As of April 25, 2026, OpenAI Workspace Agents are research-preview ChatGPT agents for Business, Enterprise, Edu, and Teachers workspaces, built for repeatable team workflows that use workspace data, apps, tools, files, skills, and custom MCPs. Check access before you build: rollout is gradual, eligible Enterprise workspaces need admin enablement, Enterprise workspaces using EKM are not included at launch, and the preview is free only until May 6, 2026.
The right first build is not an autonomous writer. Choose one repeatable, low-risk workflow: gather approved context, draft a brief or action list, test it in Preview, and keep write actions behind human approval.
Use GPTs and workspace agents side by side while your team tests the new route. OpenAI describes workspace agents as an evolution of GPTs, not an immediate replacement; GPTs remain available during the transition and can later become part of agent workflows.
Slack is a channel option, not a shortcut around workspace readiness. It requires the ChatGPT agents app for Slack, a paid Slack Enterprise Grid setup and admin configuration, plus an eligible ChatGPT Business, Edu, or Enterprise account; governance still comes from admin controls, connector permissions, previews, approvals, and auditability.
Stop if the first candidate workflow needs unsupervised writes, broad private data access, external customer action, or unclear rollback. Build the narrow reviewable draft first, then expand after the agent has proven its value and control boundaries.
Quick Answer: Check Access Before Setup
OpenAI announced workspace agents in ChatGPT on April 22, 2026. The product is not just another personal chatbot. It is a shared ChatGPT workspace feature, powered by Codex, that can keep working in the cloud on team workflows while staying under organization permissions and controls.
The first practical question is whether the workspace is eligible today.
| Reader question | Current answer on April 25, 2026 | What to do next |
|---|---|---|
| Can a personal Pro user enable Workspace Agents? | OpenAI's public eligibility list names Business, Enterprise, Edu, and Teachers, not Pro. | Do not hunt for a Pro toggle. Use an eligible workspace or wait for a future change. |
| Is every eligible workspace enabled already? | No. OpenAI's release notes say rollout is gradual, and the Help Center says Enterprise has the feature off by default at launch. | Ask the workspace admin to check the Agents controls before drafting a build plan. |
| Can Enterprise customers with EKM use it at launch? | No. The release notes say Enterprise workspaces using EKM are not included at launch. | Keep the workflow in GPTs, ChatGPT, or existing automation until OpenAI changes the EKM status. |
| Is it free? | OpenAI says workspace agents are free until May 6, 2026, then move to credit-based pricing. | Treat any cost projection as temporary until pricing is rechecked after May 6. |
| Is Slack included automatically? | No. The Slack app route has its own paid Slack Enterprise Grid and admin setup requirements. | Validate Slack prerequisites only after ChatGPT workspace eligibility is clear. |
That access order prevents wasted setup work. If the workspace is not eligible, or if an Enterprise admin has not enabled the feature, the builder steps do not matter yet. If the workspace is eligible, the next decision is whether the task is a good agent workflow at all.
What Workspace Agents Are, And What They Are Not
Workspace agents are shared agents inside ChatGPT. They can use workspace knowledge, tools, connected apps, files, skills, and custom MCPs to complete multi-step workflows for a team. OpenAI's Workspace Agents Academy page frames strong candidates as repeatable, structured, event-driven or time-driven, and tool-based work. That is the mental model to keep.
A useful workspace agent is not "ChatGPT, but with more autonomy." It is closer to a governed workflow assistant that can gather context, produce an output, ask for review, and run on a schedule or trigger when the workflow has clear boundaries.
This distinction matters because many tasks still belong elsewhere:
- Use regular ChatGPT when the work is exploratory, personal, or one-off.
- Use an existing GPT when the job is a custom assistant pattern that does not need shared workspace execution, tools, schedules, or channels.
- Use deterministic automation when the task has rigid rules and no language judgment.
- Use a workspace agent when the same team workflow repeats, needs workspace context, and benefits from a reviewable draft or action.
OpenAI's launch post also makes the GPTs relationship specific: workspace agents are an evolution of GPTs, and GPTs remain available while teams try the new route. That means the safer migration plan is coexistence, not a rushed rewrite. Keep a working GPT in place until the agent version has proven that it can use the right data, stay within permissions, and produce a better team workflow.
If the word Codex is confusing here, separate the product layers. Workspace agents are Codex-powered ChatGPT team agents. That is not the same decision as choosing a coding tool such as Codex versus another coding agent. For developer route choice, use a dedicated comparison such as Codex vs Claude Code; the workspace-agent decision is about ChatGPT team automation, permissions, and rollout.
Should This Workflow Become An Agent?
The best first candidate is narrow, repeatable, and reviewable. A good test workflow has an obvious trigger, a small set of approved data sources, a draftable output, and a human checkpoint before anything changes outside ChatGPT.
Good first candidates include:
- A weekly internal status brief that pulls from approved docs and calendar context.
- Meeting notes that become action items after a human review.
- A customer handoff summary that drafts from permitted CRM or document sources.
- A knowledge-base update proposal that never publishes without approval.
- A team reminder or digest that uses a fixed schedule and fixed source set.
Weak first candidates share the opposite shape. Avoid starting with broad "handle customer escalations," "manage our inbox," "update the database," or "post to Slack whenever needed" prompts. Those workflows are not impossible forever, but they mix open-ended judgment, private data, external writes, and unclear rollback. They should come after the team has already proven smaller agent behavior.
Use this quick filter:
| Workflow signal | Agent-ready? | Reason |
|---|---|---|
| Repeatable trigger, known sources, draft output, human approval | Yes | The agent can be evaluated against a stable job. |
| One-off brainstorming, no fixed source, no stable output | No | Regular ChatGPT is usually safer and faster. |
| Needs external writes, customer action, or policy-sensitive decisions | Not first | Keep it manual until controls and audit paths are mature. |
| Needs tools, files, skills, or custom MCPs to save repeated team effort | Often yes | The feature's value appears when the workflow crosses multiple approved resources. |
This is the central adoption rule: do not start with the most impressive workflow. Start with the workflow whose value and failure mode are easiest to see.
Build The First Safe Workspace Agent
OpenAI's Help Center describes a builder surface where admins or users can add apps, tools, custom MCPs, skills, files, and channels, then test the agent in Preview before creating and sharing it. The safest build sequence is:
- Define the job in one sentence.
- Name the trigger: manual mention, schedule, channel, or event.
- List the exact data sources the agent may use.
- Decide whether the output is a summary, recommendation, checklist, draft message, or file update proposal.
- Add the minimum tools, apps, files, skills, or MCP connections.
- Test in Preview with realistic examples.
- Keep write actions on approval.
- Share only after the team agrees on the success signal and stop rule.
The OpenAI developer cookbook example is useful because it does not treat an agent as magic. The sales meeting prep agent combines calendar context, SharePoint materials, web search, scheduling, skills, and sharing. That kind of example shows the real shape: the agent saves time because it coordinates known sources and produces a reviewable output.
For a first build, write the instruction like an operating contract:
Prepare a weekly customer success brief every Monday at 9 a.m. using only the approved account notes, the shared customer-health spreadsheet, and last week's meeting transcripts. Draft the brief in ChatGPT, cite the source documents used, list open risks, and ask the owner before posting to Slack or updating any file.
That prompt is less exciting than "act as our customer success agent," but it gives the team something measurable. It defines the trigger, sources, output, citation expectation, and write boundary.
Tools, Skills, MCP, Auth, And Write Approvals
The Help Center setup surface names several building blocks, and they do different jobs:
| Building block | Use it for | Risk to watch |
|---|---|---|
| Apps and connectors | Calendar, Drive, Slack, SharePoint, and similar workspace systems when enabled by admins. | Availability depends on workspace settings, connector permissions, and user or shared authentication. |
| Files | Stable reference material, templates, process docs, or examples. | Stale files can become hidden policy drift if the team never reviews them. |
| Skills | Reusable procedures, templates, or instructions that improve repeatable outputs. | A vague skill can hide bad assumptions inside the workflow. |
| Custom MCPs | External tools or data systems the agent needs for the workflow. | MCP scope must be narrow; broad access creates an unnecessary data and action surface. |
| Channels | Where the agent can be used or surfaced, such as ChatGPT or Slack. | A channel is not a permission model by itself. Admin settings still matter. |
Authentication deserves its own decision. OpenAI's Help Center says app authentication can use the end user's account or an agent-owned account. For agent-owned shared connections, it recommends service accounts when possible. That recommendation is practical: if a shared agent depends on one employee's personal credential, the team inherits that employee's access scope, departure risk, and personal-data exposure.
Write actions are the other hard boundary. The Help Center says write actions default to Always ask during an agent run. Keep that posture for first builds. A safe workspace agent should be allowed to draft, preview, cite, and recommend. It should not silently send emails, edit files, post messages, create tickets, or update records before the workflow has a proven review path.
Use a simple permission rule:
- Read only the sources needed for this workflow.
- Draft outputs inside ChatGPT first.
- Ask before every write.
- Log the source material and decision point.
- Expand scope only after several successful runs.
That rule is stricter than a launch demo, but it is how a team avoids confusing "agent can do this" with "agent should do this in our workspace."
Slack Route And Admin Governance
Slack is useful when the workflow lives where the team already works, but it is not the first prerequisite. The ChatGPT Agents app in Slack Help Center page says the route requires Slack admin setup, a paid Slack Enterprise Grid workspace, and a Business, Edu, or Enterprise ChatGPT account. The Workspace Agents Help Center also says Slack use requires shared auth connections for all app connections on that agent.
Treat Slack as a deployment channel after the agent has passed the ChatGPT-side test. A safe order is:
- Build and Preview the agent in ChatGPT.
- Confirm the workspace admin has enabled agents and the required apps.
- Decide which channels or user groups may access the agent.
- Use shared or service-account style connections where appropriate.
- Test mention-based use before schedule-based posting.
- Keep write approvals and audit visibility in place.
Slack changes the risk profile because messages can reach more people faster. A bad summary inside ChatGPT is contained. A bad Slack post can become an operational signal for the whole team. That is why channel access, user groups, connector sources, and approval policies should be reviewed together.
Use a stop rule before enabling Slack:
| If the workflow needs... | Do this first |
|---|---|
| Customer-facing messages | Keep output as a draft until the human owner approves. |
| File or database updates | Require explicit approval and log what changed. |
| Private channel context | Confirm the agent's access matches the channel's privacy expectations. |
| Shared auth | Prefer service accounts and document who owns the credential. |
| Scheduled runs | Run a manual preview cycle before scheduling. |
The goal is not to avoid Slack. The goal is to make the Slack route boring: clear audience, clear trigger, clear sources, clear approval, clear audit trail.
Pricing, Rollout, And Roadmap Watch
Workspace agents are launch-fresh. Treat every availability and cost statement as time-bound.
As of April 25, 2026, OpenAI says workspace agents are free until May 6, 2026, and credit-based pricing starts after that. That does not tell a team what a normal monthly cost will be after real usage begins. It tells the team to use the free preview window for small, measurable tests, then recheck pricing before expanding to scheduled, broad, or tool-heavy workflows.
The same caution applies to rollout. The Enterprise and Edu release notes say rollout is gradual over the next few weeks for Business and Enterprise workspaces. If one workspace sees the feature and another does not, that may be rollout timing rather than a configuration mistake. For Enterprise, the Workspace Agents Help Center says the feature is off by default at launch and needs admin enablement for eligible workspaces.
Roadmap language should also stay bounded. OpenAI says GPT conversion support and Codex app support are coming later, but those are not the same as current live controls. Plan the current rollout around what the Help Center documents today: builder setup, apps and tools, skills, files, channels, Preview, auth choices, schedules, approvals, and admin controls.
Safe Rollout Checklist
Use this checklist before the team calls a workspace agent ready:
- The workspace plan is eligible, and admin enablement is confirmed.
- EKM status, connector availability, and app permissions are checked.
- The workflow has one clear trigger and one clear output.
- The agent uses the smallest necessary set of apps, files, tools, skills, and MCPs.
- Auth is documented as end-user or agent-owned, with service accounts preferred for shared agent-owned connections.
- Write actions require approval.
- Preview tests use realistic inputs and record expected outputs.
- Slack is enabled only after ChatGPT-side behavior is stable.
- The team knows what the agent must not do.
- Pricing is rechecked before moving from preview tests to wider scheduled usage.
The best launch metric is not how many things the agent can touch. It is whether a repeatable workflow gets faster while the team can still explain the sources, approvals, and failure boundaries.
FAQ
Are OpenAI Workspace Agents available to ChatGPT Pro users?
OpenAI's current public eligibility language names ChatGPT Business, Enterprise, Edu, and Teachers plans for the research preview. It does not list personal Pro as an eligible plan. If you are on Pro and cannot find the feature, treat that as expected unless OpenAI later changes the plan list.
Are workspace agents replacing GPTs?
No. OpenAI describes workspace agents as an evolution of GPTs, and says GPTs remain available while teams test the new experience. The practical path is to keep useful GPTs running, then move a workflow to a workspace agent only when shared execution, tools, schedules, channels, or governance make the agent route better.
Why can one Business or Enterprise workspace see it while another cannot?
OpenAI's release notes describe a gradual rollout, and Enterprise workspaces have the feature off by default at launch. Admin settings, rollout timing, EKM status, and connector availability can all affect whether the feature is visible or usable.
What does EKM change?
The release notes say workspace agents are not available to ChatGPT Enterprise customers using Enterprise Key Management at launch. If your Enterprise workspace uses EKM, do not plan a production rollout until OpenAI documents support for that configuration.
Can a workspace agent send emails, edit files, or post to Slack automatically?
The feature can be configured with tools and apps that support actions, but the Help Center says write actions default to Always ask during a run. Keep that default for first builds. Let the agent draft and propose; require a human to approve external writes.
What is the difference between skills and custom MCPs?
Use skills for reusable procedures, instructions, templates, and workflow logic that the agent should apply consistently. Use custom MCPs when the agent needs a specific external tool or data surface. A first agent often benefits from a skill before it needs a custom MCP, because a good skill narrows behavior without adding a new integration risk.
How should Slack be enabled safely?
Enable Slack after the agent works in ChatGPT Preview. Confirm paid Enterprise Grid, Slack admin setup, eligible ChatGPT plan, shared auth needs, channel/user-group scope, and write approvals. Start with mentions or manual tests before scheduled posts.
What should a first workspace agent do?
Pick a workflow that repeats, uses known data, produces a reviewable draft, and has a clear human owner. Weekly briefs, meeting action items, handoff summaries, and knowledge-base update drafts are better first tests than broad inbox management or autonomous customer actions.
Why does OpenAI say workspace agents are powered by Codex?
It means the agent runtime uses OpenAI's Codex-powered cloud work loop for multi-step tasks. It does not turn workspace-agent adoption into a coding-assistant choice. For most teams, the useful question is still: can this ChatGPT workspace workflow be shared, permissioned, previewed, and approved safely?
Bottom Line
OpenAI Workspace Agents are worth testing when the team has an eligible workspace and a repeatable workflow that becomes better with shared context, tools, and governance. They are not a reason to automate the broadest process first.
Start with access, admin enablement, and EKM status. Then choose one workflow that can be drafted, previewed, approved, and measured. Keep GPTs in place while the agent proves itself. Add Slack only after the ChatGPT-side workflow is stable. Recheck pricing before broad scheduled use after May 6, 2026.
