OpenAI debuts ChatGPT Work, an agentic tool for automating business workflows

· Source: AI – SiliconANGLE · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Intermediate, extended

Summary

OpenAI launched ChatGPT Work and the GPT-5.6 model family on July 9, 2026, expanding its AI capabilities for business workflows. ChatGPT Work is an "agentic" tool integrated into ChatGPT, enabling autonomous actions across connected applications, files, and web tools, capable of generating spreadsheets, presentations, and web applications from a single prompt. It is powered by the new GPT-5.6 model, which offers a 54% improvement in token efficiency for agentic coding tasks compared to GPT-5.5, reducing operational costs. The GPT-5.6 family includes Sol for advanced reasoning, Terra for enterprise workloads, and Luna for high-volume, low-cost applications. OpenAI also introduced "Sites" for transforming work materials into interactive websites and an updated ChatGPT desktop app with integrated browser functionality. The company emphasized safety, detailing 700,000 A100 equivalent hours of red teaming and a U.S. government review for GPT-5.6.

Key takeaway

For AI Product Managers evaluating new automation solutions, OpenAI's ChatGPT Work and GPT-5.6 offer significant advancements. You should explore its agentic capabilities for automating complex, multi-application workflows, noting the 54% token efficiency improvement in GPT-5.6 for potential cost savings. Consider piloting ChatGPT Work for tasks like content generation or data analysis, leveraging its enterprise controls to manage AI agent access and sensitive actions.

Key insights

OpenAI's new agentic ChatGPT Work, powered by GPT-5.6, automates complex business workflows and content creation across diverse applications.

Principles

Method

ChatGPT Work uses plugins to connect with diverse applications, allowing users to invoke specific tools via "@" commands for autonomous, multistep task execution, with human approval for sensitive actions.

In practice

Topics

Best for: CTO, VP of Engineering/Data, Product Manager, Director of AI/ML, AI Product Manager, Automation Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI – SiliconANGLE.