ChatGPT Workspace Agents are OpenAI’s new shared, Codex-powered agents for team work. Introduced on April 22, 2026, these workspace agents are designed to help organisations turn recurring jobs into reusable workflows that can run across ChatGPT, Slack, files, code, and connected business tools. That makes the feature more important than another prompt enhancement. It pushes ChatGPT from individual assistance toward repeatable operational execution.
In OpenAI’s product announcement, the company describes workspace agents as an evolution of GPTs that can keep working in the cloud, use connected apps, remember what they learn, and ask for approval before sensitive actions. OpenAI also says the feature is available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans, with credit-based pricing starting after May 6, 2026. The related ChatGPT pricing page reinforces the broader business story: a secure, shared workspace, 60-plus apps, admin controls, and no training on business data by default.
For teams already investing in AI strategy, workflow automation, business process automation, and intelligent automation, the launch matters because it combines shared context, long-running execution, and governance in one place. Instead of asking one employee to rebuild the same prompt chain every week, the team can create an agent once, improve it through use, and share it where work already happens.
| Question | Practical answer |
|---|---|
| What are ChatGPT Workspace Agents? | Shared, Codex-powered agents that handle multi-step work across ChatGPT, Slack, files, and connected apps |
| Why do they matter? | They turn repeatable team work into reusable workflows instead of one-off prompts |
| What can they do? | Research, summarize, write reports, run code, use tools, ask for approval, and keep working in the cloud |
| Where do teams use them? | In ChatGPT today, in Slack today, and across more work surfaces over time |
| Who gets them now? | ChatGPT Business, Enterprise, Edu, and Teachers plans in research preview |
| What should leaders evaluate? | Workflow fit, approvals, connected-data boundaries, analytics, and measurable time savings |
| Buyer takeaway | ChatGPT Workspace Agents are a workflow product, not only a model feature |
ChatGPT Workspace Agents at a glance
ChatGPT Workspace Agents are easiest to understand as shared workers for recurring team processes. OpenAI’s examples are deliberately operational: software review, product feedback routing, weekly metrics reporting, lead outreach, and third-party risk management. Each example has the same pattern. Someone describes the job, ChatGPT helps define the steps, chooses the right tools, adds skills, and tests the workflow until the agent behaves the right way.
That framing matters because the product is not being sold as an isolated prompt template. It is being positioned as a long-running, cloud-based agent that can keep going when the user is away, run on a schedule, and respond in shared team surfaces. In other words, the launch is not only about better answers. It is about better follow-through.
This also explains why the launch is relevant to business leaders. A normal AI assistant can speed up an individual task. These agents aim to reduce coordination cost across a whole workflow. When an agent can gather context, use the right systems, produce a structured output, and route the next step, the value moves from personal productivity into team operations.
How ChatGPT Workspace Agents differ from GPTs
ChatGPT Workspace Agents differ from earlier GPTs because the product is built for shared execution instead of mostly personal customisation. OpenAI explicitly calls the new system an evolution of GPTs, and that distinction matters. A GPT can package instructions and knowledge. The new agents go further by running in the cloud, holding onto working context, using connected tools, and persisting as a reusable team workflow.
That difference changes the adoption story. Older GPT-style usage often depended on one motivated user who knew the right prompts and maintained the playbook informally. The new model gives teams a way to formalize that playbook. The workflow lives in a shared agent, not only in one employee’s chat history or one internal prompt document.
OpenAI also notes that GPTs will remain available while customers test the new system, and the company plans to make conversion easier. That is a sensible bridge. Teams do not need to discard what they already built, but they should understand that workspace agents are the more serious operating model when the goal is recurring work, shared ownership, and cross-tool execution.
Where ChatGPT Workspace Agents fit across Slack, files, and tools
ChatGPT Workspace Agents fit best where work is already fragmented across chat, documents, spreadsheets, source systems, and team coordination channels. OpenAI says they are powered by Codex in the cloud and have access to a workspace for files, code, tools, and memory. The practical implication is simple: the agent can do more than answer a question. It can inspect context, work through multiple steps, and continue the job across several assets.
Slack is especially important here. OpenAI says teams can deploy agents in Slack so they can respond to requests as they come in, answer questions in the flow of work, and file tickets when needed. That turns ChatGPT Workspace Agents into a shared service instead of another destination employees must remember to open.
The connected-tool story matters just as much. Business buyers care less about whether an agent sounds fluent and more about whether it can interact with the systems that hold the real work. If the product can move between file inputs, connected apps, and action steps without losing context, then teams can reduce handoffs and lower the friction between insight and execution.
Why ChatGPT Workspace Agents turn team knowledge into reusable workflows
ChatGPT Workspace Agents matter because knowledge at work is usually scattered. Policies live in docs, decisions live in chat threads, data lives in dashboards, and process knowledge often lives in one experienced employee’s head. OpenAI’s library-and-sharing model is aimed directly at that problem. Teams can build an agent once, improve it in use, and then share or duplicate it for related workflows.
That is a stronger operating model than leaving every recurring task as a fresh conversation. A reusable agent means the team can encode the right process, the right tool access, and the right review steps in one place. When a finance team builds a month-end close agent or a sales team builds a lead qualification agent, the organisation is turning tacit knowledge into a repeatable system.
Because the agents can be corrected in conversation and have memory, the workflow can get better over time instead of freezing on day one. That is one of the more commercially important parts of the launch. Teams rarely need a perfect first version. They need a system that becomes more reliable as usage reveals gaps, edge cases, and better ways to route work.
How ChatGPT Workspace Agents handle approvals and governance
ChatGPT Workspace Agents only become useful in real organisations if governance is built into the workflow, and OpenAI is clearly leaning into that point. The company says users stay in control by choosing which tools and data an agent can use, what actions it can take, and when approval is required. Sensitive steps such as editing a spreadsheet, sending an email, or creating a calendar event can be gated behind explicit permission.
That approval model is what separates a flashy demo from a deployable internal agent. Teams want automation, but they also want boundaries. ChatGPT Workspace Agents are more credible because the product assumes different steps carry different levels of risk. Low-risk research and summarization can run more freely. Higher-risk actions can pause for human review.
The admin story is also stronger than a typical consumer feature launch. OpenAI says Enterprise and Edu admins can control which tools and actions user groups can access, manage who can use, build, and share agents, and monitor configuration and runs through the Compliance API. OpenAI also says built-in safeguards are designed to help agents stay aligned when they encounter misleading content, including prompt injection attacks. For security-conscious teams, that governance layer is not optional. It is the reason the launch is worth evaluating at all.
Who gets ChatGPT Workspace Agents and how pricing works
ChatGPT Workspace Agents are available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. That availability matters because OpenAI is positioning the feature as a team capability, not a general consumer add-on. The product needs a shared workspace, admin controls, connected apps, and governance, so it fits more naturally inside the business-oriented tiers.
OpenAI says the feature will be free until May 6, 2026, with credit-based pricing starting after that date. That pricing model suggests the company expects agents to become an execution surface rather than a flat feature toggle. Teams will likely pay according to how much agent work they run, which is closer to an operations budget than a simple seat upgrade.
The surrounding Business plan details make that easier to read. ChatGPT Business is pitched as a secure, collaborative workspace with admin controls, SAML SSO, dedicated workspace features, and more than 60 app connections that bring company tools and data into ChatGPT. In that context, the new agent layer is less like a novelty and more like the workflow engine OpenAI wants to sit inside the broader business product.
How to evaluate ChatGPT Workspace Agents in a real team
ChatGPT Workspace Agents should be evaluated on one recurring workflow first, not on abstract enthusiasm. The right pilot is a process with clear repetition, too much manual stitching, and a human review point that is easy to define. Weekly reporting, software approval routing, lead qualification, vendor checks, support summarization, and internal knowledge triage are all better first candidates than a vague promise to automate “everything.â€
The evaluation checklist should stay practical. Measure how much context gathering the agent removes, whether the output quality is consistent enough for review, how often approvals are triggered, which connected systems are actually required, and whether the team uses the shared agent instead of falling back to ad hoc prompting. The feature only matters if the workflow becomes faster and cleaner in practice.
This is also where operating-model work matters. Teams need clear owners, approval rules, and success metrics before broad rollout. If you want help mapping the product onto a governed automation roadmap, contact Progressive Robot to design the workflow around the business loop instead of just adding another AI tool.
ChatGPT Workspace Agents FAQ
What are workspace agents in ChatGPT?
ChatGPT Workspace Agents are shared, Codex-powered agents in ChatGPT that can handle multi-step team workflows across files, tools, memory, and connected apps.
Are they the same as GPTs?
No. OpenAI describes the system as an evolution of GPTs. GPTs remain available, but workspace agents are designed for shared, cloud-running, tool-using workflows inside organisations.
Where can teams use them today?
OpenAI says teams can use them in ChatGPT and Slack today, with more work surfaces coming later.
Do they require admin controls?
In business settings, yes. The value depends on connected-tool permissions, approval gates, analytics, and visibility into how agents are configured and used.
Who can access them right now?
The feature is in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans, according to OpenAI.
The product is significant because it shifts ChatGPT from personal prompting toward shared operational workflows. If OpenAI executes well on tools, approvals, analytics, and pricing, it could become one of the clearest bridges between conversational AI and everyday business process execution.
