Stack Overflow says a ChatGPT plugin tripled agent signups overnight

OpenAI’s October 16, 2026 customer story credits a ChatGPT plugin with tripling overnight registrations for Stack Overflow for Agents, and with pushing agent creation as high as 10 times. Those figures are Stack Overflow’s, published by OpenAI. They are not an independent count.

Alexandra Lato, vice president of product at Stack Overflow, is quoted on the page: “We built a ChatGPT plugin for Stack Overflow for Agents, and our registrations tripled overnight.” Her line on agent creation is looser: “I think we were up to 10X last time I checked.” OpenAI’s headline compresses that into “up to 10× growth in agent creation.” Cards on the same page read “3x More signups for Stack Overflow for Agents overnight with the ChatGPT plugin” and “10x More agents created with the ChatGPT plugin.”

This is a growth case study, not a launch note. The page starts from a product two engineers had already shipped, then explains how a plugin changed discovery and setup.

Shipped in a few weeks with Codex

OpenAI says Stack Overflow for Agents began as a hypothesis and a two-engineer effort using Codex. The engineers gave Codex a development environment and a plan, then worked with it to build the product. Within a few weeks, Lato says, they had a full-fledged product. “So it really made their team feel bigger,” she says. “They were able to throw a container at it and say, ‘This is our plan, help us build it.’”

The page sets that build inside a knowledge intelligence platform: context scattered across systems, from Slack to Google Docs, turned into verified knowledge for people and agents. It says models including GPT-6 Astra and GPT-6 Sol power the platform’s AI, with tasks routed by the reasoning required and by cost. Codex is how the engineers build. The ChatGPT plugin is how the newer agent product is meant to be found and used. The story lists Codex and the API as the OpenAI products involved.

Fewer steps than an MCP setup

Distribution was the next problem. Once Stack Overflow for Agents was ready, OpenAI says, the team needed it inside people’s workflows. They built the plugin so users could discover and set the product up in ChatGPT, with fewer steps than a traditional Model Context Protocol (MCP) setup. Lato’s line: “Once you have a plugin, it becomes a frictionless setup with ChatGPT.”

The page does not compare plugin install counts with MCP install counts. Signup growth and agent creation are attributed to the plugin. The MCP point is about steps, not a measured win over an existing install base.

The connector claim, separately

The same story pairs the growth numbers with a Codex claim about connectors for Stack Internal, Stack Overflow’s enterprise product. A stat card states a maximum 87% reduction in time to build one connector, from 3–6 weeks to 4 days. The quote behind the 4-day figure is Lato’s, and it names Codex: “On average, it takes us 3 to 6 weeks to build one connector. With Codex, we were able to build it within 4 days.” OpenAI also says the team more recently built a connector in a single day, and that Lato describes a path to market of three months rather than a year for the enterprise product. Connectors, the page notes, still need more human involvement than the two-engineer agent product.

Treat the 87% figure as OpenAI’s summary of Stack Overflow’s stated range, not as a lab measurement. The underlying claim is Lato’s 3-to-6-week average versus a 4-day build with Codex.

What the live listing claims

A ChatGPT listing for Stack Overflow For Agents is published under developer Stack Exchange, Inc., in Developer Tools, at version 1.5.0. Its text says the product provides free, trusted, production-tested knowledge so agents spend less time and fewer tokens rediscovering answers. Agents can view, reply to, upvote, and verify posts created by other agents. An agent stuck on a technical problem can ask a question and see if other agents can solve it.

Skills listed are sofa, sofa-contribute, and sofa-status. The actions shown are checking connection status, finding knowledge for a task, and contributing a reusable learning as a post. The website field points to agents.stackoverflow.com.

Before you repeat the overnight tripling, read OpenAI’s October 16 customer story and keep the 3x and 10x figures labeled as Stack Overflow’s. Then open the ChatGPT plugin listing and compare those three actions—status, search, and contribute—with the MCP setup you would otherwise ask someone to finish.