OpenAI Workspace Agents Just Replaced Custom GPTs — And Plugged ChatGPT Directly Into Slack, Gmail and Salesforce

Apr 22 2026 — Workspace Agents launch in ChatGPT Business / Enterprise / Edu  ·  3 — Anchor integrations at launch: Slack, Google Workspace, Salesforce  ·  May 6 2026 — Date credit-based pricing replaces the free preview window  ·  40%+ — Share of OpenAI revenue now coming from enterprise — and rising

On April 22 2026, OpenAI launched Workspace Agents inside ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, and the Teachers plan — the official successor to Custom GPTs and, in plain terms, the moment ChatGPT crossed the line from 'helpful chatbot teams use' to 'autonomous worker that lives inside the company's tool stack.' Workspace Agents ship with native, first-party integrations for Slack, Google Workspace (Gmail, Drive, Calendar, Docs, Sheets), and Salesforce. They run in the cloud, persist their state across sessions, can be shared and improved across a team, and — critically — they keep working even when nobody is in front of a screen.

The launch is free in research preview until May 6 2026, after which it shifts to credit-based pricing. For UK CTOs, COOs, RevOps leaders, and anyone responsible for how AI is integrated into the daily tool stack, this is the most consequential single AI integration release of the year so far. Custom GPTs were a clever experiment; Workspace Agents are infrastructure. Here is exactly what changed, what they unlock, and how to plan your 30-day rollout.

What Changed: Custom GPTs vs Workspace Agents

• Custom GPTs were chat-only — Workspace Agents take real actions in real systems (send messages, create records, update calendars, file tickets) with proper authentication and audit trails.
• Custom GPTs were single-session — Workspace Agents persist state across sessions and across users, so a team can build a shared agent that gets smarter the more the team uses it.
• Custom GPTs were synchronous — Workspace Agents run asynchronously in the cloud, so they can complete long-running work (overnight reports, multi-step research, scheduled outreach) without a human supervising.
• Custom GPTs lived inside ChatGPT — Workspace Agents live inside Slack as well, with @-mention triggers, thread-aware conversation memory, and native channel ownership of repeating workflows.
• Custom GPTs had no formal permission model — Workspace Agents have proper enterprise governance: who can use which agent, what tools each agent can access, what approvals are required for sensitive actions, and full audit logs of every action taken.

The Three Integrations That Matter Most at Launch

Slack — Agents That Live in Channels

The Slack integration is the most heavily promoted at launch and, for a lot of UK businesses, the single most consequential change in how teams will use ChatGPT day to day. Workspace Agents can be installed into Slack channels, respond to @-mentions in threads, follow conversations, take ownership of recurring tasks (e.g. 'this agent owns the daily standup summary'), and execute multi-step workflows that span Slack and the connected tools. The behaviour change is significant: instead of teams 'switching to ChatGPT' to do AI work, AI work happens where the team already is.

Google Workspace — Native Gmail, Calendar, Drive, Docs, Sheets

The Google Workspace integration covers the full daily-work stack for the majority of UK SMBs and a substantial slice of enterprises: Gmail for outbound and triaged inbound mail, Calendar for scheduling and prep, Drive for file access, Docs and Sheets for document and spreadsheet creation. Workspace Agents can read context from these surfaces, take action against them, and orchestrate work that spans them — all under proper enterprise OAuth scopes and audit logging. For functions like SDR motion, exec admin support, finance close, and content production, this is a step change in what 'AI integration' means in practice.

Salesforce — RevOps Workflows in the Native CRM

The Salesforce integration matters specifically for revenue operations. Workspace Agents can read Account, Contact, Opportunity, and Activity records; create and update them; and run multi-step RevOps workflows like account research, meeting prep, account-plan generation, and opportunity-stage hygiene. For UK B2B businesses, this is the integration that makes Workspace Agents directly relevant to the GTM stack — and where the productivity gains in the next 90 days are likely to be most visible.

Five Workspace Agents Worth Standing Up in the First 30 Days

1. The SDR Pre-Meeting Prep Agent

Triggered by upcoming Salesforce meetings, this agent runs account research, pulls relevant news, summarises recent activity, drafts a tailored agenda, and posts the prep brief into the relevant Slack channel an hour before the meeting. Replaces 30–60 minutes of human prep per meeting. The single highest-ROI agent for most B2B sales orgs.

2. The Inbox Triage Agent

Sits on Gmail, classifies inbound mail, drafts responses for routine messages, escalates anything that needs human judgement, and produces a daily 'three things you actually need to read' digest in Slack. Frees roughly 45 minutes of executive time per day with a robust enough confidence model to avoid mis-routing.

3. The Standup Summariser

Lives in your engineering or ops Slack channel, ingests the team's standup updates, reconciles them against open tickets and Drive docs, and posts a clean daily summary plus a 'risks and blockers' note for the team lead. Replaces the daily admin overhead of standup cross-referencing.

4. The Quarterly Business Review (QBR) Builder

Connected to Salesforce and Google Sheets, this agent assembles the customer-facing QBR deck — pulling activity history, usage metrics, NPS scores, recent tickets, and exec-relevant updates — and produces a draft slide deck the day before the QBR meeting. Saves the customer success team 4–8 hours per QBR.

5. The Recruiter Co-Pilot

Watches a Drive folder of incoming CVs, screens against a defined rubric, drafts shortlisting notes, books interview slots via Calendar, and drafts feedback messages after interviews. Materially compresses the time-to-shortlist for high-volume hiring funnels.

The Pricing Cliff on May 6 — And How to Plan For It

Workspace Agents are free in research preview until May 6 2026, after which OpenAI shifts to credit-based pricing. The exact pricing curve is still being communicated to enterprise customers, but the right planning posture is to assume meaningful per-agent and per-action costs once the meter starts. Two practical implications follow from this.

• Use the free preview window deliberately. Stand up your top 3–5 candidate agents now, while the cost of experimentation is zero, and let the team build muscle on what works before the meter starts.
• Architect for cost observability. Every agent action should be logged with a cost-attribution tag from day one — by agent, by user, by team, by use case — so you can rationalise the agent portfolio cleanly when pricing kicks in. Doing this retroactively is much harder than doing it on day one.

Workspace Agents vs Microsoft Copilot vs Google Gemini Enterprise — Quick Decision Framework

Workspace Agents do not exist in a vacuum. Microsoft Copilot Studio (with the Copilot Pages and Agents in Microsoft 365), Google Gemini Enterprise (with Agent Space and the Vertex AI agent platform), and Adobe CX Enterprise Coworker (for marketing) are all credible alternatives or complements. The right choice — and increasingly, the right combination — depends on three specific questions about your existing estate.

1. Where does your team's daily work live? If it is overwhelmingly in Microsoft 365 (Outlook, Teams, OneDrive), Copilot is the path of least integration friction. If it is overwhelmingly in Google Workspace and Slack, Workspace Agents are the natural fit. If it is split, expect to run both side by side — and that is fine.
2. What is your CRM? Salesforce-resident orgs get a meaningful immediate productivity uplift from Workspace Agents specifically. HubSpot-resident orgs may want to lead with Copilot (which has a strong HubSpot connector) or wait for the OpenAI HubSpot integration.
3. What is your model preference? Copilot is OpenAI-anchored, Gemini Enterprise is Google-anchored, Adobe CX Enterprise Coworker is multi-model. Workspace Agents are OpenAI-anchored but increasingly speak MCP, which mitigates lock-in. If multi-model flexibility is high priority, weigh accordingly.

OpenAI Frontier Alliances and Why Integration Will Get Easier

Workspace Agents are not a standalone launch. They are part of OpenAI's broader 'Frontier Alliances' programme — formalised partnerships with McKinsey, BCG, Accenture, Capgemini, AWS, Databricks, and Snowflake — designed to make it dramatically easier for enterprises to integrate OpenAI's intelligence into their existing data and infrastructure stack. For UK enterprises wrestling with how to land Workspace Agents on top of legacy data warehouses, regulated workflows, and tightly-governed identity systems, the Frontier Alliances ecosystem is now genuine professional-services capacity rather than scattered consulting. That accelerates the deployment curve materially.

Sources

1. OpenAI — Introducing Workspace Agents in ChatGPT (April 22 2026)
2. OpenAI Cookbook — Building Workspace Agents in ChatGPT to Complete Repeatable, End-to-End Work
3. OpenAI Academy — Workspace Agents
4. VentureBeat — OpenAI Unveils Workspace Agents, a Successor to Custom GPTs (April 22 2026)
5. Reworked — OpenAI Replaces Custom GPTs With Workspace Agents Built for Team Workflows
6. TechBuzz AI — OpenAI Launches Workspace Agents for Enterprise Automation
7. OpenAI Help Center — ChatGPT Business Release Notes (April 2026)