OpenAI is turning ChatGPT into something closer to a 24/7 digital colleague, not just a smart chat box you ping a few times a day. With the launch of “workspace agents,” the company is betting that the next wave of AI adoption will be less about clever prompts and more about agents that quietly run your team’s workflows in the background.

At a high level, workspace agents are shared AI workers that live inside ChatGPT and your existing tools, like Slack, and handle the boring but critical parts of knowledge work: pulling data, following checklists, chasing approvals, and keeping long-running tasks moving. They are built on OpenAI’s Codex stack, run entirely in the cloud, and are designed to understand context across tools so they can keep doing their thing even when nobody is actively chatting with them. Think of them as upgraded custom GPTs that have grown up from being personal assistants into team-level operators.

OpenAI is positioning workspace agents as the evolution of GPTs, the custom assistants users have been building since late 2023. GPTs were mostly single-user and prompt-driven; agents, by contrast, are meant to embody shared processes: a standard way your sales team qualifies leads, how your finance team closes the books, or the checklist product uses to triage feedback. The company’s own sales and accounting teams are already showcasing these use cases: one internal agent pulls call notes and account research, qualifies new leads, and drafts follow-up emails; another prepares journal entries, reconciles balance sheets, and generates month-end workpapers in minutes.

OpenAI really wants to make the setup feel approachable. Instead of asking you to design a flowchart from scratch, ChatGPT now has an “Agents” tab in the sidebar: you describe a workflow your team does often, or drop in a file that outlines the process, and ChatGPT walks you through turning that into an agent. Under the hood, it helps define the steps, plug in the right tools, add skills, and test the flow until it behaves the way you expect. For teams that would rather not start from a blank page, there are ready-made templates for roles like finance, sales, and marketing, complete with common tools and skills preconfigured so you can tweak instead of reinvent.

To make the concept concrete, OpenAI highlights a handful of agents it has already built in-house, pitching them as blueprints for customers. There is a software review agent that triages employee app requests, checks them against company policy, and automatically opens IT tickets with next steps. A product feedback router listens across Slack, support channels, and public forums, tags and prioritizes the feedback that matters, and turns it into weekly product actions. Another agent pulls metrics every Friday, generates charts, drafts the narrative, and emails a weekly business report, essentially doing the grunt work of status reporting for you. On the revenue side, a lead outreach agent researches inbound leads, scores them against your rubric, drafts personalized follow-ups, and updates the CRM. There is even a third-party risk manager that screens vendors for sanctions, financial and reputational risk, then ships a structured report.

All of this is powered by a Codex-based environment that can work across “dozens of tools,” as OpenAI puts it. Instead of just generating text, an agent can write and run code, call connected apps, use memory to store what it’s learned, and stretch its work across multiple steps and longer time spans. OpenAI emphasizes that agents can run on a schedule or sit inside Slack, waiting to respond to requests as they come in. One internal example is a product agent that lives in Slack channels, answering employee questions, linking relevant documentation, and filing a ticket when it discovers something that needs follow-up. The promise is that teams spend less time coordinating in threads and more time acting on the output.

The strategic angle is clear: OpenAI wants ChatGPT embedded “in the flow of work,” not just in a browser tab. Today, teams can interact with workspace agents in ChatGPT and Slack, and the company says more surfaces are coming soon. This dovetails with ChatGPT Business and Enterprise’s existing push into integrations with tools like Slack, GitHub, Google Drive, and more. The new agents sit on top of that foundation, trying to orchestrate actions across systems rather than forcing users to bounce between apps themselves.

Crucially, OpenAI is not just talking about productivity; it is heavily leaning on governance, which has become table stakes for enterprise AI deals. Workspace agents come with role-based controls so admins can specify which tools and actions different user groups can access, who is allowed to build and share agents, and which workflows require explicit human approval. For higher-risk actions like editing spreadsheets, sending emails, or adding calendar events, admins can force the agent to stop and ask before proceeding. There is also analytics built into the editor, so you can see how often each agent runs, how many people use it, and how it is performing over time.

On the security front, OpenAI says the agents inherit enterprise-grade monitoring and controls, with a particular focus on keeping them aligned even when they encounter malicious or misleading content like prompt injection attacks. The company’s Compliance API gives admins visibility into each agent’s configuration, updates, and runs, and allows them to suspend agents if something looks off. Soon, admins will also get a consolidated view of every agent built across their organization, including usage patterns and connected data sources, which should help large companies keep “shadow agents” in check.

Early customer feedback, at least the handpicked kind, is predictably positive. Rippling, one of the early testers, says the hard part of agents is not the model but everything around it: integrations, memory, UX. According to a testimonial from Ankur Bhatt on OpenAI’s site, a sales consultant at Rippling was able to build a “Sales Opportunity” agent from scratch—no engineering team required—that researches accounts, summarizes Gong calls, and posts deal briefs directly into Slack, cutting 5 to 6 hours of manual work per rep each week. Other names like SoftBank, Better Mortgage, BBVA, and Hibob are listed as early users as well, underscoring that OpenAI is targeting global enterprises, not just startups.

In terms of access, workspace agents launch in “research preview” and are only available for paying organizational plans: ChatGPT Business, Enterprise, Edu, and the dedicated Teachers plan. That keeps them squarely in the domain of teams and institutions rather than individual free users. OpenAI says the feature will be free until May 6, 2026, when it transitions to a credit-based pricing model, similar to how other advanced capabilities on the platform are monetized. Meanwhile, existing GPTs are not going away: OpenAI notes that GPTs will remain available while teams test workspace agents and that it will soon provide a straightforward path to convert GPTs into agents.

If you zoom out, workspace agents fit neatly into a broader shift in how AI is being sold to businesses. The first phase was about copilots and chatbots that helped individual workers move faster in familiar tools. The next phase, which OpenAI is now explicitly chasing, is about systematizing that assistance into reusable workflows that reflect how teams actually operate. Rather than every employee crafting their own prompts, the organization defines a “best practice” once—say, how to qualify a lead or handle a vendor risk review—and bakes it into an agent that everyone can use and gradually refine.

Of course, a lot of this will come down to real-world friction: how easy it is for non-technical teams to wire agents into their tools, how reliably those agents behave in messy edge cases, and whether governance controls are strong enough to comfort risk-averse IT and compliance teams. But OpenAI clearly sees agents as the next logical step for ChatGPT’s enterprise story: a move from chat to continuous collaboration, from isolated prompts to shared workflows, and from AI as a clever assistant to AI as a dependable coworker that never sleeps.