OpenAI case study: two engineers shipped Stack Overflow for Agents in weeks with Codex, and a ChatGPT plugin tripled signups overnight

Start with what the case study actually says. According to OpenAI, Stack Overflow is building a knowledge intelligence platform: both people and AI agents need context to get work done, but that context is usually scattered across systems from Slack to Google Docs, and finding it and understanding how it fits together is a challenge in itself. OpenAI models GPT-6 Astra and GPT-6 Sol power the platform's AI, with tasks routed to different models based on the reasoning required and cost; Codex helps the company's engineers build the products, while a ChatGPT plugin makes Stack Overflow for Agents easier to discover and use. Stack Overflow for Agents is positioned as extending that intelligence layer to agents.

The second point is how the product got built. OpenAI says Stack Overflow for Agents began with a hypothesis and two engineers using Codex: they gave Codex a development environment and a plan, then worked with it to build the product, and within a few weeks they had a full-fledged product. In the words of VP of Product Alexandra Lato, it really made their team feel bigger — they were able to throw a container at it and say, this is our plan, help us build it. Note that Codex's role in this story is the build tool, not the model inside the product; the product itself runs on GPT-6 Astra and GPT-6 Sol.

The third point is the concrete numbers around connectors and distribution. Stack Internal, the enterprise product, needs to connect to all the systems where organisations keep their knowledge, and OpenAI says Codex is accelerating that too: on average a connector took three to six weeks, and with Codex it was built within four days, with a more recent one finished in a single day; the enterprise product's path to market also shortened from a year to three months. Lato's words: on average it takes us 3 to 6 weeks to build one connector, and with Codex we were able to build it within 4 days. On distribution, the team built a ChatGPT plugin so users could discover and set up the product inside ChatGPT, which OpenAI says takes fewer steps than a traditional Model Context Protocol (MCP) setup. Lato's take on the plugin: once you have a plugin, it becomes a frictionless setup with ChatGPT; she says it tripled registrations, and that agent creation was up to 10x the last time she checked.

The boundaries deserve the same clarity. First, the three metrics at the top of the page are 10x more agents created with the ChatGPT plugin, 3x more signups, and up to an 87% reduction in time to build a single connector, from three to six weeks down to four days. All three come from a vendor-published customer case study, spoken by the customer and written up by OpenAI, with no third-party audit; the page gives no baseline, no conversion rates, no cost savings and no comparison against organic growth over the same period. Second, the plugin-versus-MCP comparison rests on a single customer quote, with no published time or drop-off data for either path. Third, the team's interest in the Decisions API is still an ambition to explore, aimed at automating some model routing and configuration work; it has not shipped and the case study offers no time or cost savings for it. Fourth, OpenAI and Stack Overflow have a commercial relationship: Stack Overflow uses OpenAI's models, Codex and plugin ecosystem, which makes this a customer testimonial rather than an independent evaluation.