Independent of any product, we put three ideas on the table — a vocabulary shift, a branching model, and a practice. None of them are Intent-System features; you can adopt any of them on the stack you already use.
AI capability moved faster than the practices and the vocabulary around it. Words like harness and guardrail came from a moment when models needed restraint. The branching models we use — Git-Flow, GitHub-Flow — were built before agents could open their own pull requests. And teams still jump from idea to code with no structured step for intent in between. We propose updates to three specific places.
The right metaphor for guiding capable AI today is not the harness — it is a compass. A harness restrains; a compass gives direction.
A branching model that places a buffer between AI velocity and human oversight: agents work continuously on ai-develop; humans promote selected states into main on their own schedule.
The practice you run before you write code: structuring technology and intent as the product owner, so the work walks out as an intent tree instead of meeting notes. A group workshop, or solo with the AI interviewing you.
All three proposals are embodied in Intent-System but don't depend on it. Compass: the Intent Tree supplies the direction for each slice. Git-AI-Flow: the implementation cycle runs naturally on ai-develop. Intent Storming: intent-cli supplies the interview structure that seeds the tree. You can adopt any of them without using Intent-System.