From machine code to plain English, and maybe one day straight from thought

Every big jump in programming has done the same thing: moved us a little further from the machine and a little closer to how we actually think. Follow that line and you end up somewhere interesting. The next programming language may not be a language at all.

From machine code to C

At the start, you talked to the hardware on its terms. Machine code was raw numbers. Assembly gave those numbers names, but you were still doing everything by hand.

Take the simplest program there is: printing "Hello, world!" on the screen. In assembly, that takes several lines. You tell the processor exactly which registers to fill: one says you want to write, one says where to write, one points to the text, one holds its length. Then you hand the whole thing to the operating system.

Then C came along and the same job became a single line. You name a print function, give it the text, and a compiler works out the registers for you. You still had to manage memory yourself, but you were finally writing logic instead of wiring.

Java and Python

High-level languages pushed further. Java took memory off your plate with garbage collection, so you stopped cleaning up after every piece of data. Its "Hello, world!" is one line too, just a wordier one.

Python stripped away even more ceremony. "Hello, world!" is one short line that reads almost like plain English: the word print, then your text.

Still, every one of these is a language. When I started learning databases, my first query just asked for everything in a table of vendors. That's always been the deal: learn the syntax and the machine does what you say.

Natural language and coding agents

Now the syntax itself is starting to go away. With a coding agent, the instruction can be one ordinary sentence: "Add a page that lists every vendor we haven't paid in 60 days, with a button to send a reminder." The agent writes the code, in whatever language fits.

That's why AI has turned your phone into a full-fledged IDE. Imagine noticing a bug while you're out, describing it in plain words, reviewing the agent's fix on your screen, and shipping it. You don't need a keyboard built for code when the input is sentences.

And I don't think it stops there. In January 2024, Rabbit introduced the r1 at CES, built on what it called a "large action model," an AI that learns how people use apps and does tasks in them. A few months later I wondered what comes after that: a large thoughts model that executes intent prompts?

My guess is the next UI may take thoughts as input and execute on them. Look at the arc again. Machine code, then C, then Java and Python, then plain English. Each step removed some translation between what's in your head and what the machine does. Thought as input is the step that removes it almost entirely.

You'll still need to know what you want and whether the result is right. But the grammar is going away. The thinking stays.