History of Programming: What Does It Mean to Be a Programmer?

With each step, the programmer said more about what they wanted and less about how. Is there a point where saying "what you want" stops being programming?

Hu argues that time sharing turned programmers into "users," made to feel like intimate partners of the computer while the other users on the same machine disappeared from view. Galloway argues that the web works like continuity editing in film: it hides its own apparatus so you never notice the cuts. Each step up the staircase hides more of the machine. Is that progress, or continuity editing? At the top step, programming in English, what is hidden from you?

ENIAC had no stored program. To give it a new problem, people worked out the math on paper, then set switches and plugged cables by hand. Setting up one problem could take days to weeks.

Its first programmers were six women: Kay McNulty, Betty Jennings, Betty Snyder, Marlyn Wescoff, Fran Bilas and Ruth Lichterman. The work was considered clerical. The engineers who built the hardware got most of the credit.

To program, you had to understand the machine's circuits.

ENIAC's first programmers were treated as clerical workers, and the people who built the hardware got the credit. What kind of work in programming today gets treated as "not the real work"?

↑ Back to 75 Years of Programming

• Input: a panel of 32 buttons and switches, one for each bit of a line of memory. The operator picked a line, set each bit to 0 or 1, then moved to the next line.
• Memory (you could actually see): the memory was a cathode ray tube. Each bit was a charged spot on the screen, a dash for 1 and a dot for 0.
• Run: press start. The machine stepped through memory line by line.
• Output: a display tube showed the pattern of dots and dashes. The operator read the answer off the screen, in binary.

The Baby's programmers thought in bits (0s and 1s) because the switches were bits (on/off 0s/1s). What does today's interface (a text editor, or a chat with an AI) make you think in?

↑ Back to 75 Years of Programming

Batch programmers read their code carefully because running it was expensive. Running code costs almost nothing now. What happens to reading?

0001 0110
0010 0111
0011 0101
1111 0000

        LOAD  A     ; copy A into the register
        ADD   B     ; add B to it
        STORE X     ; save the result in X
        HALT        ; stop the program
A:      DATA  2     ; the first number
B:      DATA  3     ; the second number
X:      DATA  0     ; space for the result

X = A + B

let x = a + b;

Add the two numbers and show me the result.

↑ Back to 75 Years of Programming

• FORTRAN (1957, "FORmula TRANslation"): for scientists and engineers.
• LISP (1958): for artificial intelligence research.
• COBOL (1959): for business. Designed to read like English sentences so managers could follow it.

↑ Back to 75 Years of Programming

With the web, open source and package managers, most of the code in a project is code someone else wrote.

Programming became finding the right library, reading its documentation, and connecting the pieces.

GitHub and Stack Overflow (both 2008) made other programmers' code and answers searchable. Copying and adapting became a normal, even expected, part of the job.

↑ Back to 75 Years of Programming

GitHub Copilot (2021) suggested the next lines of code. ChatGPT (2022) produced whole programs from a conversation. Coding agents (2025) write, run and fix code across an entire project.

In early 2023, AI researcher Andrej Karpathy wrote: "The hottest new programming language is English."

English (or natural language) functions as the interface to code that a machine runs. Arguably the skill moves towards precision in English: saying exactly what you want, then checking that you got it.

What does precision mean?

↑ Back to 75 Years of Programming

Fill in the "?" row of the criticism table for AI models. What is the worry, and where is the skill moving?

Why or when would someone prefer to think more like the machine?

• decides what the building is for and who will use it
• understands structure well enough to know what will stand up
• makes the plan precise enough for builders to follow
• inspects the work and catches what is wrong

• Give it a sketch or an image of what you are looking for.
• Give it sample code (from this class, a tutorial, or an existing website) to pinpoint the functionality you want.
• Give it the actual data or files it should work with, not a description of them.
• Say what must stay the same, so it does not change the parts that already work.
• Give it a test the result has to pass, for example: "clicking the button adds the item to the list and clears the input."

An architect doesn't lay bricks but still has to understand structure. How much do you need to understand about code to judge what an AI hands back? How would you know if you don't know enough?

• ENIAC's first programmers were treated as clerical workers, and the people who built the hardware got the credit. What kind of work in programming today gets treated as "not the real work"?
• Batch programmers read their code carefully because running it was expensive. Running code costs almost nothing now. What happens to reading? Do you read the code an AI gives you?
• Fill in the "?" row of the criticism table for AI models. What is the worry, and where is the skill moving?
• An architect doesn't lay bricks but still has to understand structure. How much do you need to understand about code to judge what an AI hands back? How would you know if you don't know enough?
• With each step, the programmer said more about what they wanted and less about how. Is there a point where saying "what you want" stops being programming?