OpenAI Launches GPT-5.6 and ChatGPT Work to Automate Complex Workflows

OpenAI has officially entered the era of autonomous agents with the public rollout of GPT-5.6 and the introduction of ChatGPT Work. This move transitions ChatGPT from a conversational chatbot into a proactive worker capable of managing entire end-to-end projects across multiple software environments.

The Evolution from Chatbot to Autonomous Agent

The core of this launch is ChatGPT Work, a new product available for mobile, web, and desktop that integrates Codex technology directly into the ChatGPT ecosystem. Unlike standard LLM interactions that require step-by-step prompting, ChatGPT Work is designed to operate across various apps and files for extended periods. It can ingest a single prompt and deliver finished, professional outputs—such as completed Excel spreadsheets or Word documents—without constant human intervention.

This development suggests a strategic merger of ChatGPT and Codex into a unified "super app." By bringing Codex's specialized capabilities to general knowledge workers, OpenAI is positioning itself to compete directly with Anthropic’s recent launch of "Cowork," signaling a massive arms race in the AI agent space.

Unified Plugins and Third-Party Integration

A major hurdle for OpenAI’s previous plugin ecosystem in 2023 was the lack of model maturity. With the deployment of GPT-5.6, OpenAI is attempting to fix this through the new "Unified Plugins Directory." This directory provides a centralized hub for third-party integrations, allowing the agent to access data from:

  • Productivity & Communication: Google Drive, SharePoint, Slack, Microsoft Teams, Gmail, Outlook, and Zoom.
  • CRM & Professional Tools: Salesforce, LinkedIn, and GitHub.
  • Creative & Storage: Adobe, Canva, and Dropbox.

Users can explicitly trigger these tools using "@" mentions or allow the GPT-5.6 model to autonomously determine which data source is required to complete a task, such as transforming raw customer research into a full marketing campaign brief.

Security, Deployment, and Usage-Based Billing

As these agents gain the ability to execute actions and access sensitive company data, security is paramount. OpenAI has introduced an "Auto-Review" feature, where advanced models act as a secondary layer to vet critical actions before they are executed. During adversarial red-teaming, OpenAI reported that this feature blocked 100% of attempts to extract protected data.

The rollout is phased, prioritizing Pro, Enterprise, and Edu users on web and mobile, while the ChatGPT desktop app for Mac and Windows is immediately available to all tiers, including free users.

Billing for ChatGPT Work departs from standard chat subscriptions. It utilizes an agent-based consumption pool shared with Codex and Workspace Agents. The cost of a task is dynamic, scaling based on its complexity, size, and the specific model used. Enterprise admins are provided with "Spend Controls" to manage these costs at the workspace or individual user level.

Key Takeaways

  • Agentic Workflow: ChatGPT Work uses GPT-5.6 and Codex to move beyond simple chat, enabling autonomous execution of long-term, multi-step professional projects.
  • Deep Integration: The Unified Plugins Directory connects the agent to essential enterprise tools like Salesforce, Slack, and Google Drive for seamless data handling.
  • New Economic Model: Moving away from flat-rate chat, OpenAI is implementing a usage-based consumption model tailored to the computational intensity of autonomous tasks.