OpenAI Revolutionizes ChatGPT with Autonomous Work Agents

OpenAI Introduces Autonomous Work Agents in ChatGPT

OpenAI has recently launched a new feature for ChatGPT: autonomous work agents. These agents, powered by Codex, go far beyond the capabilities of custom GPTs by handling complex, multi-step team tasks. Unlike custom GPTs that responded to individual queries, these new agents manage multi-step tasks for teams.

These agents operate in a dedicated workspace, with access to files, code, tools, and persistent memory. They can integrate information from multiple systems, request approvals, and directly integrate with Slack. Existing custom GPTs will remain available for the time being, and a tool to convert GPTs into work agents is currently in development.

A Major Advancement for GPTs

OpenAI views this innovation as a significant evolution of GPTs. Operating on Codex in the cloud, the agents continue their work even when the user is offline. The work agents are designed to leverage the context of various systems, track a team's workflow, and advance tasks across multiple tools.

Each agent has its own workspace, with the ability to write and execute code, use connected applications, and retain information between sessions. Teams can also program them or connect them to Slack, where they autonomously retrieve requests.

Quick Setup and Ready-to-Use Templates

Setting up a work agent is simplified through ready-to-use templates. Users can describe a recurring workflow or upload a file, and ChatGPT takes care of transforming it into an agent, defining the steps, connecting the tools, adding skills, and testing the outcome. OpenAI offers templates for various sectors such as finance, sales, and marketing.

OpenAI has already shared some examples functioning internally. A "Software Reviewer" checks employee software requests against approved tools and policies, creating IT tickets if necessary. A "Product Feedback Router" monitors Slack, support channels, and public forums, turning feedback into priority tickets and weekly summaries. A "Weekly Metrics Reporter" extracts data every Friday, creates charts, and shares a report with the team.

A "Prospecting Agent" searches for incoming leads, evaluates them, drafts personalized follow-up emails, and updates the CRM. OpenAI's sales team uses an agent that extracts details from call notes and account research, qualifies new leads, and deposits drafts of follow-up emails into representatives' inboxes. The accounting team has built an agent that manages parts of the monthly close, from journal entries to balance sheet reconciliations and variance analyses.

Centralizing Knowledge in Reusable Workflows

The work agents aim to centralize scattered knowledge within companies into reusable workflows. With their memory and continuous improvement capabilities, agents can be refined and shared within teams. Work agents can be managed and discovered via a new Agents tab in the ChatGPT sidebar.

Users can interact with these agents through ChatGPT and Slack, with additional integrations coming soon. Administrators can define which tools and data are accessible by the agents, and built-in protections prevent injection attack vulnerabilities. Sensitive actions like sending an email or creating a calendar entry may require approval. Agents can be monitored via the compliance API and paused if something goes wrong. More details are available in the OpenAI Academy and Help Center.

Availability and Terms of Use

Currently in Research Preview, these agents are available for ChatGPT Business, Enterprise, Edu, and Teachers plans. They are free until May 6, 2026, after which a credit-based billing system will be introduced. Administrators of Enterprise and Edu plans have role-based controls to manage the creation and sharing of agents.