On April 22, OpenAI shipped workspace agents in ChatGPT, a feature that reframes the product from a conversational AI tool into something closer to a shared automation platform for business teams. The launch is available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans.

The core idea is straightforward. Instead of each person running their own ChatGPT session for repetitive work, a team now builds one agent that handles a recurring task, publishes it to their organization, and lets the whole team use it. The agent runs in the cloud, which means it keeps working in the background whether or not anyone is actively watching.

What Workspace Agents Actually Do

Workspace agents are powered by Codex, OpenAI’s software development model, and are positioned as an evolution of GPTs. Unlike GPTs, which were largely one-person tools, workspace agents are built to be shared, improved collaboratively over time, and connected to business systems.

OpenAI describes four example workflows the feature targets out of the gate:

Software Reviewer — monitors incoming software requests, checks them against predefined criteria, and files IT tickets automatically.

Product Feedback Router — watches communication channels for customer feedback, classifies it, and creates prioritized tickets in a project management tool.

Weekly Metrics Reporter — pulls data from connected sources, formats it, and distributes a report to the relevant team members on a schedule.

Lead Outreach Agent — researches new leads, drafts personalized outreach, and updates CRM records without requiring manual data entry.

All of these run inside the organization’s ChatGPT workspace, with access to uploaded files, code, external tools, and persistent memory across sessions. They can request human approval before taking certain actions, making them practical for workflows where oversight matters.

Agents integrate with Slack, allowing output and notifications to land in the communication channels teams already use.

Pricing and Availability

Workspace agents are free until May 6, 2026. After that, usage will be billed against OpenAI credits, following the flexible credit model already in place for ChatGPT Business plans. Organizations can set spend controls to manage usage across teams.

Why This Matters Now

OpenAI’s direction here tracks a shift that has been visible across every major enterprise AI platform over the past six months. Microsoft launched Agent 365 to make autonomous agents a native part of the Microsoft 365 workplace. Google’s opening keynote at Cloud Next 2026 this week was titled “The Agentic Cloud” and framed agents as the new unit of enterprise infrastructure. Salesforce has rebuilt Slack around agentic workflows.

Workspace agents fit that same direction but land inside the tool 40% of enterprise workers already use daily. OpenAI does not need to convince teams to adopt a new platform. They are shipping agents into the workflow people already have.

The practical implication is that the barrier to deploying a working business automation just dropped significantly. A non-technical operations manager can now describe a workflow they run every week, and ChatGPT will walk them through turning it into a shared agent that runs automatically. No API integrations, no development team, no vendor negotiation.

What This Means for Business

The ChatGPT workspace agents launch continues a clear pattern: the tools businesses use every day are becoming agentic by default. Repetitive knowledge work, data collection, ticket routing, report generation — these are the tasks that agents now handle without a standing human instruction each time.

For most businesses, this is a useful starting point. Automating a reporting workflow or an IT intake process through workspace agents is achievable this week for any team on a Business or Enterprise plan.

But the architecture of these agents has a ceiling. Workspace agents operate within ChatGPT’s environment and connect to the tools OpenAI has chosen to integrate. They are optimized for common knowledge-work tasks, not for industry-specific processes, proprietary data environments, or multi-agent workflows that span an entire operation.

Businesses that need AI agents to connect with their own systems, make decisions using their own data, and coordinate across functions need a different architecture than what a SaaS platform ships by default. That is the distinction between adding AI capabilities to software and building an AI workforce for your specific operation.

Enterprise DNA put together a free field guide on exactly this: the full Claude ecosystem, Claude Code, and how to roll agents out without breaking things. Get the guide.