Stack Overflow’s vice president of product says a plugin inside ChatGPT tripled registrations for the company’s new product for AI agents. The claim comes from a customer story that OpenAI published on its own website, so it is a testimonial gathered and posted by the vendor, not independent reporting and not a measurement.
The quote, from Alexandra Lato: “We built a ChatGPT plugin for Stack Overflow for Agents, and our registrations tripled overnight.” The post offers no baseline, no absolute signup count, no time window beyond that one word, and no outside verification. Lato later adds that agent creation was “up to 10X last time I checked,” and she prefaces that with “I think,” so the figure is her recollection. A tripling from ten signups and a tripling from ten thousand are very different stories, and the post does not say which this was.
The mechanism matters more than the multiple. Stack Overflow for Agents is the company’s newer product for software agents, programs that act on a user’s behalf. The plugin lets people find and set up that product without leaving ChatGPT, in fewer steps than a traditional setup through MCP, the standard for connecting agents to outside tools. “Once you have a plugin, it becomes a frictionless setup with ChatGPT,” Lato says.
That makes the assistant a storefront. A developer who would never have searched for this product can meet it in the same window where they already work, and the company that runs the window decides how that meeting happens. The post says nothing about how plugins are ranked or what terms apply.
The second claim concerns speed. According to the post, the agents product started as a hunch tested by just two engineers, with Codex, OpenAI’s coding agent, doing much of the building. Their input was a container to work in and a written plan, and they then worked alongside the tool. Within a few weeks, Lato says, they had a full-fledged product. “So it really made their team feel bigger,” she says.
The enterprise product, Stack Internal, gets a more modest version of the same story. Lato says the team recently built one connector to another system in a single day, after an earlier one took four days with Codex. Without it, she says, a connector typically takes three to six weeks. She puts the product’s path to market at three months instead of a year. This project involved more people than the small team behind the agents product, the post concedes.
One architecture detail is concrete. Stack Overflow’s platform uses two OpenAI models, GPT-6 Astra and GPT-6 Sol, and sends each job to one or the other based on the reasoning it needs and its cost. The post does not say which tasks land where. Stack Overflow handles that routing and configuration itself today. Lato hopes OpenAI’s Decisions API could automate it, which she says would save “a ton of time and money.” A vendor’s customer story in which the customer pitches the vendor’s own product for a job it now does itself is worth reading with that incentive in mind.
The wider pitch is scattered context. Information sits in Slack, Google Docs and other systems, and Stack Overflow says it is building a platform that gathers it and turns it into verified knowledge for people and agents.
The source does not comment on this, so the observation is ours: a company that built its name on people answering developers’ questions is now selling checked answers to machines. The post reports signups and speed, with no revenue and no paid use.
For any team shipping a developer tool in the next quarter, a listing inside ChatGPT is now a distribution channel worth testing, and the first thing to measure is your own baseline, because the one number Stack Overflow’s story leaves out is the one that makes a tripling mean anything.
Based on OpenAI’s customer story about Stack Overflow, published on openai.com (the source gives no publication date).
