Shareable Assistants: ChatGPT Plugins, Gems, and Claude Projects
Package instructions and knowledge into an assistant other people can use — designing, testing, sharing, and maintaining it as a small product.
The fastest AI tool you can give your team is a configured assistant: a set of instructions and reference files, packaged in Gemini, Claude, or ChatGPT, that anyone you share it with can open and use. You can build one in twenty minutes. That's exactly the problem. Twenty-minute assistants get shared on Friday, behave strangely on Monday, and are quietly abandoned by the end of the month.
The assistants that last are treated like small products. They have a defined user and job, instructions written for people who won't read the instructions, a test set, an owner, and a change log. In this lesson you'll design, test, share, and maintain one that way, using a release-notes assistant as the worked example, and you'll learn what each platform's sharing model means for who can see what.
An assistant is a product with one screen
A shareable assistant is a chat assistant pre-loaded with standing instructions (like a system prompt), optional reference files the model can draw on, and sometimes connected apps or tools. You configure it once; other people open it by name and start typing. The three well-established homes for them, as of November 2026:
| Platform | What it's called | How sharing works | Worth knowing |
|---|---|---|---|
| Google Gemini | Gems | Shared like a Google Drive file, with view or edit access. Workspace admins control it through Drive sharing settings. | Knowledge files come from your device or Google Drive. |
| Claude | Projects | On Team and Enterprise plans, shared with people in your organization as "Can use" or "Can edit." | Members with use access can see the project's instructions and knowledge, not just chat. |
| ChatGPT | Plugins (replacing custom GPTs) | On Business and Enterprise workspaces, shared with workspace members. | OpenAI is retiring custom GPTs; see below. |
The details differ, but the product questions are the same everywhere: who is this for, what does it do, what does "good" look like, and who fixes it when it breaks.
If you built custom GPTs
OpenAI announced in 2026 that custom GPTs are being retired in favor of ChatGPT plugins, which bundle reusable instructions (called skills), reference files, and connected apps. At the time of writing, OpenAI's help center lists December 11, 2026 as the standard retirement date, with later dates for some enterprise workspaces, and many accounts can no longer create new GPTs. Its migration tool turns a GPT's instructions into a skill and copies knowledge files into reference files. Two things don't carry over: the model you picked and any custom actions (connections to outside APIs), which you'll need to rebuild. The migrated plugin also starts private, so your users lose access until you share it again. Check the notice for your own account for exact dates.
It's a good reminder of why this lesson insists on a master copy of everything outside the tool. Teams with their instructions and test cases in a document can migrate in an afternoon and verify the result. Teams without them are reverse-engineering their own assistant.