An historical arc I’ve been thinking about: how “programming” keeps drifting upward in abstraction — from hardware to assembly to high-level languages… and now to (some degree of) text & speech with prompts in "vibe coding" with today's AI Devtools (e.g. GitHub Copilot Agent) and (reasoning) LLMs (e.g. GPT-5).
1) Hardware-era thinking (1840s → 1940s)
Ada Lovelace — the world's first programmer — wrote algorithmic “tables of operations” for Babbage’s Analytical Engine — as close to programming a machine’s gears as you could get (Wikipedia). By the 1940s, early systems (ENIAC, ACE) blurred hardware and software: you programmed with plugboards, switches, and very sparse instruction sets; Turing’s ACE design helped move us toward stored-program computing (CHM, rutherfordjournal.org).
Ada Lovelace and the Analytical Engine
Alan Turing and the "Bomba" computer
2) Assembly takes over (1950s → late 1980s)
Symbolic opcodes on top of machine code made programming portable across similar machines — still close to the metal, but far more productive (GitHub Octoverse 2022).
Source: snippet of assembly code from github.com/microsoft/MS-DOS/blob/main/v4.0/src/BOOT/MSBOOT.ASM
3) High-level languages become mainstream (1990s → 2010s)
C/C++, Java, Python, etc. became the default for apps, servers, and OS components. For a nice time capsule: Microsoft recently open-sourced early MS-DOS versions — and yes, it’s assembly all the way down (1.25/2.0 and 4.0 are on GitHub) (GitHub, opensource.microsoft.com, Microsoft for Developers).
Most popular programming languages
4) Text (and voice) as the new IDE (2020s → )
Large language models let more of us “program by describing” — what some call vibe coding: describe intents in natural language, iterate, and let the AI propose code. It’s a new layer of abstraction, not a replacement for fundamentals. (Karpathy popularized the term in 2025; debate continues around where it’s best used.) (Wikipedia, Business Insider)
What changes — and what doesn’t
Abstraction rises (good!), but the enduring skills remain: clear problem framing, good taste in architecture, testing, and the humility to iterate. From Lovelace’s tables to today’s prompts, we’re still telling precise stories to machines — just in progressively more human languages.
Curious: how has your own programming workflow shifted along this curve?