OpenAI launched workspace agents in ChatGPT on 22 April, marking the most significant shift in the product's architecture since the introduction of custom GPTs. Workspace agents are powered by Codex and run in the cloud, meaning they continue working even when users are offline. Unlike custom GPTs — which respond to individual prompts within a single conversation — workspace agents are designed for teams and can handle complex, multi-step workflows autonomously. They write and execute code, access files and tools, maintain persistent memory across sessions, pull context from multiple systems, and request human approval for sensitive actions like sending emails or modifying calendar entries. OpenAI described the feature as 'an evolution of GPTs' rather than a replacement, with existing GPTs remaining available and conversion tools forthcoming.
The initial integration is with Slack, where workspace agents can autonomously pick up tasks, respond to queries, and execute workflows without requiring users to open ChatGPT directly. Teams can build an agent once, share it across the organisation, and improve it over time through corrections mid-conversation — effectively creating institutional knowledge that persists and improves with use. The agents can also run on schedules or continuously, enabling background automation for recurring tasks like report preparation, code review, and data analysis. The research preview is available on ChatGPT Business, Enterprise, Edu, and Teachers plans, with all usage free until 6 May 2026, after which credit-based billing begins.
For context engineers, workspace agents represent OpenAI's answer to the agentic infrastructure race that Anthropic entered with Claude Code Routines and Google expanded at Cloud Next with Agentspace. The strategic bet is clear: ChatGPT is no longer positioning itself as a chatbot that answers questions but as a team automation platform where AI agents operate as persistent, autonomous colleagues. The Codex backbone means these agents inherit the full software engineering capabilities of OpenAI's coding models, while the Slack integration ensures they sit inside existing team communication workflows rather than requiring users to adopt a new tool. The credit-based pricing model — replacing the flat subscription approach — also signals that OpenAI expects these agents to handle substantial workloads that justify usage-based billing, which is a fundamentally different commercial model from selling chatbot subscriptions.
