Just-in-Time Software
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
A design philosophy for AI-native applications in which software is kept as minimal as possible, providing only the scaffolding necessary to let the underlying model perform — analogous to just-in-time manufacturing, where components are produced only as needed rather than stockpiled in advance.
Core Idea
The concept emerges from the observation that the best AI software adds the minimum amount of code needed to let the model shine. Rather than building heavyweight, feature-laden applications in the traditional software engineering style, just-in-time software deliberately defers to the model's capabilities, treating excess code as waste to be eliminated. The principle was articulated at Inside YC's AI Playbook (34:46), where the speaker declared it "time for a just-in-time software" as a guiding posture for AI product development.
Implications for AI Product Design
The concept implies a inversion of conventional software instincts: instead of encoding logic, rules, and workflows exhaustively in application code, developers should ask what the model can handle natively and write code only where the model genuinely cannot. The result is thinner application layers, with intelligence concentrated in the model rather than distributed across a large codebase. This approach connects directly to questions about how much structure to impose on AI agents versus how much latitude to grant them.