JuliaHub launches Dyad 3.0 as Julia passes 1 million users
JuliaHub has released Dyad 3.0, the latest version of its AI platform for engineering teams developing complex physical systems. The launch comes as Julia, the open-source programming language that began as an MIT research project, has grown to more than 1 million users across thousands of companies and universities worldwide.
Dyad 3.0 is designed for work on systems such as rockets, heat pumps and satellites. JuliaHub says engineers are already using Dyad to direct autonomous AI agents through physics simulations, safety analyses and quality controls. Viral Shah, JuliaHub co-founder and CEO, said users can upload data and design documents for the system to work through an aircraft design process.
From MIT research project to industrial software
Julia emerged from discussions among researchers around 2009 about the limits of scientific programming tools. The goal was to create a language that was accessible for scientists and engineers while delivering the performance needed for demanding numerical workloads. Its creators included Shah, MIT mathematics professor Alan Edelman, Jeff Bezanson and Stefan Karpinski.
The language uses just-in-time compilation based on the types of data being processed. MIT describes this as the basis for Julia’s combination of speed and flexibility in scientific research, data analysis and modelling. The researchers announced Julia in 2012, and demand for support later led them to establish JuliaHub in 2015.
Julia has since been applied to modelling atoms, semiconductors, neural networks, race cars and aircraft, as well as black-hole imaging. MIT also cites applications in computer-circuit simulation, health-disparity detection, climate and ocean modelling, and brain-activity analysis. A pharmaceutical modelling platform built in Julia was used to accelerate Moderna’s Covid-19 vaccine development.
In another cited case, a Julia-based aircraft-collision avoidance program ran about 50 times faster than an earlier Python version. Meta engineers also used Julia to develop an audio codec for WhatsApp’s 4 billion users. These examples show how a language initially targeted at interactive research workflows has moved into a broader range of industrial, government and academic settings.
Physics constraints are central to Dyad
JuliaHub launched Dyad 1.0 in June 2025 as a research agent for faster programming, followed by Dyad 2.0 in December. The company positions version 3.0 as a more autonomous system for physical-system design rather than a general-purpose AI assistant.
Shah describes Dyad as a physics compiler that enforces physical laws. He said general AI systems can produce proposed solutions to physical problems that violate those laws, while the Dyad agent is intended to identify such violations and guide work toward physically correct solutions. The company expects this approach to reduce product-engineering design times substantially.
For businesses, the release highlights a practical distinction in engineering AI: agentic workflows need domain models, physics-based checks and formal verification alongside automation. Teams evaluating such platforms should define where generated outputs enter simulations, safety reviews and quality-control processes before relying on them in design work.