DEV Community: Cyrille Huteau

OpenAI launches ChatGPT plugin discovery, sort of

Cyrille Huteau — Thu, 01 Oct 2026 13:37:33 +0000

We probed OpenAI's plugin search with 3,000+ queries and a paired prompt evaluation. The catalog is small, skewed toward business software, and models rarely consult it unless the user asks.

Yesterday OpenAI announced that it would support dynamic discovery of plugins, in other words, would suggest using plugins when a user query matched the capabilities of a plugin present in the store.

To test to what extent these capabilities had already been rolled out, we ran 3,000 search queries through the plugin search that OpenAI exposes to its models in Codex, and across all of them it surfaced 152 distinct plugins. Of those, more than 50% serve business users only, 37% serve both audiences, and the remainder are aimed squarely at consumers. In a separate evaluation still in progress, the models we tested virtually never searched for a plugin on their own, even when the task plainly called for one. Asked directly whether a plugin could help, they searched almost every time.

How plugin discovery works today

Plugin discovery runs through a plugin-management skill with two tools, one that searches the catalog and one that formally suggests a plugin to the user, who can then install it. It would appear that the model is instructed to reach for it when the user asks about plugins or when a task would materially benefit from an external app, service, or data source it can't otherwise access. Nothing forces a search, so whether discovery happens at is left to the model's judgment on each turn.

A catalog of 152, weighted toward business software

The 1,000 queries covered several hundred verticals, from restaurant reservations and telehealth to maritime logistics. The business-only plugins are mostly CRMs, analytics, finance, and project management tools, while the consumer side comes down to four travel brands (Booking.com, Expedia, Skyscanner, Tripadvisor), Instacart, and a handful of fitness trackers.

The search tool appears to look for keywords separately, so a “food tracking” query will return both food-related plugins and analytics-related plugins.

Linear, Attio, and HubSpot each appeared in roughly 240 of the 1,000 result sets, largely because a query like "retail CRM sales pipeline" or "banking CRM sales pipeline" returns the same cluster of sales tools whatever the industry.

Consumer categories are sparse enough that 24 queries returned nothing at all, among them music streaming, telehealth, pet sitting, secondhand fashion, meditation, and homework tutoring.

Some kinks to work out…

776 of the 1,000 queries hit the ten-result ceiling, though many of those results had little to do with the query. "Restaurant reservations" returned WordPress.com, and "medical appointment booking" returned Calendly alongside QuickBooks, NetSuite, Skyscanner, and GitBook.

A model receiving these results has to filter the noise itself, and our evaluation shows it often concludes that nothing fits.

Models look for plugins when asked, and rarely otherwise

Alongside the catalog probe, we're running a prompt evaluation to see when models reach for plugin search on their own. It takes 200 user tasks, half business and half consumer, and phrases each one twice, once as a user naturally would and once followed by "Does ChatGPT have a plugin that can help with this?" Every prompt runs in a fresh Codex context with account access disabled.

The evaluation is still running, but our initial findings suggest that plain prompts virtually never trigger a plugin search, while the explicit question triggers more than 90% of the time. However, even when the search is triggered, the times when a formal suggestion is made drops off steeply. We'll update this article once the sample is large enough to support firmer conclusions.

What’s next?

Plugin search gives a first look at how OpenAI intends to route users to apps, but the flow is clearly still in progress, and the catalog, the search results, and the models' willingness to use them will likely shift in the coming weeks. We've set up Atlas to track plugin search over time, so you can follow those changes as they happen! Check it out at https://atlas.alpic.ai/discoverability

OpenAI Dev Day 2026: from personal agents to new distribution possibilities

Cyrille Huteau — Wed, 30 Sep 2026 18:13:17 +0000

OpenAI’s Dev Day keynote made a case for AI that does more than answer questions. The announcements moved from Dots, personal agents that can take on work, to the models and developer tools behind them, and finally to ways to find users and build businesses through ChatGPT. The distribution announcements were the clearest expression of that last ambition: OpenAI wants developers not just to build with its technology, but to reach customers through its products.

Dots: a “delegate” with your context

Dots were presented as personal AI delegates that know what a user is working on and can take work off their plate. In the demo, “Dottie” surfaced a changed launch review, feedback from test users, and a last-minute design revision. It took voice requests, accepted tasks handed off through Slack, and used Codex to work on an app.

The keynote also showed dots participating in group chats and previewed specialist dots that Enterprise teams could set up for shared work.

Unlike the personal AI agents released by Meta and Instinct in recent weeks, OpenAI has taken a more enterprise approach to the personal AI assistant, showing mostly professional use cases and arguing that most of the value in this segment resides in highly complex tasks.

Space: a place to work together

Dots were about delegation, Space about collaboration. OpenAI introduced it as a place inside ChatGPT where teammates and their dots could work together on interactive charts, sheets, prototypes, and other shared material. The folks at Notion are most likely sweating, as the demo looked an awful lot like the current collaboration tool.

Codex: work that continues beyond your laptop

The Codex announcements centered on moving work across devices and into the cloud. OpenAI said a task could start on a phone and continue in a browser or desktop app without keeping a laptop open. It described the open-source Codex harness as a foundation for its own agents and for the new Agents API, announced in public beta with hosting, memory, and multi-agent controls. The keynote also introduced Codex Security Cloud to find vulnerabilities and prepare fixes for review, and demonstrated computer and browser use, a refreshed CLI, and UltraFast.

Distribution: turning ChatGPT into a route to customers

The final major section addressed a different problem: once you’ve built something, how do people discover it, try it, and pay for it?

The first route was Sign In with ChatGPT. Users will be able to sign into an app and use tokens included in their ChatGPT subscription. OpenAI’s pitch was that this would make a new product easier to try: users would already have an AI subscription, while developers would not have to cover the initial token cost. The keynote said the launch would begin with 16 partners and expand from there.

The second route was plugin extensions, an extension to MCP Apps that allows for richer personalization such as an interactive editor, dashboard, or workspace inside ChatGPT and Codex. This could include a collaborative design workflow in Figma or Adobe, as shown in the demo and the image below.

Building an extension, though, is only useful if someone can find it. OpenAI therefore unveiled the much awaited discovery mechanism within a conversation: when a plugin is relevant to a question, ChatGPT can surface it and let the user connect it there, assuaging many fears that plugins would be a side feature that was difficult to find and required installation before use.

The third route was aimed at enterprise sales. OpenAI announced a Marketplace with more than 30 launch partners. Enterprise customers, it said, could use part of an existing OpenAI commitment to purchase partner products; companies with their own commitments could spend them in the marketplace too.

Taken together, the distribution announcements described three distinct ways to build a business around OpenAI’s platform: let ChatGPT subscribers use your app, make your app part of ChatGPT, or sell it through a marketplace for enterprise customers. That was the larger story of the keynote’s closing section: not simply more tools to build with, but an attempt to make ChatGPT a place where apps can be discovered and do business. While the focus of the day’s conference was still largely B2B-oriented, alongside its many improvements in its ad platform, OpenAI is laying the groundwork for use cases that span both work and large-scale B2C use cases.