“The hottest new programming language is English.”–Andrej Karpathy
This simple statement is taken from his speech “Software Is Changing (Again)” at Y-Combinator captures a major shift in how we build and interact with software.
Software 1.0: Engineers wrote rules, equations, and algorithms.
Example: Writing code to calculate stress in a beam.
Software 2.0: Engineers trained models using data.
Example: Using machine learning to predict material behavior without running every simulation.
Software 3.0: AI becomes an engineering collaborator.
Example: An engineer says, “Design the lightest bracket that meets strength, thermal, fatigue, and manufacturing requirements.” The AI can set up simulations, run optimization studies, evaluate results, and recommend the best design.
But engineering is different from many AI applications.
The answer cannot simply look right. It must obey physics, mathematics, boundary conditions, manufacturing constraints, and engineering principles.
That is why mechanistic AI matters.
By combining foundation models with science and engineering knowledge, we can build systems that are not only faster, but also more reliable and trustworthy.
The future is not AI replacing engineers. It is engineers using AI to manage complexity while humans focus on objectives, constraints, trade-offs, judgment, and accountability.
The real challenge is not building smarter AI. It is building AI that engineers can trust.
What engineering workflows do you think Agentic AI will transform first?
