Resolving 'Intent Debt' in AI-Era Development Teams: Collaboration Strategies Centered on Guides and Context Sharing

While AI adoption has increased individual developer efficiency, the resulting surge in outputs has created new challenges, including review burdens and 'intent debt.' Intent debt refers to the loss of context regarding why certain decisions were made, and it is identified as the only type of debt that AI cannot repay on its own. To resolve this, decision-making processes trapped within AI conversations must be externalized into team-level guides and skills.

Context Sharing Through Guides and Skills

Guides are not merely lists of rules but records of the team's direction and rationale, designed to be understood by AI. Decisions made during work are initially recorded provisionally and only promoted to guides after verification through team reviews. It is crucial to adhere to the principle that humans must first clarify intent, while AI records only facts, to prevent AI-inferred intent from being mistaken for human intent. As seen in Anthropic's case, where removing over 80% of Claude Code's system prompts resulted in no performance loss, guides should provide strong directives on core matters while ensuring flexibility.

Division of Roles Between Humans and AI and the Review Process

AI excels in areas with abundant training data, such as syntax and standard library usage, but human review is essential for domains with wide probability distributions, such as domain-specific rules and exception handling. During reviews, the history—including the prompt, loaded skills, and decisions—should be referenced alongside the Diff to examine both the output and the process. Rules that can be mechanically checked should be enforced via gates, while aspects that cannot be mechanically verified, such as design and domain language, should be confirmed through inter-stage contracts and reviews.

Team Growth and Psychological Safety

Research suggests that the performance of star developers is a product of the organization and context rather than individual ability alone, making team collaboration culture key in the AI era. As demonstrated by Google's Project Aristotle study, psychological safety is a factor in high-performing teams, and a culture of sharing failure experiences and concerns is part of the process of finding the right way to utilize AI. It is important to convert individual trial-and-error into team assets by sharing external news through internal channels and conducting immediate retrospectives.