OpenAI drops workspace agents in ChatGPT to automate team workflows

Most people still treat ChatGPT like a personal scratchpad. You ask a question, it answers, and the moment you close your laptop, the work stops.

OpenAI is trying to break that loop. The company just introduced workspace agents in ChatGPT, rolling out in research preview for Business, Enterprise, Edu, and Teachers subscribers.

Powered by Codex, these aren't just polite conversational bots. They are persistent, cloud-based assistants designed to tackle long-running tasks and complex team handoffs even while you are logged off.

Asynchronous workers living in your Slack

Workspace agents represent the next evolution of custom GPTs. Instead of waiting for prompts, they run in the cloud on scheduled triggers or sit inside tools like Slack, picking up requests as they happen.

Building one requires describing a routine workflow in plain English or dropping in a reference file. ChatGPT walks you through defining actions, integrating tools, and testing permissions. Pre-built templates are also available for finance, sales, and marketing.

Inside OpenAI, teams are already leaning heavily on them. Sales reps use an agent to scrape call notes and research accounts before drafting follow-up emails right into their inboxes.

Their accounting team uses another agent to handle core month-end closing tasks, reconciling balance sheets and running variance analysis in minutes. Other internal examples include an IT software reviewer, a product feedback router, and an automated weekly metrics reporter.

OpenAI confirmed that existing GPTs will stick around for now, with a direct migration path to workspace agents planned soon.

Admin controls and the ticking free trial

Letting an autonomous agent roam inside corporate databases usually gives security leaders nightmares. OpenAI seems well aware of the hesitation.

To keep agents in check, teams can set mandatory human approvals for sensitive steps like editing spreadsheets, firing off external emails, or scheduling meetings. Built-in defenses also guard against prompt injection attacks.

For IT managers, a Compliance API monitors configurations, agent runs, and connected data sources, alongside tools to audit or immediately suspend any rogue agent.

As for the cost, workspace agents are completely free to test until May 6, 2026. After that date, OpenAI plans to switch usage to a credit-based pricing model.

Why it matters

Up until now, enterprise generative AI has mostly helped individuals draft emails faster. But real office bottlenecks rarely come from typing speed; they happen during handoffs, status checks, and data collection between teams.

We think this is OpenAI's most direct move yet toward replacing glue work inside companies. For business teams bogged down by weekly pipeline hygiene, invoice reconciliation, or vendor risk assessments, testing a few templates during the free window is an easy call.

For the broader industry, OpenAI is moving the battleground from chat boxes to operational pipelines. It immediately puts pressure on workflow automation startups and forces platforms like Slack and Microsoft Copilot to defend their turf as the central operating layer.

That said, cloud agents running asynchronous tasks can easily turn into credit-guzzling black boxes once metered billing kicks in. Keeping guardrails tight and starting with narrow, repetitive tasks will be essential before handing over any mission-critical keys.