as malicious prompt manipulation.
Not Early to the Market
To say that major tech vendors are investing heavily in agent-based systems is an understatement. At this point it appears releasing AI agents is the leading market push by tech companies. Consequently, OpenAI faces pressure from competitors developing similar capabilities, particularly those integrating AI into workplace tasks.
Big picture, it is almost surprising that OpenAI waited until now to release this type of no-code agent solution. On the other hand, there may be a strategic pattern here. OpenAI has entered categories after initial experimentation has happened elsewhere, then worked to standardize the solution. ChatGPT itself was not the first language model interface, but it defined the category for mainstream use. Workspace agents may follow a similar path if adoption takes hold.
The release also marks another change for OpenAI: it represents a move away from custom GPTs, which OpenAI introduced as a way to tailor ChatGPT for specific use cases. While those tools will remain available for now, the company plans to provide a path for converting them into agents.
