How OpenAI’s New Workspace Agents Turn Any Team Task into an Automated Agent

OpenAI announced that Workspace Agents are now integrated into ChatGPT, positioning them as "shared agents that can work across tools and teams on complex, long‑running tasks." They are built on Codex, run entirely in the cloud, and extend GPTs with memory, tool usage, approval flows, scheduling, and multi‑step execution.

To create an agent, users open the "Agents" pane in ChatGPT, describe the desired workflow, and the system guides them through writing instructions, configuring tools, adding skills, testing, and iterating. No prompt‑engineering expertise is required; a document can even be uploaded for the model to process automatically.

OpenAI provides five ready‑to‑use templates that reflect real internal use cases:

Scout (product‑feedback routing) : aggregates feedback from Slack, support tickets, and public comment sites, tags and prioritizes items, and produces a weekly action list.

Spark (sales‑lead follow‑up) : filters inbound leads, scores them, drafts personalized outreach emails, and syncs to a CRM.

Tally (weekly report bot) : pulls data every Friday, generates charts, writes analysis, and emails the team.

Slate (software‑request review) : validates employee software requests against approved lists and policies, creates approval flows, and opens IT tickets.

Trove (third‑party risk) : assesses supplier sanctions, financial health, and reputation, then produces a structured risk report.

Angle (marketing planning) : adds creative scenario generation.

Additional templates exist for finance, sales, and marketing functions, ready to be customized.

Real‑world examples include:

Sales scenario: an agent automatically compiles call notes, conducts account research, filters new leads, drafts follow‑up emails, and delivers them to reps, saving 5‑6 hours per week.

Month‑end accounting scenario: an agent processes journal entries, adjusts balance sheets, performs variance analysis, and generates review materials with supporting documentation, all within minutes.

Employee‑support scenario: an agent answers questions in a Slack channel, provides relevant documentation links, and opens tickets for new issues.

Rippling’s AI engineer Ankur Bhatt noted that the hardest part of building agents is integration, memory, and user experience—not the model itself— and that Workspace Agents let non‑engineers create end‑to‑end sales‑opportunity agents without engineering help.

The platform’s cross‑tool capability lets agents pull context from documents, email, chat, code, and business systems, and perform approved actions such as updating Linear issues, creating Google Docs, or sending Slack messages. Currently agents are usable in ChatGPT and Slack, with more platforms promised.

Enterprise governance features include admin‑controlled tool and data access, approval requirements for sensitive actions, fine‑grained group permissions for ChatGPT Enterprise and Edu, a built‑in prompt‑injection guard, and a Compliance API that offers visibility into configuration, change history, and runtime logs, plus one‑click pausing of any agent.

OpenAI highlights three strengths: (1) upgrading conversational GPTs to a full‑featured agent platform with memory, tools, approvals, scheduling, and cross‑app integration; (2) comprehensive enterprise‑grade governance; and (3) a free‑trial window until May 6, after which usage is billed by credits.

Three caveats are noted: (1) the feature is limited to ChatGPT Business, Enterprise, Edu, and Teachers plans, excluding Plus and Pro users; (2) credit pricing has not been disclosed, creating budgeting uncertainty; and (3) OpenAI plans to eventually deprecate the original GPTs in favor of Workspace Agents, requiring existing GPT‑based workflows to migrate.

Sam Altman previously predicted an “extremely multi‑agent” future; Workspace Agents represent the first concrete product bringing that vision to enterprise settings.