A single physical part generally has many, many inputs and outputs (expressed as its physical contact surfaces, tolerances, forces and stresses of all possible kinds from all directions they are supposed to withstand, etc., all with the added dimension of time) compared to a line of code or even a function. You cannot design a 'functionally pure' part, or design a part that creates itself for every discrete 'use'. And unlike a line of code, it can't be quickly replaced so you must get it right. And if you get it wrong, people may die in this particular example. Etc.
And that's before we become concerned with actual production of that part, whereas all you need for to write a line of code is a text editor or pen and paper.
Most human beings can handle, at most, 5 to 9 variables in their head at once. Software systems grow to immense complexity of thousands to millions of variables, with various runtimes, databases, caching layers, frameworks, libraries, coding styles accrued over multiple decades in some cases, data aggregation pipelines, dozens to hundreds of APIs with their own unique interfaces, etc. It is a constant battle against entropy, which happens inevitably in such systems.
Meanwhile, there isn't a single modern "real engineer" who doesn't get their daily work done thanks to multiple software programs supporting their own productivity so much that they can replace what used to be teams of dozens to hundreds of assistants, techs, engineers, and other various human support systems. They draw lines on AutoCAD that automatically compute moments of inertia, shaft calcs, bill of materials, etc. Then they take that file output and zap it over Slack instantly to their colleague who works from home hundreds of miles away to get their feedback on it. Behind all of that is a team of software engineers and product devs that not only has to understand the constraints of their own technical systems, but the technical needs of their engineering clientele.
