From frustrated emails to 1M users: the MIT origins of Julia

Julia began in 2009 as an MIT research effort to fix a persistent pain point in scientific computing: languages like MATLAB and Python were easy to use but slow, forcing researchers to rewrite working prototypes in C for production speed. Co-creators Viral Shah, Alan Edelman, Jeff Bezanson, and Stefan Karpinski set out to build one language that combined high-level expressiveness with C-level performance, leaning on just-in-time compilation that specializes code to the data types in play. Announced via a 2012 blog post, the open-source language now claims more than a million users across academia, industry, and government.

The project spun out of MIT’s CSAIL into the company JuliaHub in 2015, which shifted from user support toward advancing the language itself. Julia’s track record spans black-hole imaging, climate and ocean modeling, and circuit simulation. Notable wins include a pharmaceutical modeling platform used in Moderna’s Covid-19 vaccine development, an aircraft collision-avoidance system that ran roughly 50x faster than its Python predecessor, and an audio codec Meta built for WhatsApp.

JuliaHub is now betting on AI-driven engineering with Dyad, whose 3.0 release landed in April. Pitched as a “physics compiler,” Dyad steers autonomous agents through simulations and safety analyses while enforcing physical laws that general-purpose AI models tend to violate. Shah frames the goal as agentic hardware design — uploading design documents and having the system compile, verify, and build complex systems like aircraft, with customers such as Boeing already engaged. The company claims it could compress engineering timelines from months to hours.