Matija Šošić is the co-founder and chief executive of Wasp, an open-source framework for building full-stack applications with React, Node.js, and Prisma. Working alongside his twin brother Martin, Wasp’s co-founder and chief technology officer, he has made declarative application architecture central to a practical vision of AI-assisted programming: give coding tools a coherent application structure, and their output becomes easier to generate, inspect, and maintain.
Šošić earned a master’s degree in computer science from the University of Zagreb. His earlier career included internships at Singapore’s A*STAR Bioinformatics Institute and Noom, founding TalkBook, working with Techstars London, and leading an engineering team at biotechnology company Lifebit. The brothers began developing Wasp in 2019, committed to it full time around early 2020, and joined Y Combinator’s Winter 2021 batch. Originally short for Web App Specification, Wasp expressed their goal of describing an entire application at a higher level while generating familiar, editable code.
Wasp entered beta in 2023, when Šošić also helped introduce Mage, an AI application generator built on the framework. Mage converts a short product description into a full-stack project with authentication, routes, database models, and frontend and backend code; by October 2023, it had generated 20,000 applications. Šošić positioned it as a customizable starting point for CRUD applications, with developers still responsible for reviewing and extending the result.
- Deterministic scaffolding before model calls. Mage generates predictable configuration, authentication, styling, and application structure without invoking a language model, then separates the remaining work into planning, implementation, and repair. Common errors, such as incorrect imports, can be corrected heuristically. His working demonstration follows a generated task-management application through database inspection and deployment.
- Model specialization for affordable generation. Šošić assigned application planning to GPT-4 and much of the implementation to the cheaper GPT-3.5. Constraining generation to Wasp’s established stack reduced ambiguity, inference costs, and opportunities for inconsistent code.
- TypeScript-first application specifications. In 2026, Šošić reconsidered Wasp’s custom language, concluding that its adoption, positioning, editor-support, and maintenance costs outweighed its advantages. Wasp subsequently introduced TS Spec, preserving declarative full-stack coordination while moving application definitions into familiar, type-checked TypeScript.
Šošić’s position is that more capable coding models increase the importance of readable code, reliable framework conventions, and software that remains understandable after generation. His critique of prompt-only development defends versioned, reviewable engineering against the assumption that prompts alone can replace maintainable application architecture.
