OpenAI's Workspace Agents Arrive in ChatGPT for Teams

OpenAI just put team agents inside ChatGPT

Yesterday, OpenAI introduced workspace agents — shared, Codex-powered agents that sit inside ChatGPT and run multi-step work for entire teams. Same chat sidebar, very different product. For small businesses watching the agent space, this is the moment ChatGPT stopped being a chatbot.

The pitch is straightforward: describe a workflow your team does often, and ChatGPT builds it into a reusable agent. It runs on a schedule, can be triggered from Slack, remembers what it has done, and keeps working while you sleep. OpenAI is calling workspace agents “the next evolution” of Custom GPTs.

What actually shipped

Workspace agents are available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. They are free through May 6, 2026, after which OpenAI will switch to a credit-based pricing model.

A few details worth pinning down:

• Powered by Codex — OpenAI’s software engineering agent model. Agents can write code, run code, call connected apps, and loop through multi-step tasks.
• Cloud-resident — they run even when you are offline. You don’t need to keep a browser tab open.
• Shareable — build an agent once, use it across the team in ChatGPT or Slack.
• Scheduled or triggered — kick off on a cron-like schedule or when a Slack message lands.
• Permissioned — admins can limit what data and tools an agent touches, and require approval for sensitive actions.

OpenAI’s own examples: preparing reports, writing code, and responding to messages.

Why this matters for small businesses

The “AI employee” idea is officially mainstream

We have been making the case for shared, persistent, job-specific AI agents for a while now — it is the whole premise of AI Employees. Workspace agents validate that the direction is right. OpenAI, with nearly a billion weekly users, just told every business in its orbit that agents should be shared across teams, not stuck inside individual chats.

That is a big cultural shift. A year ago, ChatGPT was “a thing I use at my desk.” Today, OpenAI is pitching it as “a thing that runs jobs for my whole team, on a schedule, even when nobody is logged in.” The same logic we apply to Dispatch, Torque, and the rest of the lineup.

The fine print is where small business hits a wall

Workspace agents are Business/Enterprise only. ChatGPT Team starts at around $30 per user per month, and after May 6 you will also pay credits for agent runs. For a five-person HVAC office, that is a real line item — and you still have to build the agents yourself, connect them to your systems, and maintain them.

Most small businesses don’t have an in-house person who wants to spend their week describing workflows to ChatGPT and fixing them when something changes. They want the phone to stop ringing after hours, the reviews to get responded to, and the leads to get qualified. That gap — between “here is a platform” and “here is a specific job, done” — is where the market for SMBs still sits.

Our take

The bottom line: OpenAI just made the “AI agents at work” model official. For small businesses, the platform is now credible, but the work of turning it into a business outcome is still on you.

A few things the coverage is missing:

• Credit pricing is the real story. Free through May 6 is a hook. The credit model that follows will determine whether this is an occasional convenience or a line item that needs budgeting. Watch what OpenAI charges per agent run for common tasks — that number will decide whether SMBs can justify building workflows at all.
• Slack is a quiet win for Microsoft’s rivals. Workspace agents drop directly into Slack. Microsoft is building competing agents inside Teams and Copilot. If your business standardized on Slack already, OpenAI just gave you a reason to stay.
• “Build your own” is not the same as “get the job done.” Custom GPTs had the same promise and most small businesses never touched them. The ones that worked were the ones someone else built and handed over. Workspace agents will follow the same curve — expect a marketplace or template layer within six months.

Questions that remain

• What happens when an agent breaks silently at 2 a.m. because Slack rate-limited it?
• How will OpenAI price long-running agents (the kind that actually replace human work)?
• Can an agent built on Business tier be moved to Enterprise without rebuilding it?

What you should do this week

If you already pay for ChatGPT Business or Enterprise, the answer is simple — try it before May 6 while it is free. Pick one workflow you already do by hand every week. A status report. A weekly marketing digest. Triage of inbound emails. Describe it to the new Agents tab and see what ships.

If you are on a personal or free ChatGPT plan:

1. Don’t upgrade yet. Wait until OpenAI publishes the credit pricing after May 6.
2. Watch which use cases actually stick. Small businesses are going to tell the real story on LinkedIn and in Slack communities over the next month.
3. Audit what “AI work” you are already doing manually. The value of any agent — OpenAI’s or ours — is directly proportional to how well you have defined the job.

If you have been waiting to see whether AI agents were real before acting, the question is answered. They are here, and they are going to get cheaper and more capable every quarter.

Where this leaves the rest of us

OpenAI is building a horizontal platform. That is a good thing — it validates the whole category and pulls the price of agent capability down for everyone. It also means SMBs still need vertical, opinionated, “here is the job, done” options alongside it. That is where we have been building, and this week’s news only makes the case sharper.

Thinking about putting an AI agent to work on a specific job in your business — after-hours intake, dispatching, review responses, lead qualification? Explore our AI Employees or get in touch to talk through what fits.