Brainstorming Requirements
Leveraging high-reasoning models for adversarial product discovery to challenge assumptions and define rock-solid acceptance criteria.
5 August 2026
The Gap Without Adversarial Discovery
By default, most LLMs are engineered to be helpful, agreeable "yes-men." If you ask an AI to build a feature, it will eagerly agree and start generating code based on your first draft of the requirements, no matter how flawed, incomplete, or insecure those requirements might be.
Without a dedicated brainstorming and discovery phase, three common failures occur:
- Happy-Path Obsession. The AI optimizes for the simplest success scenario and completely ignores edge cases, error handling, and malicious inputs.
- Missing Product Context. The AI doesn't ask why a feature is being built, resulting in technically functional code that fails to solve the actual user problem.
- Premature Implementation. Jumping straight into code before finalizing requirements leads to massive rework when fundamental logic flaws are discovered during later testing phases.
To prevent this, we must explicitly prompt the AI to switch personas—from an agreeable assistant to an adversarial product manager and principal engineer whose job is to poke holes in your ideas before they become code.
Never accept the first draft of requirements from an AI. Force the model to generate edge cases, ask clarifying questions, and play devil's advocate until the acceptance criteria are robust.
The Skill: /sprint-brainstorm
In the sprint-workflow operating system, after initializing the workspace, we invoke the discovery phase:
/sprint-brainstorm
Executing /sprint-brainstorm triggers an interactive, adversarial discovery session:
- Context Loading: Reads the
sprint-manifest.jsonto understand the sprint scope and loads any initial user prompts or legacy context. - Adversarial Interrogation: The AI acts as a principal product manager, asking targeted questions about edge cases, scale limits, and failure modes.
- Requirement Refinement: Collaboratively drafts a robust Product Requirements Document (PRD) containing clear user stories and acceptance criteria.
- Artefact Generation: Saves the final PRD to
.sprint/<slug>/docs/brainstorm.mdand updates the manifest.
Unlock the Full Article
This article covers the full walkthrough, model choice, and pitfalls. Please connect to find out more.