Our approach to manufacturing is as quaint as punchcards

Recently, I had the privilege to hang out with some septuagenarian programmers. I was showing off the bicycle shifter I've been working on. I complained to the veterans that I was shocked by the amount of time required to bring a physical design to implementation. It took two weeks to come up with a CAD model; a month and 30 failed attempts to make a usable 3D print; 4 months to manufacture a part out of aluminum.

"That's what programming was like back in the day" was their reply. Really? "You'd turn in your punch cards and hope to get the output a week later — sooner if you were lucky". Can you imagine writing a program like that? I had never considered the analogy to physical design.

It's unimaginable how anyone prototyped physical objects before 3D printing. Debugging a program with one execution attempt per week is similarly inconceivable. We're spoiled these days. We grew up reliant on read-eval-print loops (REPLs), interactive debuggers, quick compilers, and instant gratification. Though we've progressed beyond punchcards, many people think our tools still aren't fast enough. Our generation is obsessed with speed and convenience (perhaps for good reason).

The question is: What will bring physical manufacturing up to parity with software? Earlier this year I would have said it's 3D printing. But 3D printing is slow. Even though laser sintering can produce precision parts like rocket engines, it doesn't scale. SpaceX isn't printing 1,000,000 engines, they're printing 100.

To build cars, cell phones, and soda cans you need to produce high volumes quickly. Waiting for a 3D printer to finish a job isn't cost effective for most terrestrial products. Instead, you need to design for manufacturing with mills in factories where the machine time is measured in seconds, not hours. But today designing for factories is too way too difficult. Our approach to manufacturing is as quaint as running assembly code on punchcards once a week.

Though there are high-level tools to turn CAD models into milling toolpaths, that's only a small portion of the problem. What we need is a way to click a button and launch a manufacturing process: a generalized, automated workflow from CAD to produced parts. Maybe affordable milling tools will help us get there. Maybe cheap, inflatable robots will enable scale.

As much as I loved the Diamond Age, I don't think 3D printers will make it possible. But I think it's a goal worth pursuing. It would be wonderful to tell our grandchildren the same line: "That's what manufacturing was like back in the day". Who knows what they'll be creating then.