OpenAI Lets Your Projects and Skills Follow You Between Agents

OpenAI is quietly attacking the most annoying part of living in multiple AI tools: your context doesn't travel. The new sync means projects, chats, skills and plugins stay consistent across Codex, ChatGPT Work, and other AI agents — so the instructions you wrote once don't get re-typed in every window.

The interesting part isn't convenience, it's positioning. Skills — reusable folders of instructions an agent loads on demand — have become the portable unit of AI configuration, and whoever owns where they live owns the workflow. Anthropic pushed the same idea with Claude's Skills; OpenAI syncing them across surfaces makes the config layer the sticky part of the product, not the model.

For anyone rotating subscriptions between labs, this cuts both ways. Shared state across your own tools is a real time-saver. But a config layer that lives in one vendor's account is also the tidiest lock-in ever built — it's much harder to leave a platform that holds every skill and project you've assembled than one that just holds a chat history.

Why it matters: the competitive battleground is shifting from the model to the context layer around it.