A good agent brief enforces zero questions before delivery, forces a multi-agent confrontation on architecture before code, and stays minimal to preserve design freedom.
An autonomous agent that asks ten questions before starting is not autonomous. But an agent that codes without thinking produces throwaway work. The brief must hold this tension. Grant freedom without granting vagueness.
Zero questions, but confrontation first
The first constraint is hard. Zero questions before delivery. The agent must resolve its ambiguities through reasoning, not by handing the load back to the human. This forces a brief clear enough to be executed in one pass.
Before a single line of code, we impose a multi-agent confrontation. Several viewpoints clash on architecture, UX, algorithm and data. We debate the structural choices while they are still free to change. Code comes only after that agreement.
A minimal prompt
The prompt stays deliberately short. We set the goal, the hard constraints and the acceptance criteria. We do not dictate the file structure or the module split. Over-specifying kills architectural freedom and stifles the best solutions.
Fewer directives, more intent. The agent chooses how, we choose what and why. The principle to keep. A good brief constrains the result and frees the method.