OpenAI workspace agents
- OpenAI launched “workspace agents” that can run multistep tasks across apps and keep working while users are offline. - OpenAI also said GPT‑5.4 “Thinking” is stronger at document reasoning and producing polished frontend code. - These agents change team workflows and require clearer permission, audit, and delegation rules across Slack, Salesforce and cloud apps. (theverge.com)
OpenAI has launched workspace agents in ChatGPT, pushing the product from one-off prompts toward shared software that can run team workflows on its own. (openai.com) The company says the agents can automate multi-step jobs across Slack, Google Drive, and Microsoft SharePoint, and can run on schedules for recurring work like lead review, support summaries, and report generation. OpenAI is offering the feature in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. (openai.com) OpenAI also says the agents are built to be shared across a company, with one person creating an agent and other employees using the same workflow inside ChatGPT. Admins can set role-based access, approval gates for sensitive actions, and audit logs to track what the agent did. (openai.com) A workspace agent is closer to a background worker than a chatbot. OpenAI says these agents run in the cloud, can keep working when a user is offline, and can take actions across connected tools instead of waiting for each next prompt. (9to5mac.com) That shift builds on GPT-5.4, which OpenAI released on March 5, 2026, as its main model for “professional work.” The company says GPT-5.4 combines reasoning, coding, tool use, and agentic workflows in one model. (openai.com) In ChatGPT, OpenAI says GPT-5.4 Thinking can show an upfront plan before it answers, hold context better on longer tasks, and improve deep web research on highly specific queries. The company also says the model produces more consistent work on documents, spreadsheets, presentations, and code. (openai.com) In Codex and the application programming interface, OpenAI says GPT-5.4 is its first general-purpose model with native computer-use capabilities. It also supports up to 1 million tokens of context, which OpenAI says helps agents plan, execute, and verify long workflows across apps. (openai.com) OpenAI is also using workspace agents to replace an older product tier. The company introduced custom GPTs in ChatGPT last year, and 9to5Mac reported on April 22 that workspace agents are the successor OpenAI is positioning for teams. (9to5mac.com) The practical question for companies is no longer whether ChatGPT can draft an answer, but what an employee is allowed to delegate to software that can read files, send messages, and update records. OpenAI’s own product page puts those controls at the center, with permissions, approvals, centralized management, and monitoring built into the pitch. (openai.com) That leaves OpenAI selling not just a smarter model, but a managed coworker inside the apps companies already use. The closer those agents get to Slack threads, customer records, and internal documents, the more the product depends on clear rules about who can let them act. (openai.com)
