OpenAI Workspace Agents
OpenAI Workspace Agents Replace Custom GPTs for Enterprise Teams
OpenAI launched Workspace Agents on April 22 — Codex‑powered AI agents that run in the background, connect to Slack and Salesforce, and operate on schedules. They replace Custom GPTs with team‑owned, always‑on automation. Here's what builders need to know.
What Workspace Agents Actually Do
OpenAI launched Workspace Agents on April 22, 2026 — Codex‑powered AI agents within ChatGPT that automate complex, repeatable, multi‑step workflows across enterprise tools and teams. Unlike regular ChatGPT, they run in the cloud and continue working even when the user is offline, per OpenAI's blog.
Each agent has three components: a Trigger (schedule or manual run), a Process with Skills (reusable open‑source packages based on the agent skills standard at agentskills.io), and Tools and Systems (approved integrations like Slack, CRM, and internal docs). No code is required — users describe the job in plain English through a conversational builder, per VentureBeat.
Why Custom GPTs Had to Die
Custom GPTs failed as enterprise primitives for three reasons: they were tied to a single user (not team‑owned), they could not write back to external systems reliably, and they had no meaningful admin layer, per AI Automation Global. Workspace Agents fix all three:
- Team ownership One edit updates for everyone — no more 10 people building 10 versions of the same workflow
- Deep tool access Native writes to Slack, Salesforce, Google Workspace, plus custom MCP servers — not shallow read‑only integrations
- Admin controls RBAC, Compliance API, and approval gates — agents can request human sign‑off before sending emails or editing spreadsheets
- Background execution Schedulable with cron‑style wake‑ups; runs in Codex cloud sessions even when you're offline
Integrations and Pricing
At launch, Workspace Agents ship with 60+ enterprise connectors and 90+ new plugins covering Slack, Google Workspace (Gmail, Drive, Calendar, Docs, Sheets), Salesforce, Notion, Atlassian Rovo, CircleCI, GitLab, Neon by Databricks, and Render. Custom MCP servers can wire proprietary systems, per AI Automation Global. Notably, Microsoft 365 integrations are listed as "in development" — SharePoint works at launch, but Teams, Outlook, Word, and Excel are not yet available.
Workspace Agents are free during the research preview until May 6, 2026. After that, pricing shifts to credit‑based pay‑per‑use with no minimum commitments. The per‑credit price has not been published. Availability is limited to ChatGPT Business ($20/user/month), Enterprise, Edu, and Teachers plans — not Plus, Pro, or Free tiers, per Pillitteri. Custom GPTs for business accounts are being deprecated with a reported shutdown date of August 26, 2026.
The Competition: Microsoft, Google, Anthropic
The timing is not coincidental. Microsoft launches Agent 365 on May 1 at $15/user/month, while Google announced its Gemini Enterprise Agent Platform the same day as OpenAI's launch. Here's how they stack up:
| Factor | OpenAI Workspace Agents | Microsoft Agent 365 | Google Gemini Agents |
|---|---|---|---|
| Philosophy | Dynamic exploration; agents try new paths | Governance‑first; tight guardrails | Platform‑first; IT‑focused |
| M365 integration | SharePoint only at launch | Full native (Teams, Outlook, etc.) | Native Google Workspace |
| Distribution | ChatGPT Business/Enterprise | 350 million MAU via Copilot | Google Cloud enterprise |
| Target user | Non‑technical teams | Knowledge workers in M365 | IT and technical teams |
| Pricing | Credit‑based (TBD) | $15/user/month | Cloud consumption |
Microsoft's advantage is distribution, not model quality, per Business20Channel. Google Cloud CEO Thomas Kurian's pitch, as Google's blog put it: "Rival vendors are handing you the pieces, not the platform."
Beta Results and Early Signals
Beta customer Rippling reports that a sales consultant built a sales opportunity agent with zero engineering support. It researches accounts, summarizes sales calls, and posts deal briefs automatically — saving reps 5–6 hours per week, per ChatlyAI. Gartner survey data shows 79% of enterprises have at least one AI agent in production, with a median of 3 frameworks per team.
But the duplicated‑work problem persists: 10 people building 10 different AI workflows for the same task, 3 teams maintaining overlapping agent prompts, nobody knowing which is the source of truth. As The New Stack noted: "Organizations struggle to find ROI on AI not because the models are unusable, but because the work stays fragmented."
Box CEO Aaron Levie posted on X that this is "probably the biggest news yet in software going headless."
"Probably the biggest news yet in software going headless — agents have access to any of the tools and data you want to work with, with complete coding and tool use available to them."
CISO Caveats and What to Watch
Two warnings from AI Automation Global: Workspace Agents are not available for EKM‑enabled Enterprise accounts at launch — a meaningful gap for regulated industries. And the default is "on" for Business plans — IT teams on Business tier should audit tenants immediately to avoid shadow deployments.
For builders, the key question is whether credit‑based pricing will penalize high‑volume automation. If costs scale linearly with adoption, the economics could favor sporadic use over the always‑on workflows that make agents genuinely valuable, per Pillitteri. And for teams already running deterministic ETL or webhook‑driven pipelines, tools like n8n still win — most mature stacks run both side by side, as AI Automation Global noted.
What's coming: automated triggers, performance dashboards, expanded integrations, Workspace Agent support inside the Codex app, and an admin console for viewing all agents across an organization.
Apr 26, 2026
Perplexity AI Hit With Privacy Lawsuit While Appeals Court Greenlights Its Bots
Perplexity faces a class-action lawsuit for allegedly sharing user chat data with Google and Meta through hidden trackers. Meanwhile, the Ninth Circuit has paused an injunction blocking its Comet shopping bot from Amazon — setting the stage for a landmark ruling on whether AI agents need user permission or platform permission to operate.
Apr 26, 2026
Anthropic Mythos AI Found 2,000+ Vulnerabilities and Sparked a Global Scramble
Anthropic's Claude Mythos Preview found over 2,000 zero-day vulnerabilities in seven weeks, including bugs dating back 27 years. The model is too dangerous for public release — but a Discord group already leaked it, and governments worldwide are racing to respond.
Related News
Apr 26, 2026
OpenAI Knew About Shooting Suspects and Said Nothing. Now Altman Is Sorry
OpenAI flagged and banned a mass shooter's ChatGPT account eight months before the Tumbler Ridge attack but chose not to alert police. After a second shooting involving ChatGPT advice, Florida has opened a criminal investigation. The era of AI companies operating without a duty to report is ending.
Apr 26, 2026
Musk v. OpenAI Trial Begins After Fraud Claims Dismissed
Elon Musk's fraud claims against OpenAI were dismissed on April 25, but the trial proceeds Monday on trust and enrichment claims that could reshape AI's corporate governance. Here's what builders need to know.
Apr 26, 2026
Anthropic's Project Deal: AI Agents Trade Real Goods and the Losers Can't Tell
Anthropic ran a classified marketplace where 69 AI agents traded real goods for a week. Stronger models consistently got better deals — and users on the losing side couldn't perceive the disadvantage. The implications for agent commerce are sobering.