1972 & Bell Labs

The story

It begins in Murray Hill, New Jersey, in the late 1960s. Bell Labs — the research arm of AT&T, a phone company that had so much money from its government-blessed monopoly that it could afford to fund some of the best basic research on Earth. Out of that culture came the transistor (Week 2), information theory, the laser, the CCD, UNIX, and — for our story — a programming language.

In 1969, Ken Thompson, Dennis Ritchie, and a small team began writing an operating system for a discarded computer that nobody else wanted, the PDP-7. The OS was eventually called UNIX. They wrote it, painfully, in assembly. By 1971 they wanted to move it to a slightly less bad computer, the PDP-11. Rewriting the entire OS in PDP-11 assembly seemed unbearable. They needed a higher-level language — but every existing one was either too slow, too tied to its hardware, or designed by a committee with strong opinions about ALGOL.

So Dennis Ritchie wrote a new one. He started from a language called BCPL via a stripped-down dialect called B. He added types, structures, the things he and Thompson actually needed. He called it C. By 1973 the entire UNIX kernel had been rewritten in C — itself a radical move, since "writing an operating system in a high-level language" was widely considered a fool's errand. UNIX-in-C compiled, ran, and was easy enough to port that it spread, like a beneficial virus, through every university computer science department in the world.

A short timeline

Why C didn't die

Every five years since 1980, somebody has predicted C will be replaced. It hasn't been. The reasons are surprisingly simple:

1. It maps directly onto the chef. A C function call becomes a couple of assembly instructions. A for loop becomes a tiny set of jumps and adds. There is almost no hidden machinery — what you write is pretty much what runs. This is what people mean when they call C a "thin layer over assembly".
2. It compiles to fast code. Because the language doesn't pretend the computer is something it isn't, optimising compilers have decades of practice generating excellent assembly from C source. There is essentially no faster general-purpose language. (Rust is in the same league. Almost nothing else is.)
3. It is the lingua franca. Want to talk to the OS? You go through the C ABI. Want to call a Python library written in Fortran? You go through the C ABI. Want a Swift app to use a database? C ABI. Every language has a "foreign function interface", and every FFI eventually becomes "be compatible with C". It's the Latin of programming.
4. It teaches the truth. Higher-level languages spare you from understanding pointers, memory layout, allocation, stack frames, byte sizes. C insists on those things. After C, every other language feels like a vacation.

It is also, famously, a language with which you can shoot yourself in the foot in fluent and inventive ways. We will be careful. The point of this phase is not to ship production C. The point is to feel the chef.

The family C built

Look at the languages people actually use today. Almost all of them are syntactically and conceptually descended from C. The shape of a function declaration. Curly braces. for, while, if, else. Semicolons. = for assignment, == for comparison. None of those were inevitable — most of them came from C, and survived because everyone who learned C in the '80s carried them into whatever they wrote next.

Pick any of these and read its "hello world". You will be reading a slightly different dialect of the same family. The family is sixty years old now. It hasn't lost relevance. If anything, the C-with-better-safety branch (Rust, Zig) is having a renaissance precisely because, sixty years on, the chef still wants to be talked to in the same dialect.

Why this matters for AI

Pull back the JSON-and-Python wrapping on any AI model and you'll find C/C++ with high confidence. PyTorch's tensor engine: C++. NumPy: C. CUDA kernels: a C-shaped language. Llama.cpp — the project that put a 7-billion-parameter language model on a laptop CPU? It's right there in the name.

The reason isn't ideology. It's that AI workloads are ferociously sensitive to the exact arithmetic the chef does. A 5% slowdown in the inner loop of a matrix multiplication is, at scale, a 5% slowdown of the entire model. Nobody has time for a vacation language to handle the inner loop. The inner loop must be C, or something equivalently bare-metal.

By the end of Phase 2 you will, in a small but real way, be able to look at one of those inner loops and understand what the chef is doing. That is the foundation everything else in the course builds on.

Higher-level languages let you forget the chef. C insists you remember.

What's next

Before we write a single line, we need to understand what happens to it. C is a compiled language, which is one of the reasons it's fast — and it's the dividing line between C and most of the languages people learn first.

Week 10 is The Translator — compilers vs. interpreters, and why a C program is, by the time the chef sees it, no longer C at all.