
OpenAI Just Put Autonomous Agents Inside Slack and Salesforce. The Free Trial Ends in 10 Days.
Workspace Agents land inside ChatGPT Business with Codex under the hood. Free until May 6, then credit-based. Adobe and Google just got punched in the face.
The AI Post newsroom — delivering AI news at the speed of intelligence.
OpenAI quietly pushed a research preview this week that almost nobody outside the AI Twitter sphere has properly registered. Workspace Agents are now live inside ChatGPT Business, Enterprise, Edu, and Teachers. They run on Codex. They plug into Slack, Google Drive, Microsoft 365, Salesforce, and Notion. They execute long-running, multi-step workflows on a schedule or on a trigger. Free until May 6. Credit-based pricing kicks in after.
This is the same week Adobe shipped CX Enterprise Coworker. The same month Google launched Gemini Enterprise Agent Platform with a $750M partner fund. The orchestration layer war is no longer theoretical. Three of the top five enterprise software vendors have now staked a claim, and OpenAI just put the fastest, most consumer-flavored version directly into the ChatGPT product 800 million people already use.
What it actually does
A workspace agent is a long-running ChatGPT instance you describe in plain English. Map a workflow, connect the tools, set the trigger. The agent runs against a schedule or a webhook, plugs into your business systems, and reports back to a shared channel. Approval checkpoints kick in for sensitive actions like sending money or deleting data.
Examples OpenAI is seeding: a sales agent that watches new HubSpot leads, enriches them with public data, drops a summary into a Slack channel, and queues a draft email for review. A finance agent that pulls Stripe data nightly, reconciles against Salesforce, and flags anomalies. A research agent that watches a list of competitor blogs and posts daily summaries.
These are exactly the workflows Zapier, Make, and n8n have owned for a decade. The difference is that the workflow is now described in English, not built in a flowchart UI, and the agent can adapt mid-run if a step fails. Codex handles the code execution, tool calling, and recovery.
The pricing trap
Free until May 6, 2026. After that, credit-based pricing. OpenAI's own published guidance says Codex usage averages $100 to $200 per developer per month at typical workloads. Workspace Agents will sit on the same meter.
OpenAI is also running a Codex for Business promotion offering $500 in credits per workspace, with $100 per new seat that sends its first message. Translation: the company wants you to deploy these things into your team this week, get used to them, hit your free quota, and then write a check before you can rip them out without team revolt.
Classic land-and-expand. Same playbook Slack ran on Microsoft. Same playbook Notion ran on Confluence. The free window is the install. The pricing model is the lock.
Who gets crushed
Zapier and Make are obvious targets. Both companies are valued in the billions on the assumption that workflow automation is a moat built from integrations, not models. OpenAI just published native connectors to the same set of apps and pointed Codex at them. Zapier's response will be to lean harder on the long tail of integrations, where it has thousands and OpenAI has dozens. That is a temporary moat.
Adobe's CX Enterprise pitch is more vertical, more brand-managed, and aimed squarely at the CMO suite. Salesforce's Agentforce is locked to Salesforce data. Microsoft's Copilot suite is M365-centric. OpenAI's pitch is the only one that runs anywhere your team already lives, including Microsoft and Google's own properties. That cross-stack neutrality is the most disruptive piece.
The other casualty: custom GPTs. They had a moment in 2024. They never produced real revenue. Workspace Agents are the upgrade path, and OpenAI is treating them as the successor product.
The pre-IPO read
OpenAI told shareholders earlier this month it expects to hit 30 gigawatts of compute by 2030. Anthropic's revenue tripled to $30B in four months. Both companies need enterprise spend, and both need a story for bankers about what comes after the chatbot interface.
Workspace Agents are that story for OpenAI. They convert ChatGPT from a $20-per-month consumer tool into a per-seat, per-credit, integrated-into-everything enterprise product. If they take, the ARR multiplier is enormous. If they don't, OpenAI walks into the IPO with the same single-product-with-enterprise-tier story Anthropic has, except with worse safety brand.
The next 10 days are the test. Free trial, $500 in credits, every connector you can imagine. If your team uses ChatGPT Business already, expect somebody on it to ship a workspace agent this weekend. By May 6 the bills will be live, and the question of whether OpenAI just bought itself a real second product line will start to have a real answer.
Sources: OpenAI Workspace Agents announcement, VentureBeat enterprise analysis, Geeky Gadgets product breakdown, Let's Data Science integrations detail, OpenAI Codex pricing pages, Ubergizmo coverage.