ChatGPT just added workspace agents - Codex-powered automation that runs entirely in the cloud and handles multi-step tasks across your connected tools without you managing each step.
The agents are built on Codex, OpenAI's model trained specifically on code and programming tasks. That background matters here: Codex understands how software tools communicate with each other, which means the agents can navigate complex integrations - pulling data from one tool, processing it, and pushing results to another - rather than issuing a single command and stopping.
This is a meaningful shift from existing ChatGPT features. Previous automations required staying in the loop, approving each action or re-prompting when something broke. Workspace agents are designed to run a full workflow end-to-end. You hand off a task, it executes across multiple steps, and returns with results.
What This Replaces
For teams, the pitch centers on scope. One person can describe a workflow in plain English - say, pulling weekly metrics from analytics, summarizing them, and drafting a report - and the agent handles execution. That positions ChatGPT in direct competition with dedicated automation tools like Zapier or Make, where the same workflow requires building a flowchart of nodes and conditions instead of writing a sentence.
The "securely" language in OpenAI's announcement addresses the obvious concern: when an agent has access to your business tools, the scope of what it can read, write, and modify matters. Details on permission controls and audit logs will determine how comfortable IT teams are handing over the keys.
The Reliability Question
Autonomous agents that run multiple steps without human checkpoints fail in ways that are hard to catch until the damage compounds. A misunderstood instruction in step one affects every step that follows. How OpenAI handles mid-workflow errors - whether the agent stops and asks, or keeps going on a wrong assumption - will decide whether workspace agents become a daily driver or a feature people try once.
No pricing details have been published yet. Given that Codex usage typically runs on higher-tier plans, expect workspace agents to live behind Team or Enterprise subscriptions.