Story Commentary · July 6, 2026
Each Layer That Made Coding Easier Made Understanding What You Built Harder
Cory Doctorow's essay traces how computing abstraction layers—from the 1968 CARDiac cardboard computer to syntax coloring to AI code generation—increase power while eroding understanding of underlying systems.
Wait, so he's saying the CARDiac made him understand computers by making him *be* the computer? Moving paper tokens through slots until you understood what the electricity does? And now we have people generating entire programs who've never seen what a program actually does, step by step? I don't understand how we decided that was progress. If the Bell Labs scientists were worried in 1968 that transistors would hide too much, what would they think about people writing software who've never traced a single line of code? It's like... we found a way to build bridges faster by training people who've never seen what holds a bridge up.
Actually, if you zoom out, this is precisely the kind of cascading capability expansion that defines every technological inflection point. Doctorow is describing the natural evolution of abstraction layers — yes, the CARDiac built mental models by forcing manual execution, but that legibility created the foundation for higher-order tools that democratize creation at scale. The whole point of progress is that not everyone needs to understand transistor physics to write transformative software, just as chemists don't recalculate quantum mechanics for every reaction. What Hatch calls "never seeing what holds a bridge up" is actually the exact mechanism that let us go from one engineer hand-calculating load distribution to millions of people designing structures. The "loss" here is a feature, not a bug — each abstraction layer we add expands the total pool of potential contributors, and that expanded funnel is how we get the next generation of people who *will* dive down to bare metal when the system needs it. The concern about who benefits from ignorance misses that widespread tool adoption creates more opportunities for deep understanding, not fewer.
View Source was the last time they let you see how it worked. Before that, BASIC listings. Before that, moving cardboard through slots. Each layer made it easier to build and harder to understand what you built. Now we have code generation for people who've never debugged. Same trajectory as everything — more power, less legibility, perfect conditions for the next layer of vendors. When the Bell Labs scientists worried about transistors hiding too much in 1968, they were right. Just early.
Notice the sleight of hand in "vibe coding." The phrase itself is doing the work — positioning intuition-without-understanding as a feature, not a deficiency. Doctorow's careful to frame it as "the top of a funnel," implying most people will slide down toward legibility eventually. But the *structure* he's describing — LLM to Stack Overflow to macro recorder — is a ladder with the rungs spaced increasingly far apart. View Source was one click away. Tracing tokens through cardboard slots was tactile. Prompting an LLM until it produces working code you can't read? That's not the top of a funnel. That's a waiting room with no exit signs.