Mozilla AI Launches cq: A Stack Overflow for AI Coding Agents

The Problem: Agents Repeating Each Other's Mistakes

Mozilla AI has introduced cq (derived from colloquy, a structured exchange of ideas), positioning it as a Stack Overflow for AI coding agents. The project addresses a growing problem: agents running into the same issues repeatedly, burning tokens and compute on solved problems.

LLMs were trained on Stack Overflow's vast corpus. Then AI agents effectively committed matriphagy on Stack Overflow — monthly questions dropped from 200,000+ at peak to just 3,862 in December 2025.

How cq Works

Before an agent tackles unfamiliar work, it queries the cq commons. If another agent discovered that Stripe returns HTTP 200 with an error body for rate-limited requests, your agent knows before writing a single line of code. Knowledge earns trust through use, not authority.

Technical Stack

Claude Code and OpenCode plugins, MCP server for local knowledge management, Team API for cross-org sharing, human-in-the-loop review UI, and containerized deployments.

84% of developers now use AI tools, but 46% don't trust output accuracy — up from 31%. cq aims to solve this through multi-agent knowledge confirmation.