Is ChatGPT an AI Agent Now? Workspace Agents vs Custom GPTs

Custom GPTs are deprecating for business accounts. Workspace Agents replace them. Here's what's different — and what to do about your old GPTs.

OpenAI announced Workspace Agents on April 22 and buried a small sentence halfway down the launch blog: Custom GPTs are being deprecated for business accounts. No date yet. No forced migration this month. Just a quiet note that the format your team has been authoring in since late 2023 has an expiration label on it.

For the roughly two million people who’ve built Custom GPTs — and the much larger group who use them every day — this is the most consequential part of the announcement. Not the new integrations. Not the approval UX. The part where OpenAI said, without saying, “the thing you’re using is not the thing we’re building forward.”

This is the honest explainer of what changed, whether you should care, and what to do about your existing Custom GPTs. No hype, no FOMO. Just the actual decisions.

The short version

Custom GPTs are a saved ChatGPT prompt with a personality, a few instructions, sometimes a custom tool (an Action). You talk to them, they talk back. Launched November 2023. You have one, probably several.

Workspace Agents are the same idea — a saved, shareable AI configuration — but the model (Codex-based) can take actions in your tools. Read Salesforce. Post to Slack. Send an email. Update a spreadsheet. Run on a schedule. Remember across conversations. Launched April 22, 2026.

What OpenAI announced about the transition:

  • Individuals keep Custom GPTs indefinitely. Nothing changes today.
  • Business, Enterprise, Edu, Teachers accounts will be required to convert their Custom GPTs to Workspace Agents at a future date. OpenAI has not named one.
  • A one-click conversion tool is coming — Custom GPT → Workspace Agent.
  • Workspace Agents are free until May 6, 2026, then credit-based pricing kicks in.

That’s the structural picture. Now here’s what it actually means for your week.

What’s actually different

Same-lane comparison. The two products share a lot of surface area, which is probably why OpenAI is positioning Workspace Agents as the successor rather than a new category.

Custom GPTWorkspace Agent
ReleasedNovember 2023April 22, 2026
Core ideaA saved ChatGPT persona with instructionsA saved agent that executes tasks across your tool stack
Can it send emails / post to Slack / update CRM?No — chat only (unless you build an Action via OpenAPI)Yes, with pre-built connectors and per-action approval
Schedules and recurring runs?NoYes — daily, weekly, custom
Memory across sessions?Limited to chat context within a single conversationPersistent, per-user, scoped folders
Plug-in connectors ready out of the boxNone without custom ActionsSlack, Salesforce, Gmail, Google Calendar, HubSpot, Gong, SharePoint, Notion, Atlassian, Microsoft 365, more
Underlying modelRegular GPT-4/5Codex agent model
Who can build themAnyone on Plus, Pro, Business, EnterpriseBusiness, Enterprise, Edu, Teachers only
Who can use themAnyone with the linkAnyone the agent is shared with — each user authenticates their own accounts
Approval gates on actionsN/A (no actions)Yes — risk-indicator approval before every sensitive step
CostIncluded in any paid planFree until May 6, 2026; credit-based pricing after
Future of formatDeprecating for orgs (date TBD); indefinite for individualsStandard going forward

Three things matter most from that table:

Workspace Agents do things; Custom GPTs talk about things. This is the whole reason for the upgrade. A Custom GPT for “Q3 Sales Outreach” drafts emails you still have to copy-paste into Gmail. A Workspace Agent for Q3 Sales Outreach reads Salesforce, drafts the email, asks you “send to Sally Chen now?”, and hits send when you click. Same concept, different muscles.

Workspace Agents are Business-and-up only. If you’re on ChatGPT Plus or Pro individually, you literally cannot access Workspace Agents today. The announcement confused a lot of Plus subscribers, who saw “new ChatGPT feature” and went looking for a Workspace Agents menu that isn’t there. It’s not a bug. It’s a plan gate. Individuals keep using Custom GPTs; organizations get the new format. This will probably change eventually, but not this quarter.

Authentication is different, and it’s better. Custom GPTs with custom Actions required you to hand the GPT an API key and hope. Workspace Agents authenticate per user per tool — Sally’s Salesforce login, not “the team’s.” Shared resources (like a team SharePoint folder) use a service account. That separation is how enterprise security reviews actually approve agents, and it’s why Workspace Agents are the format Fortune 500 buyers will accept.

Is ChatGPT an AI agent now, though?

The short answer: partially, in a specific mode, if you’re on the right plan.

ChatGPT itself, the base product, is still a chatbot. You open it, you type, it replies. Default behavior has not changed.

But three layers of “agent-ness” now live on top of it:

Agent Mode (a toggle inside ChatGPT) lets a single chat session take actions — browse the web, run code, use a toolbelt — within that conversation. It’s been around since mid-2025. It’s a per-chat capability.

Custom GPTs are a saved persona with optional Actions. The agent-ness is limited to whatever Actions you wired up yourself via OpenAPI. Most Custom GPTs don’t have Actions; they’re just saved prompts. So “is my Custom GPT an AI agent?” — probably not, functionally.

Workspace Agents are fully agentic, shareable, persistent, and scheduled. These are AI agents in the industry sense of the word: they perceive, plan, act, and remember, on their own cadence.

If someone asks “is ChatGPT an agent now?” the accurate answer is: ChatGPT has agentic features, and Workspace Agents are the first proper agent platform OpenAI has shipped. The chatbot is still a chatbot by default.

Should you build net-new Custom GPTs in 2026?

Depends on who you are.

If you’re an individual on Plus or Pro: Yes. Custom GPTs work, they’re not going away for you, and Workspace Agents aren’t available on your plan. Keep building them the way you have been.

If you’re on Business, Enterprise, Edu, or Teachers: Probably not. OpenAI is telling you directly that the format is on a deprecation path for your account type. There’s no published date, but the commercial pressure to convert is real — Workspace Agents are the ones getting new connectors, new model upgrades, and new capabilities. Custom GPTs are in maintenance mode.

The honest advice: for new work, build a Workspace Agent. For an existing Custom GPT that’s working fine, leave it alone and migrate it when the one-click conversion tool ships (likely Q2 or Q3 2026). Do not spend this week rebuilding what already works.

If your team has a library of 20–50 Custom GPTs: Inventory them, rank them by usage, and plan to migrate the top 5 in the first month after the conversion tool ships. Let the long tail wait until end of year. Most Custom GPTs are built and used by one person for one task; those can retire without anyone noticing.

What a Custom GPT → Workspace Agent migration actually looks like

OpenAI hasn’t shipped the conversion tool yet, but the shape is predictable. A Custom GPT has four ingredients: name, description, instructions, and knowledge files (optional Actions). A Workspace Agent has all of those plus connectors, skills, memory, and scheduling.

The conversion will transfer the first four cleanly. The last four are where the human decides:

  1. Connectors. Does this agent need to act in Slack, Salesforce, Gmail? Pick the connectors. Auth them.
  2. Skills. A skill is a reusable block of instructions (like “format output as a 3-bullet brief”). Most Custom GPTs have one or two de facto skills buried in their instruction blob — you extract those into named skills so they can be reused across agents.
  3. Memory. Do you want this agent to remember what it did last week? Toggle on persistent memory.
  4. Schedule. Should it run on a timer? Set the cadence.

The migration isn’t a copy-paste. It’s an upgrade opportunity. The best-performing Custom GPTs — the ones your team actually uses daily — are the ones worth carefully re-authoring as Workspace Agents rather than auto-migrating. The conversion tool will do the mechanical part; the judgment part is yours.

The Pro plan problem

Here’s the frustration that’s lighting up forums: Pro plan subscribers at $200/month do not get Workspace Agents. Business plan at $20/user/month (minimum 5 users per account) does.

If you’re a solo consultant, a freelancer, or a single-person shop, you’re in the gap: your plan is expensive enough to suggest “premium features,” but Workspace Agents specifically are gated to team accounts. OpenAI’s logic appears to be that Workspace Agents assume a team context — shared agents, service accounts, multi-user authentication. A Pro account is a single user.

What Pro subscribers can do right now:

  • Use Agent Mode inside a regular ChatGPT chat (you already have this).
  • Build Custom GPTs (also unchanged).
  • Sign up for a 1-seat Business account if you want Workspace Agents at $20/month — it’s cheaper than Pro and gets you the new format. The catch: some Pro-tier features (like unlimited GPT-5.5 Pro access) don’t transfer. You’d have both.
  • Wait. Workspace Agents may come to Pro at some point; OpenAI hasn’t committed either way.

For a solo operator, the honest play is the 1-seat Business trial. Build one Workspace Agent, see if the ROI is real, and decide. Two weeks of free access runs until May 6.

What to do this week

Individual ChatGPT users (Plus/Pro): Nothing changes. Your Custom GPTs still work. Optionally, add a 1-seat Business subscription to try Workspace Agents before the free window closes May 6.

Business/Enterprise admins: Inventory your team’s Custom GPTs. Don’t commission new ones. Start with a single Workspace Agent pilot — ideally something high-value your team uses weekly (sales reports, vendor onboarding, weekly newsletter drafting). Track the time saved. That number is what you’ll take to finance when the post-May 6 pricing lands.

Custom GPT builders: Your skills are portable. Instructions, output templates, tone rules — all of that translates. You’ll actually have more to work with in Workspace Agents because you can wire the output into a real tool action, not just a chat response. Start sketching what your top 3 Custom GPTs would look like with actual connectors attached.

Companies evaluating ChatGPT for their org: The $20/user Business tier just got a lot more interesting. Workspace Agents shift the ROI math from “chat assistant” to “team automation platform.” Evaluate accordingly. Compare with Microsoft Copilot Cowork (M365-integrated, different approach) and Claude Cowork (desktop, local). The three-way race is real.

What this means for you

If you built a Custom GPT your team actually uses daily: Good news — it’s not getting killed tomorrow. Bad news — the platform you built it on is in maintenance mode for your account type. Start planning the migration conversation, especially if your GPT includes an Action with an API key. That Action will become a proper connector in Workspace Agent format, which is more secure and more supportable.

If you’re one of the Plus users confused about why Workspace Agents aren’t showing up: You’re not crazy, and nothing is broken. They’re Business-plan-only. Either wait for OpenAI to expand access (no commitment yet) or switch to a Business subscription if you want them today.

If your company has been sitting out the ChatGPT-for-teams conversation: The May 6 free window is the cheapest time in the next year to evaluate. A pilot team can build 2–3 Workspace Agents and see real ROI in two weeks at zero marginal cost. If the agents work, the procurement conversation writes itself.

If you manage an AI tools budget: Factor in credit-based pricing post-May 6. OpenAI hasn’t published exact rates, but expect consumption-based billing similar to the API (pennies per run, but they add up on high-frequency agents). Forecast accordingly.

The bottom line: Custom GPTs are the past tense of OpenAI’s team product. Workspace Agents are the present tense. The old ones keep working; the new ones are getting the features. If you’re on an account type that can build Workspace Agents, that’s where your next agent should live. If you’re on an account type that can’t, keep doing what you’re doing — your day hasn’t changed.

Who should use which

Use Custom GPTs if:

  • You’re on ChatGPT Plus or Pro (individual)
  • You want a saved persona for personal tasks (writing helper, coding buddy, research assistant)
  • You don’t need the agent to do things, just respond consistently

Use Workspace Agents if:

  • You’re on Business, Enterprise, Edu, or Teachers
  • You want the agent to act in Slack, Salesforce, Gmail, or similar tools
  • The task is recurring and could run on a schedule
  • Multiple teammates will use or trigger the same agent

Use neither, use Agent Mode, if:

  • You want one-off agentic tasks inside a single chat (research, computer use, coding)
  • The task is ad-hoc, not saved for future reuse

The bottom line

OpenAI made a clean bet: autonomous agents that act in your tools are the future of ChatGPT for teams, and Custom GPTs were a useful stepping-stone that doesn’t carry forward. The April 22 launch is the first real move of that bet. The two-week free window is OpenAI’s way of saying: try this before the pricing math makes you think twice.

If you’re a sales lead, an operations manager, an HR director, or anyone on a Business plan with a repeating task that touches multiple tools — this was the launch that mattered this week. Your Custom GPT library isn’t dying tomorrow, but it’s not the thing you’ll be building in six months either.

Start one Workspace Agent today. Pick something small. See what it costs you in attention to get it running. See what it saves you in a week. Make the migration call with that data in hand, not the hype.


Sources:

Build Real AI Skills

Step-by-step courses with quizzes and certificates for your resume