OpenAI head forecasts work shift

OpenAI’s product messaging is converging on a specific claim: users will stop judging AI tools as single-turn assistants and start expecting them to carry out multi-step work. In a May 22 interview, Thibault Sottiaux, who has led ChatGPT and Codex work at OpenAI, said the next six months would change knowledge work by shifting expectations around orchestration, context and natural-language control. That forecast lines up with OpenAI’s own recent releases, which have added long-running agents, memory features, approvals and audit-style logging across ChatGPT and Codex. ### What exactly is OpenAI saying users will expect now? OpenAI has been building toward software that does more than answer prompts. The company said in April that “workspace agents in ChatGPT” are Codex-powered agents that automate complex workflows, run in the cloud and help teams scale work across tools securely. In February, OpenAI described the Codex app as an interface to manage multiple agents at once, run work in parallel and collaborate over long-running tasks. (openai.com) GPT‑5.3‑Codex, released in February, was presented as a model for “long-horizon, real-world technical work.” OpenAI said users can steer and interact with it while it is working “without losing context,” and said the model can handle research, tool use and complex execution. That is close to the product behavior Sottiaux described: not one request, one answer, but a chain of work that persists across time. (openai.com) ### Where do approvals and audit trails enter the picture? OpenAI’s own safety and enterprise materials show that approvals are becoming part of the workflow, not an afterthought. The company said this month that Codex is deployed so “low-risk everyday actions should be frictionless” while “higher-risk actions should stop for review,” and its developer documentation says approvals, sandboxing and network controls are core parts of operating Codex securely. (openai.com) OpenAI’s enterprise governance page also says Compliance API exports can provide audit records for Codex activity and be used in monitoring, investigations and compliance workflows. The same page says those exports include detailed activity logs, while enterprise release notes say mobile surfaces can show project context, approvals, screenshots, terminal output, diffs and test results from an active Codex environment. (openai.com) ### How is preserved context becoming a product feature? OpenAI has been expanding memory as a visible system feature. The company’s ChatGPT release notes say “memory sources” are being introduced across consumer plans so users can see what information helped personalize a response and edit it if needed. Its Memory FAQ says saved memories can be used in future responses unless deleted, and that users can inspect and manage them. (developers.openai.com) Codex has a more operational version of that idea. OpenAI’s Chronicle documentation says Codex can use memories from prior sessions to fill in missing context, including references to documents, collaborators and recent work. That matters because the shift Sottiaux described depends on software remembering enough state to continue a task instead of restarting from zero. (help.openai.com) ### Why does that matter for editorial software? Newsroom workflows are full of chained tasks. A single assignment can involve ingest, transcription, clipping, drafting, review, revisions, approvals and export into multiple formats. If users get used to asking software in plain language to complete connected tasks while preserving context, then tools built around isolated buttons and one-off actions will face a higher bar. That is an inference from OpenAI’s product direction, not a statement OpenAI made directly. (developers.openai.com) OpenAI’s recent releases show the pieces of that expectation forming in public: multi-agent work, long-running execution, memory, approvals, mobile monitoring and audit records. For companies building workflow products, the next concrete signals will come from additional ChatGPT and Codex release notes, enterprise controls and developer documentation as OpenAI continues rolling out those features. (help.openai.com) (openai.com)