ChatGPT’s New Workspace 24/7 AI Agents is INSANE
OpenAI launched ChatGPT Workspace Agents on April 22nd, 2026, transforming ChatGPT from a conversational tool into a 24/7 autonomous team member powered by the Codex model. These agents run in the cloud, integrate with Slack, retain memory, and can execute real business workflows without constant human oversight. The video covers how to build them, real use cases, competitor comparisons, and practical tips for implementation.
Summary
The video introduces ChatGPT's new Workspace Agents, framing them as a major evolution beyond ChatGPT's previous chat-only functionality. Launched on April 22nd, 2026, and powered by OpenAI's Codex model, these agents run entirely in the cloud, meaning they continue working even when a user's computer is off. Unlike Custom GPTs, which were limited to conversation, Workspace Agents can take real actions across connected tools and run continuously on a schedule or in response to triggers.
The presenter outlines four core differentiating features: they operate 24/7 without requiring the user's device to be active; they integrate natively into Slack, allowing teams to drop agents into channels where they handle requests automatically; they have persistent memory that improves based on team corrections over time; and they include a human approval checkpoint before executing sensitive actions like sending emails or editing spreadsheets. A notable security feature highlighted is built-in protection against prompt injection attacks, described as critical for teams handling real business data.
OpenAI highlighted four specific real-world agent workflows already in use: a software request reviewer that triages, checks policy, routes approvals, and opens IT tickets; a product feedback router that aggregates input from multiple channels and produces weekly action reports; a weekly metrics reporter that auto-generates charts, narratives, and business reports every Friday; and a lead outreach agent that researches, scores, drafts emails, and updates CRM systems — with OpenAI's own sales team reportedly using this internally.
The video walks through a five-step no-code setup process: accessing the agents tab in ChatGPT, describing the workflow in plain language, letting ChatGPT map out the steps and connect tools, selecting a trigger (schedule or Slack-based), and adding guardrails to restrict sensitive actions. The presenter emphasizes this can be completed in minutes without engineering support.
A competitive comparison covers Anthropic's Claude (strongest for deep coding and long-context reasoning), Microsoft Copilot Studio with Agent 365 (best for Microsoft 365 enterprise environments), Google's Gemini Enterprise Agent Platform (notable for supporting 200+ models including Claude), and Salesforce AgentForce 360 (purpose-built for sales and CRM teams). ChatGPT Workspace Agents are positioned as the best fit for marketing, sales, and HR teams already using ChatGPT, with the Slack integration cited as a major adoption advantage, though the trade-off is being locked into OpenAI's Codex model.
Additional use cases are presented for HR (onboarding automation), marketing (brand mention monitoring), customer support (ticket drafting and routing), and project management (daily stand-up generation). The presenter closes with five practical tips: start with a single boring task, always add human approval for sensitive actions, use Slack triggers for adoption, test with messy inputs, and treat the agent like a new employee requiring ongoing feedback. The agents are currently in research preview, free on Business, Enterprise, Edu, and Teachers plans until May 6th, 2026, after which a credit-based pricing model takes effect.
Key Insights
- The presenter states that ChatGPT Workspace Agents have built-in protection against prompt injection attacks — where malicious content in emails or webpages tries to hijack the agent — describing this as a significant security feature OpenAI built in from day one for teams handling real business data.
- OpenAI's own product team is already running a Workspace Agent inside a Slack channel that answers employee questions, links to documentation, and files tickets when it spots a new issue — presented as a real internal deployment, not a demo.
- The presenter argues ChatGPT Workspace Agents are best positioned for marketing, sales, and HR teams already living inside ChatGPT, citing a near-zero adoption curve and strong Slack integration, but notes the trade-off is being locked to OpenAI's Codex model unlike Gemini which supports 200+ models.
- OpenAI's own sales team is reportedly already using the lead outreach agent internally to qualify leads — researching inbound leads, scoring them, drafting follow-up emails, and updating the CRM — suggesting these are validated internal workflows, not theoretical use cases.
- The presenter warns that Workspace Agents are currently free during research preview on Business, Enterprise, Edu, and Teachers plans, but OpenAI will switch to a credit-based pricing model after May 6th, 2026, recommending eligible users test immediately while access is free.
Topics
Full transcript available for MurmurCast members
Sign Up to Access