ChatGPT Plugins

The Promise

In March 2023, OpenAI announced ChatGPT Plugins with the kind of fanfare usually reserved for platform launches. The pitch: turn ChatGPT into an extensible platform where third-party developers could add capabilities. Plugins would let ChatGPT browse the web, run code, access databases, book flights, order food, and do anything else developers could imagine.

The vision was explicitly modeled on the iPhone’s App Store. OpenAI was positioning ChatGPT not just as a product but as a platform—the foundation for an entire ecosystem of AI-powered applications. Early partners included Expedia, FiscalNote, Instacart, Kayak, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram, and Zapier.

Sam Altman’s blog post framed it as a new computing paradigm: “Language models could be the ‘base model’ in AI the way operating systems are in traditional computing.” If true, whoever controlled the plugin ecosystem would control the future of AI-native software.

The Rise

The initial excitement was enormous. Developers rushed to build plugins. Companies saw it as the next great distribution channel. Tech media declared it the birth of the AI App Store. Waitlists formed for plugin access, both for users wanting to try them and developers wanting to build them.

OpenAI expanded the plugin ecosystem through 2023, eventually hosting hundreds of plugins across categories: productivity, search, shopping, travel, education, entertainment. The company introduced a Plugin Store interface that let users browse and install plugins, completing the App Store metaphor.

For a moment, it seemed like OpenAI had cracked the code on AI distribution. Every startup wanted a ChatGPT plugin. Every enterprise wondered if they needed one. The platform narrative drove additional investment into ChatGPT-adjacent companies.

The Fall

But the usage never materialized. Unlike mobile apps, which users actively seek out and return to daily, ChatGPT plugins were cumbersome to use and easy to forget. Users had to manually enable plugins before conversations, couldn’t use more than three at once, and often found that ChatGPT’s native capabilities were sufficient.

The discovery problem was fatal. Users didn’t browse the Plugin Store the way they browsed the App Store. They came to ChatGPT with questions, and if a plugin wasn’t immediately relevant, they didn’t use it. The “pull” model of app stores didn’t work in a “push” model AI interface.

Developer enthusiasm waned as usage metrics disappointed. Maintaining plugins required ongoing effort, but the return on investment was unclear. Many plugins were abandoned within months of launch.

By November 2023, OpenAI had already pivoted. GPTs—custom ChatGPT personas that could be created without code—were announced as the future. Plugins were legacy. On March 19, 2024, OpenAI officially ended new plugin conversations. By April 9, 2024, even existing plugin conversations were shut down.

The AI App Store had lasted barely a year. The ecosystem that was supposed to define the next era of computing simply… evaporated.

Warning Signs

• Friction in activation: Requiring users to manually enable plugins broke the conversational flow
• Discovery mismatch: App Store browsing behavior didn’t translate to chat-first interfaces
• Limited plugin slots: Only three plugins at once made complex workflows impossible
• Cannibalization by core product: ChatGPT kept getting better at things plugins were supposed to do
• Developer economics unclear: No clear monetization path for plugin developers created an unsustainable ecosystem

Epitaph

🪦 Built an ecosystem nobody came to