Mozilla has introduced an open source project to address a growing inefficiency in AI systems: their tendency to repeatedly solve the same problems in isolation.
The initiative, called cq, is described by Mozilla engineers as a Stack Overflow for agents, a reference to the long-running developer forum that once served as a central repository of programming knowledge. As AI coding tools have proliferated, traditional knowledge-sharing platforms have declined, while the systems trained on their data now operate largely without shared memory.
The goal of cq is to reduce duplicated effort by enabling AI agents to query a shared knowledge base before attempting unfamiliar tasks. If a solution already exists (like how an API behaves under specific conditions) the agent can use that feedback immediately. When new issues come up, agents can contribute their findings back into the system, where other agents and human reviewers validate the results over time.
As Mozilla engineer Peter Wilson noted on the Mozilla blog, agents frequently “run into the same issues over and over,” consuming compute resources to rediscover solutions that already exist.
This inefficiency has become more visible as AI gains ever greater adoption. Surveys show that while a majority of developers now use AI tools, confidence in their accuracy has declined.
The platform introduces a tiered structure for knowledge sharing. Information can remain local to an individual developer, expand to an organization-wide pool, or eventually be contributed to a broader public commons. Each unit of knowledge begins with low confidence and gains credibility as it is confirmed through repeated use.
Wilson summarized the philosophy: “Knowledge earns trust through use, not authority.”
Stack Overflow has seen activity fall sharply in recent years. Some observers say this is due to the rise of LLMs, which harvested much of the platform’s content during training and now provide answers directly to users. Wilson described this dynamic using the biological term matriphagy, in which offspring consume the parent. In this case, the data that sustained developer communities has been repurposed into systems thzat (in theory) no longer rely on those communities for ongoing input.
A Work in Progress
The cq project is still a work in progress, and security concerns have emerged. Because cq relies on shared contributions, it could become vulnerable to malicious inputs, ranging from subtle inaccuracies to deliberately harmful instructions embedded in seemingly valid solutions.
Mozilla has outlined several countermeasures, including anomaly detection, requirements for corroboration from multiple sources, and human oversight before knowledge is widely shared. Yet some developers remain skeptical that such safeguards can scale effectively, particularly in an open system.
In the bigger picture, there are questions about whether shared knowledge alone addresses the challenges of current AI systems. The issue may be less about access to information and more about the ability of agents to work reliably.
The current version of cq is available as a proof of concept, with plug-ins for coding tools and infrastructure for local and team-based deployments. A public, global knowledge commons remains a longer-term ambition.
