OpenAI launched workspace agents in ChatGPT, a Codex-powered system that lets teams build shared agents to handle complex, long-running workflows inside org-defined permissions. The framing is direct: workspace agents are an evolution of GPTs—same idea of packaging knowledge into something reusable, but now with cloud execution, shared org access, and the ability to keep working when nobody's watching.
The product is in research preview today for ChatGPT Business, Enterprise, Edu, and Teachers plans. It's free until May 6, 2026, when credit-based pricing kicks in. OpenAI says it'll soon be easy to convert existing GPTs into workspace agents.
From "Helps One Person" to "Runs a Process"
The core insight in the launch: AI has already made individuals faster, but most important workflows in an organization depend on shared context, handoffs, and decisions across teams. A GPT helps the person prompting it; a workspace agent runs a process. It can gather context from the right systems, follow team procedures, ask for approval at sensitive steps, and keep work moving across tools.
To create one, you click Agents in the ChatGPT sidebar, describe a workflow your team does often, and ChatGPT walks you through defining steps, connecting tools, adding skills, and testing it. Templates exist for finance, sales, marketing, and other functions, each shipping with built-in skills and suggested tools.
Five Agents OpenAI Highlights
OpenAI's launch leans on patterns its own teams already run in production:
- Software Reviewer — triages employee software requests, checks them against approved tools and policies, recommends next steps, files IT tickets when needed.
- Product Feedback Router — monitors Slack, support channels, and public forums; turns feedback into prioritized tickets and weekly product summaries.
- Weekly Metrics Reporter — pulls Friday data, generates charts, drafts the narrative, delivers a business report.
- Lead Outreach Agent — qualifies inbound leads, drafts tailored follow-ups, updates the CRM.
- Third-Party Risk Manager — screens vendors for sanctions, financial, and reputational risk, then delivers structured reports.
Codex in the Cloud, Slack as a Surface
Agents are powered by Codex in the cloud—each gets a workspace for files, code, tools, and memory. They write or run code, use connected apps, remember what they've learned, and continue work across multiple steps. Today they can be reached through ChatGPT and Slack, with more surfaces coming. OpenAI's product team built one that proactively answers employee questions in Slack channels, links the relevant docs, and files a ticket when it spots a new issue.
Agents also run on schedules. OpenAI's accounting team built one that prepares month-end close—journal entries, balance-sheet reconciliations, variance analysis—generates workpapers with control totals for review, and follows internal policies. Rippling's early test of a Sales Opportunity agent collapsed 5–6 hours per rep per week into a background process.
Enterprise Controls and Compliance
For sensitive steps—editing a spreadsheet, sending email, adding a calendar event—admins can require the agent to ask permission before acting. ChatGPT Enterprise and Edu admins control which connected tools and actions user groups can access, and who can use, build, or share agents. Built-in safeguards aim to keep agents aligned with their instructions when they encounter prompt injection attacks in external content.
The Compliance API exposes every agent's configuration, updates, and runs for monitoring. Admins can suspend agents at will, and a forthcoming admin-console view will show every agent built across an organization, with usage patterns and connected data sources.
Why It Matters for Web Developers
Workspace agents are OpenAI's clearest answer to a practical question every team has been wrestling with: where does the agent actually live? Not in a one-off prompt, not in a personal GPT, but as a governed, shared, observable thing inside the org's permissions model. For developers, the read is twofold: first, expect more org-internal automations to migrate from custom-built bots into ChatGPT agents over the next few quarters; second, the convergence of Codex execution, Slack integration, and admin controls is a direct push against bespoke agent frameworks. If the agent you'd otherwise build with LangChain plus a Slack webhook plus an internal admin UI now ships as a single configured ChatGPT object with a Compliance API behind it, the build-vs-buy math has shifted.