OpenAI is pushing ChatGPT further into the day-to-day operations of companies, introducing this week shared “workspace agents” that can carry out tasks across tools and continue running without constant input.Powered by Codex, OpenAI’s coding agent, the new feature lets teams build agents that handle multi-step work such as drafting reports, triaging requests, or responding to messages, using data and tools available inside their organization. Unlike earlier versions of ChatGPT, which relied on a back-and-forth prompt model, these agents are designed to take a task and see it through.See also: OpenAI launches GPT-5.5, calling it “a new class of intelligence.”OpenAI said the agents run in the cloud and can be shared across a team, allowing multiple people to use and refine the same system over time. They can also be deployed in tools such as Slack, where they can respond to requests or trigger actions as work comes in.Shared workspace agentsA team effortOpenAI has been building out team-focused features in ChatGPT for some time. Products such as ChatGPT Enterprise and ChatGPT Business introduced shared environments, while custom GPTs allowed teams to create tailored assistants for specific tasks.“Workspace agents are an evolution of GPTs.” – OpenAIWorkspace agents build on that concept. Instead of responding to prompts, they are set up to follow a defined process: gathering information from connected systems and returning results. In some cases, that might involve pulling together weekly reports, qualifying sales leads, or reviewing internal requests against company policies.OpenAI said teams can build these agents by describing a workflow inside ChatGPT, which then guides users through connecting tools, defining steps, and testing the system. Agents can also be scheduled to run at specific times or respond to incoming requests automatically.Custom GPTs will remain available, according to OpenAI, with plans to allow teams to convert them into agents over time.Shared context, shared systemsThe shift toward shared agents reflects a broader effort to move AI beyond individual use and into team processes. Notion, for example, is exploring similar ideas with Custom Agents, building systems that operate across internal documents, tools, and communication channels.In OpenAI’s case, the agents draw on a workspace that includes files, code, connected apps, and memory. That allows them to work across different systems while keeping track of previous steps, rather than starting from scratch each time.“AI has already helped people work faster on their own, but many of the most important workflows inside an organization depend on shared context, handoffs, and decisions across teams. Workspace agents are designed for that kind of work.” – OpenAIOpenAI says teams can customize agents with their own tools, data sources, and internal processes. The company also highlighted early use-cases internally, including agents that compile sales notes, generate reports, and respond to employee queries in Slack channels.Workspace agents in action in SlackControls and oversightAs these systems take on more responsibility, OpenAI is also adding controls over what agents can access and do.Teams can set permissions around tools and data, and require approval for certain actions, such as sending emails or modifying files. Admins can monitor how agents are used, including how often they run and what systems they interact with. A compliance API provides visibility into agent configurations and activity.That focus on oversight comes as AI systems are being used with more sensitive internal data. OpenAI has also been working on related efforts around privacy, including its open source OpenAI Privacy Filter, which is designed to filter or limit how data is handled in AI systems.Workspace agents are being released in research preview for ChatGPT Business, Enterprise, and education-focused plans, with pricing set to move to a credit-based model in May.The post OpenAI debuts always-on agents to end the friction of manual team handoffs appeared first on The New Stack.