Mozilla AI has released cq (colloquy), an open-source knowledge-sharing system for AI coding agents. Where Stack Overflow once served developers, cq lets agents share what they’ve learned. In the process, the platform’s creators intend for AI agents to avoid repeating the same mistakes. The project launched as a working proof of concept earlier this month.
Mozilla staff engineer Steve Wilson is echoing a sentiment largely shared about Stack Overflow. Namely: it is either dead or dying, with a dramatic drop-off in questions submitted over the past few years. AI only seems to have accelerated its demise, and its possible replacement is entirely centered on the technology. With AI agents relying on training data from those human-made Stack Overflow discussions, they can now turn to cq to improve further. In the rather dramatic words of Wilson, AI committed matriphagy on its parent, consuming the platform that once helped conceive it.
What cq actually does
The problem cq is solving is essentially that AI agents in isolation fail to learn anything. Together, they can correct one another. Failed AI attempts can thus become part of a larger corpus, showing agents a list of worst practices if you will. These should help prevent failed builds, burned tokens and wasted compute cycles. Wilson points out that cq is designed exactly to counter this.
cq works as a shared knowledge commons. Before an agent starts on unfamiliar work, such as an API integration, a CI/CD config or a framework it hasn’t touched before, it queries the cq commons. If another agent has already solved that problem, the knowledge is available immediately. When an agent discovers something new, it proposes that knowledge back. Other agents can confirm what holds up and flag what’s gone stale.
The project ships as an MCP server managing a local knowledge store, with plugins for Claude Code and OpenCode, a team API for sharing across organizations, a human-in-the-loop UI for review, and containers for spinning the whole stack up. It’s already installable.
Trust signals, not just documents
One more irony to pile onto here is that the shift to AI is explicitly tracked by Stack Overflow’s surveys, which may very well be the only remnant of that platform in short order. According to last year’s Developer Survey, 84 percent of developers now use or plan to use AI tools. Nevertheless, 46 percent don’t trust the accuracy of the output, up from 31 percent the year before. Knowledge confirmed by multiple agents across multiple codebases, Mozilla AI argues, carries more weight than any single model’s best guess.
Mozilla’s AI unit plans to build out confidence scoring, reputation signals, and other trust mechanisms beyond static documentation. The current approach of dropping instructions into markdown files and hoping agents adhere to them only goes so far. There’s clearly still work to do then.
Andrew Ng raised a similar question publicly in early 2026, Wilson notes, asking whether a Stack Overflow equivalent for AI coding agents was worth building. Mozilla AI says it had already started, and is now seeking community input to shape how agent knowledge sharing gets standardized. cq is open source and under active development.