Custom GPTs let anyone configure ChatGPT without code

The conversation

On 6 November 2023, at its first DevDay, OpenAI announced Custom GPTs, "tailored versions of ChatGPT for a specific purpose" that a user could build "without any code." The announcement states the GPT Store "is rolling out later this month" and that GPT creation was open immediately to "Plus and Enterprise" users. OpenAI's help center, updated the following day, confirms the timing in nearly the same words, "Plus and Enterprise users can start creating GPTs this week," describing a two-tab editor with a conversational GPT Builder alongside a manual Configure tab for instructions, knowledge files and enabled capabilities.

What the documents show

The launch announcement frames GPTs as a successor to the plugins beta already under way, stating that "the design of actions builds upon insights from our plugins beta" and that a developer could reuse "your existing plugin manifest to define actions for your GPT." The 7 November help article, by contrast, lists only the built-in capabilities available at launch, web browsing, DALL-E image generation and Advanced Data Analysis, with no mention yet of externally defined API actions, which OpenAI's documentation added afterward. Custom Instructions, the July 2023 feature the announcement names as the demand signal for GPTs, is the one earlier capability both documents agree GPTs absorbed and extended.

The system boundary

The original design separated two things a builder controls: instructions, which set behavior, and knowledge files, which supply reference material a GPT can search but that the announcement says a builder should only include if authorized to share it in a user's session. The announcement ties GPTs to the builder's own ChatGPT account rather than a separate developer platform, and, at launch, to the Plus and Enterprise plans, a restriction the 7 November help article repeats independently rather than merely echoing.

Where it fails

OpenAI's own help article, as retrieved on 16 September 2026, now states a further reversal: "we're planning to retire custom GPTs" for affected Enterprise workspaces, recommending a move to a system it labels "Plugins," with retirement "planned for Dec 11, 2026" and a migration experience "targeted for Sep 17, 2026." The current document draws no line back to the 2023 plugins beta it once described GPTs as succeeding, so a reader should not assume the two uses of the word "Plugins," three years apart, name the same mechanism.

• Does today's documentation for a feature still describe the same capability its launch announcement claimed?
• When a vendor reuses a name from an earlier, retired feature, does its current documentation say whether the mechanism is the same?
• What migration path, if any, does the vendor's own notice offer before a stated retirement date?

Read together, the three documents show an interface built to replace one integration model turning, on OpenAI's own more recent notice, into an interface scheduled for replacement itself, which is exactly the pattern this batch keeps finding in this layer of the stack.