A five-person startup based in Paris and San Francisco is emerging today with the audacious claim of being able to create enterprise-grade software with generative artificial intelligence entirely from natural language descriptions, slashing development times by 80% in the process.

Shreds.AI is self-funded and unknown, but nothing is modest about its ambitions. It wants to “revolutionize software development by automatically generating the full software, from architecture to code,” said founder and Chief Executive Soufiane Amar. “The developers’ role is only to review and validate the generated code.”

The company says it achieves these efficiencies by harnessing the power of up to a dozen generative AI models, drawing selectively on the best features of each through a process called meta AI. “We are not trying to create a big model that builds everything,” Amar said. “Instead, we’ve built a small model that is trained to emulate the software engineering process.”

As input, Shreds accepts everything from a general description to a detailed specification, with the latter producing better results. If the description is vague, the generative AI chatbot walks the user through the process of adding detail. It then generates an architectural diagram and microservices called shreds from its components.

“Each service is divided into multiple shreds, and each shred can be easily described and validated,” Adam said. “Shreds manages complexity so it understands which components interact with which.” He demonstrates the code generation process in this video.

WordPress in three hours

As proof of the product’s effectiveness, Shreds.AI plans today to post the source code of a fully functional version of the WordPress content management system rewritten in Java by Shreds. The WordPress Foundation estimates the CMS is the product of 91 person-years of development, but Amar said his generator completed the job in less than three hours. Next week, it plans to release the source code of a clone of the Magento e-commerce platform.

The company is incorporating the human element through a testing process that submits generated code to human developers for validation at fees set by the testers. “Our code is 90% to 95% error-free, but there’s no code that can go to production without human validation at some level,” Adam said.

Shreds.AI said its technology is already in beta test at automaker Stellantis N.V. and French utility Réseau de Transport d’Électricité Ltd. “We’ve used Shreds.AI for successfully generating some internal software, and we are considering integrating it into our development workflow to gain speed and efficiency,” said Jérôme Menard, datadev department manager at RTE, in a prepared statement. SiliconANGLE submitted questions asking for more details about beta testers’ experiences and will update this article with their comments.

The company is self-funded and plans to seek Series A financing soon.

Photo: Pixabay