Stack Overflow is remaking itself into an AI data provider

As part of Microsoft’s Ignite conference, Stack Overflow on Tuesday revealed a new set of products that aims to position it as a valuable part of the enterprise AI stack. This new version of the company, built around the Stack Internal enterprise product, looks to remake its classic problem-solving forum into a tool for translating human expertise into an AI-accessible format.

At its simplest, Stack Internal is an enterprise version of the web forum, but with the additional security and admin controls you’d expect. The new tools are specifically designed to feed into internal AI agents using the model context protocol, with certain variations designed specifically for Stack Overflow.

As CEO Prashanth Chandrasekar tells it, Stack Overflow was already seeing a number of enterprise customers use its API for training, which inspired the new product direction. The company also has content deals with a number of AI labs, allowing them to train models on public Stack Overflow data in exchange for a blanket fee. 

While Chandrasekar wouldn’t name specific clients or figures, he described the arrangements as “very similar to the Reddit deals,” which have brought in more than $200 million for that platform.

A critical part of the new products is a layer of metadata that Stack Internal exports alongside the question and answer pairs. That data includes basic information like who answered the question and when, as well as content tags and more complex assessments of internal coherence. These factors are then used to create a general reliability score, which informs the AI agent how much each answer can be trusted.

“The customer can set up their own tagging system or we can dynamically create that for them,” said CTO Jody Bailey. “What we’ll be doing in the future is really leveraging that knowledge graph to connect concepts and pieces of information, rather than requiring the AI systems to do that on their own.”

While Stack Internal is producing tools for enterprise agents, it isn’t building those agents itself, so it’s difficult to say what the final product will be able to do. But Bailey is particularly excited about the writing function, which would allow agents to create their own Stack Overflow queries if they can’t answer a question or notice a knowledge gap.

As Bailey sees it, that read-write functionality means that, “as we continue to evolve, it will require less and less effort from developers to capture the unique information about the way they operate their business.”