OpenAI’s new AI agent targets everyday workflows

68

TL;DR:

OpenAI has launched ChatGPT Work, an AI agent that can complete workplace tasks across apps, files and workflows. It matters because enterprise AI is shifting from answering questions to producing finished work, raising both productivity gains and governance risks.

Article:

OpenAI has launched ChatGPT Work, a new AI agent inside ChatGPT designed to handle longer workplace tasks across connected apps, files and workflows. The timing matters: enterprise AI is moving beyond prompt-based assistance toward agents that can create documents, spreadsheets, presentations, reports and websites with less manual stitching.

ChatGPT Work is powered by GPT-5.6 and brings Codex technology into broader business use, making OpenAI’s coding capability relevant for sales, finance, marketing, operations and other non-technical teams. OpenAI says the tool can gather information from apps and workflows, break complex projects into smaller steps, and continue working for hours while users review, redirect or approve important actions.

The clearest signal is usage. OpenAI says more than 5 million people use Codex every week, with more than 1 million using it for work outside software development. Its enterprise AI report also found weekly ChatGPT Enterprise messages rose roughly 8x over the past year, while average worker messages rose 30%.

“ChatGPT Work is an agent that can take action across your apps and files,” OpenAI said, positioning the product as a shift from answering questions to producing finished work.

The launch also raises governance questions. Agents that touch email, CRMs, drives, calendars and internal files need strict permissions, audit trails and approval gates. OpenAI says enterprise admins can control access, connected tools and actions, with compliance visibility for oversight.

The takeaway for businesses is blunt: AI productivity will now depend less on who writes better prompts and more on who redesigns workflows safely. ChatGPT Work may save time, but only for teams disciplined enough to decide what should be automated, reviewed and kept firmly human.