MIT researchers have introduced a new software design model aimed at making code more legible, modular, and easier to manage, especially in the age of large language models (LLMs). The approach, developed by a team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), centers on two key ideas: “concepts” and “synchronizations.” Concepts represent self-contained, well-defined pieces of functionality—like sharing, liking, or following—each with its own state and actions. Synchronizations are explicit, high-level rules that define how these concepts interact, replacing scattered, low-level integration code with clear, declarative connections. The method was presented in a paper titled “What You See Is What It Does: A Structural Pattern for Legible Software” at the Splash Conference in Singapore. The team, led by MIT professor Daniel Jackson and EECS PhD student Eagon Meng, designed a small domain-specific language (DSL) to express synchronizations in a way that is both human-readable and LLM-friendly. This allows developers to describe how different parts of a system work together without getting lost in implementation details. A major problem in current software development is “feature fragmentation,” where a single feature—like a share button on a social media app—spans multiple services and codebases. This makes it hard to understand, test, and modify. Jackson explains that today’s code often hides the relationships between components, forcing developers to hunt through multiple files to trace a single function. The new model aims to make these connections visible and explicit. By grouping features into clean, independent concepts and using synchronizations to define their interactions, the system becomes more transparent. Synchronizations act like contracts, ensuring that changes in one part of the system are properly reflected in others. This not only improves clarity but also enables better verification, analysis, and automated generation by LLMs, reducing the risk of unintended side effects. In a real-world case study, the researchers applied the model to a social media platform, assigning each feature—like liking, commenting, and sharing—to a single concept. Previously, these features were spread across multiple services. With the new approach, they became centralized and easier to understand. Synchronizations also helped standardize common concerns like error handling and data storage, reducing duplication and increasing consistency. The model also supports advanced use cases, such as synchronizing distributed systems or managing shared databases, while allowing for flexible consistency models. The team envisions a future where developers work with “concept catalogs”—shared libraries of well-tested, domain-specific concepts—enabling software development to shift from writing code from scratch to selecting and connecting pre-defined concepts. Jackson believes this could mark a cultural shift in software engineering, where systems are built on abstractions that align with human understanding rather than low-level machine operations. Meng adds that the goal is to make software as readable and intuitive as a book, with concepts mirroring real-world actions and synchronizations capturing our natural expectations of how they work together. Experts outside the team have praised the work. Kevin Sullivan of the University of Virginia called it a vital step toward building software that is safer and more understandable, especially in high-stakes areas like healthcare. Thomas Ball of Lancaster University and the University of Washington noted that the model provides a strong foundation for aligning LLM-generated code with human intent, making it a promising direction for the future of software design.