Anyone who has spent a few hours with an AI coding assistant knows the frustration: you describe what you want in chat, the AI builds something close but not quite right, and you spend the next hour course-correcting through follow-up messages. OpenSpec, a new open-source framework from Fission AI, tries to fix this by adding a structured specification layer between you and the AI.
The idea is simple. Instead of dumping requirements into a chat window and hoping for the best, you run a command like /opsx:propose "my-feature" and the system generates a dedicated folder with structured documents: a proposal explaining the rationale and scope, a requirements spec, a technical design doc, and a task checklist with subtasks. Once you and the AI agree on what to build, you run /opsx:apply and the AI implements against those concrete specs rather than fuzzy chat history.
OpenSpec works with over 20 AI coding tools including GitHub Copilot, Cursor, and Claude. The framework recommends using high-reasoning models like Opus 4.5 or GPT 5.2 for both planning and implementation. It runs via slash commands, so the integration is lightweight - no plugins to install or configs to wrestle with.
The practical appeal here is for teams where multiple people (or multiple AI sessions) touch the same codebase. Specs persist as files in your repo, so a new AI session can pick up context without you re-explaining everything. That said, this is early-stage software with minimal community traction so far, and the real test will be whether the overhead of writing specs pays off versus just prompting more carefully. For solo developers on small projects, the extra structure might feel like overkill. For teams shipping features across multiple sessions, it could save real time.