The Crowd Still Gathers for Handmade Code

One of the most read programming questions on Stack Overflow recently was about calculating a 32-bit CRC for a PNG file by hand. 1,171 views on its public counter. Its top answer, a patient walk through the polynomial arithmetic, scored 27, a number almost nothing on today's site reaches. Nobody needs to do this. zlib has computed CRCs correctly since 1995, every PNG library on earth wraps it, and any model you can name will emit the working code before you finish typing the question. By every economic measure the topic is settled, automated, and closed.

Over a thousand people came to watch anyway.

Scan what else drew crowds on the site lately and the pattern repeats until it stops looking like coincidence. Conway's Game of Life, a toy from 1970, at 855 views. "Should I make my own vector in C++ and use that more than the one built in?" at 373, a question whose textbook answer is a flat no, and whose real answer is that building one is how you find out what the built-in one is doing. "What are the requirements for C code that can run without an operating system?" at 274. A thread on hardware interrupts in protected mode. A duplicate, closed question about getting a C compiler for HelloWorld.c that still drew twenty-five comments, and a discussion of implementing post-quantum encryption in plain C that drew twenty-six. The upper floors belong almost entirely to things nobody is paid to build from scratch anymore: checksums, cellular automata, containers, bootstrap code, cryptography.

I wrote in an earlier essay about what the site's closures now say: the fact-shaped middle of programming Q&A has been automated out of scarcity, and the questions that thrive are the ones machines cannot settle. But the viewing figures are measuring something the closures miss. A closed question tells you what people no longer need to ask. Eleven hundred views on a hand-rolled CRC tell you what people still want to watch.

One methodological note, so the numbers are honest: view counts and scores are public counters on every Stack Overflow question, and mine are a snapshot from the stretch when I was writing this. The pattern does not depend on the snapshot. Look in any month and the handmade things float.

Nobody needs the checksum

Be precise about what those views are not. They are not need. A developer who actually needs a CRC in 2026 does not read a long answer about polynomial division; they call the library function, or they ask the model at their elbow and paste the result, and both are the correct professional move. If need drove that traffic, the question would have the readership of its topic's job market, which is to say, nearly none.

They are not nostalgia either, or not mostly. Nostalgia would gather around the old technologies themselves, and there is no crowd gathering around Delphi or SOAP. The crowd gathers specifically around fundamentals built by hand: the checksum computed step by step, the vector implemented badly on purpose, the C program with no operating system underneath it. Not old things. Bare things.

The nearest human activity I can find is watching a potter. Industrial ceramics won a century ago; your cupboard is proof. And yet put a wheel in a shop window and people stop walking. Nobody watching intends to throw their own dinnerware. They are watching because the process is legible in a way the factory is not, because skill exercised openly is inherently worth attention, and because some part of understanding plate only exists in watching one get made. The pot is not the product of the demonstration. The picture in the watcher's head is.

That last sentence is the same one I keep arriving at from other directions. When I argued that getting stuck is the point, the claim was that understanding forms where your picture of the machine meets the machine and loses. Watching a CRC computed by hand is the spectator version of the same event. The reader of that answer is not acquiring a checksum. They are watching bits move through a polynomial until the black box labeled "integrity check" becomes a mechanism they hold, and that mechanism transfers to every checksum, hash, and error code they meet afterward. The answer is a demonstration that the floor under the abstractions is solid and walkable. You do not need to walk it daily. You need to have walked it once.

The asking went, the watching stayed

Here is the split those numbers make visible. The asking has largely gone to the machines, and it should have. Asking is need, need wants speed and privacy, and a model delivers both without judgment at three in the morning. The watching stayed human, and I think it stayed for a reason that matters to anyone who writes or teaches: watching is how the deep stuff actually propagates. It always was. The lecture, the master class, the pair-programming session, the conference live-coding demo that everyone remembers years later, all of it is the potter's wheel, skill made legible in motion. Stack Overflow's answer format at its best was always closer to that than to a lookup table, and the crowd is voting for exactly the answers that perform understanding rather than dispense it.

Which reframes what a question like "should I build my own vector" is for. Read as a request for a decision, it is trivially answerable and mildly embarrassing, which is why the old constitution kept such questions at arm's length. Read as what it actually is, an invitation for someone skilled to take the machine apart in public, it is among the most valuable things left on the site. The 27-point CRC answer was not written for the asker. It was written for the eleven hundred, and the eleven hundred knew to come.

I took the same bet with my own apprenticeship advice: learn to code precisely because the machines can, since the value moved from producing the artifact to holding the picture. These numbers are that essay's argument running in the wild, with viewing figures. When generation became free, the artifact stopped drawing the crowd. The making kept drawing it.

Write for the watchers

If you make things, there is a practical conclusion hiding in those numbers. The handmade demonstration is not obsolete content; it is nearly the only content whose audience the machines have not absorbed, because its value was never the output. Nobody will ever need your from-scratch CRC, your toy allocator, your renderer built over a holiday week. Publish the making of it anyway, with the reasoning visible, the wrong turns kept in. The model at everyone's elbow has taken over dispensing answers, and it is welcome to that job. What it has not taken over is the shop window, the wheel, the clay, and the small crowd that still stops, reliably, a thousand strong on a Tuesday, to watch someone who knows what they are doing make a thing by hand.