OpenAI shifts ChatGPT to workflows

OpenAI spent the past week pushing ChatGPT away from one-off prompts and toward software that can finish jobs across apps, files and team workflows. (openai.com) (help.openai.com) The product move started on April 22, when OpenAI began rolling out workspace agents to ChatGPT Business and Enterprise workspaces over the following weeks. Those agents can run in ChatGPT or Slack, connect to Google Drive, Google Calendar, Slack and SharePoint, and execute scheduled recurring runs. (help.openai.com) Admins also got controls that look more like information-technology governance than consumer chat settings: version history, analytics, publishing controls, and app permissions that can allow all actions, read-only actions, or custom action policies. (help.openai.com) A day later, OpenAI released GPT-5.5 in ChatGPT and said the model is built for “real work,” not just conversation. The company said GPT-5.5 can write and debug code, research online, analyze data, create documents and spreadsheets, operate software, and keep moving across tools until a task is finished. (openai.com) (help.openai.com) OpenAI’s own benchmark table framed the pitch in workflow terms. It reported GPT-5.5 scoring 82.7% on Terminal-Bench 2.0, 78.7% on OSWorld-Verified and 84.4% on BrowseComp, all tests tied to long, tool-using tasks. (openai.com) The same release said nearly 200 trusted early-access partners tested GPT-5.5 before launch, and OpenAI updated the system card on April 24 when it made GPT-5.5 and GPT-5.5 Pro available in the application programming interface. (openai.com) ChatGPT’s surrounding features now match that pitch. Deep research lets users approve a research plan, limit which sites or connected apps it can use, follow progress while it runs, and get a report with citations or source links. (help.openai.com) Skills add another layer of repeatability. OpenAI describes them as reusable workflows that can bundle instructions, examples and code, then be shared inside a workspace so teammates can run the same process without rebuilding it. (help.openai.com) Sam Altman tied the product direction to a broader company message on April 26, when OpenAI published a new principles document. It says users should “reliably be able to accomplish increasingly valuable tasks” while the company gives broad latitude, minimizes harm and errs on the side of caution under uncertainty. (openai.com) That combination leaves ChatGPT looking less like a chatbot with extra buttons and more like a supervised worker that needs access rules, audit trails and source controls before companies trust it with recurring tasks. (help.openai.com 1) (help.openai.com 2)