Artificial Intelligence

OpenAI Workspace Agents: Catching Up to Claude with Cloud-Powered Team Agents That Actually Work Where You Do

|Author: Viacheslav Vasipenok|4 min read| 12
OpenAI Workspace Agents: Catching Up to Claude with Cloud-Powered Team Agents That Actually Work Where You Do

OpenAI just fired a serious shot across Anthropic’s bow. On April 22, 2026, the company quietly launched Workspace Agents in ChatGPT — a new class of Codex-powered, cloud-native agents designed for real business workflows. These aren’t one-off chat toys. They run autonomously in the background, triggered by schedules or Slack messages, and handle repeatable, multi-step tasks across your tools — all while staying under tight organizational controls.

OpenAI Workspace Agents: Catching Up to Claude with Cloud-Powered Team Agents That Actually Work Where You DoThis is OpenAI’s clearest move yet to match (and in some ways leapfrog) what Anthropic has been building with Skills and Claude Apps. But instead of asking teams to come to a new interface, OpenAI is bringing the agents to where the work already happens: ChatGPT and Slack.

How Workspace Agents Actually Work

  • They run in the cloud, not in your chat window. Schedule them to fire every Friday or trigger them instantly from Slack. No need to keep ChatGPT open. 
  • Shared by design. Build once in your workspace — the whole team (or entire org) can use, remix, and improve it.
  • Memory + continuous learning. Agents remember context across runs and get better with feedback. You can correct them mid-conversation, and the changes stick.
  • Safety-first. Before any sensitive action (sending email, editing a spreadsheet, creating a calendar event), they ask for explicit human confirmation.
  • Multiple surfaces today, more coming. Start in ChatGPT or Slack; OpenAI says additional platforms are already in the works.

OpenAI also shipped built-in prompt injection protection when agents pull in external content — a quiet but critical enterprise feature.

Real Examples Already Live

OpenAI Workspace Agents: Catching Up to Claude with Cloud-Powered Team Agents That Actually Work Where You DoOpenAI showed off several ready-to-use (or ready-to-customize) agents that feel genuinely useful:

  • Software Reviewer — Scans incoming software requests against company policy and automatically creates IT tickets.
  • Product Feedback Router — Pulls feedback from Slack, support tickets, and public channels, turns it into prioritized Jira tickets, and generates weekly summaries.
  • Weekly Metrics Reporter — Every Friday it pulls data, creates charts, writes a narrative, and sends the full report.
  • Lead Outreach Agent — Qualifies inbound leads, drafts personalized follow-up emails, and updates your CRM.
  • Third-Party Risk Manager — Checks vendors for sanctions, financial health, and reputational risks.

There are also pre-built templates for finance, sales, and marketing teams.

For admins, there’s a full Compliance API with visibility into every agent’s config, updates, and runs, plus granular controls: group-based access to connectors and a one-click “pause agent” button.

The Anthropic Comparison (and Why This Feels Different)

Anthropic got to the agent party first with Skills and custom Claude apps. But those were largely individual-developer focused — powerful for solo power users, less natural for team-wide adoption.

OpenAI is coming at it from the opposite direction: Slack-first, team-first, enterprise-ready. Agents live where communication already happens. They’re built for repeatable processes that mix code, documents, and connectors — exactly where Codex shines. It’s a smarter entry point for most companies that already live in Slack and Google Workspace or Microsoft 365.

Availability and Pricing

OpenAI Workspace Agents: Catching Up to Claude with Cloud-Powered Team Agents That Actually Work Where You DoWorkspace Agents are now in research preview for:

  • ChatGPT Business;
  • Enterprise;
  • Edu;
  • Teachers plans.

Everything is completely free until May 6, 2026. After that, it switches to credit-based pricing (exact rates still to be detailed, but expect the usual OpenAI usage model).

Admins in Enterprise and Edu workspaces can enable agents org-wide and control them tightly from day one.

Also read:

The Big Question Everyone Is Asking

OpenAI Workspace Agents: Catching Up to Claude with Cloud-Powered Team Agents That Actually Work Where You DoThe economics. Until May 6 it’s a no-brainer — experiment wildly. After that, teams will need to do the math: which processes actually save enough time (or money) to justify the credits, and which are still cheaper to have a junior human handle?

OpenAI clearly believes the answer will tilt heavily toward agents for anything repetitive and rules-based. The rest of us will be running the numbers over the next two weeks.

If you’re on Business, Enterprise, Edu, or Teachers, head to ChatGPT right now and start building. The era of agents that actually live in your existing tools just began — and it’s moving fast.

Official announcement

Share:
0