Session 1 · 75 minutes · slides

Choosing a Programming Language

What software is written in, and how the choice is made

1.1 What a programming language is
1.2 Fit for purpose
1.3 Popularity, maturity, and community
1.4 The ecosystem: packages and libraries
1.5 Speed and memory
1.6 Platforms and portability
1.7 Making the decision in the age of AI

Open the book chapter Download as PowerPoint

Arrow keys or space to move, F for full screen. Every slide has a Book button back to the full text.

From human ideas to machine instructions

Processors understand only machine code: numbers
A programming language is a human-readable notation with strict grammar (syntax)
What the programmer writes is source code
Something must translate source code into machine code

Two ways to translate

Compiler: translate the whole program ahead of time into an executable. Fast; tied to one platform. C, C++, Rust, Go, Swift
Interpreter: carry out the source code on the spot. Portable; slower. Python, JavaScript, Ruby
Analogy: a published translation versus a live interpreter in the room

Language, library, framework

Language: the notation itself (Python)
Library: ready-made code for one job (send email, draw a chart)
Framework: the skeleton of a whole kind of application (Flask, React)
Hundreds of languages exist because each makes one kind of work easier at the cost of another

In the age of AI: syntax is the part AI has fully absorbed. The trade-offs below have not gone away.

Was the language built for the job?

General-purpose languages can write almost anything: Python, JavaScript, Java, C#, C++, Go, Rust
Domain-specific languages do one thing superbly: SQL (database questions), HTML (structure), CSS (style)
Nobody writes a website in SQL: it cannot draw a button, react to a click, or send an email

Where each language lives

Where each language lives. Some places admit only one language (the browser, the database). The server admits almost any, which is why the back-end choice is the one most worth thinking about.

Browser: only JavaScript. Database: only SQL
The server admits almost anything: that is the real choice

A product is never one language

Typical web product: HTML + CSS + JavaScript in the browser, Python (or JS) on the server, SQL in the database
Add Swift and Kotlin for native phone apps
"Choosing a language" really means choosing the back-end language and the mobile strategy

In the age of AI: fit for purpose is unchanged. Using several languages in one product is cheaper than it was.

A language is a community, not just a notation

Hiring: millions of developers versus a few thousand
Answers: every problem in a popular language has been solved publicly
Longevity: COBOL from the 1970s still runs banks; who maintains it now?
Security: many eyes find and fix flaws quickly
Tooling: editors, testing, hosting support the popular languages first

The mainstream, and the caution about the new

Mainstream today: Python, JavaScript/TypeScript, Java, C#, C/C++, Go, SQL; Rust, Swift, Kotlin in their niches
A brand-new language may be better designed and still be the wrong choice: few libraries, breaking changes, unproven survival
Most successful languages needed ten years to become safe bets

In the age of AI: popularity matters MORE. AI learns from public code, so mainstream, stable languages get the best help.

Nobody writes it from scratch

Email, spreadsheets, payments, images, charts: someone already wrote it and published it free
The ecosystem = the libraries plus the tools to find and install them
Package manager: pip (Python), npm (JavaScript). "We added a dependency"
Almost all of it is open source, which is also a supply chain to watch

The ecosystem steers the choice

If the product is mostly about…	The ecosystem pulls toward
AI, machine learning, data	Python
Rich interactive web front end	JavaScript / TypeScript
Conventional web back end	Python, JavaScript, Ruby, Java, C#, Go
Enterprise integration	Java, C#
Games	C# (Unity), C++ (Unreal)
Native phone app	Swift / Kotlin

In the age of AI: ecosystems matter as much as ever. Watch for hallucinated packages that do not exist.

The fastest language is usually the wrong question

Compiled languages can be 10 to 100 times faster than Python at raw computation
But most business apps wait (database, network, disk); they do not compute
Developer time costs far more than machine time
Python's AI libraries are C inside: "Python is slow" and "Python runs AI" are both true

When performance genuinely decides

Situation	Why	Typical choice
Embedded devices, wearables	Kilobytes of memory, a battery	C, C++, Rust
Games, video	16 ms per frame	C++, C#, Rust
Trading, real-time control	Microseconds are money or safety	C++, Rust
Very large services	10× efficiency = thousands of servers	Go, Rust, Java
Mobile apps	Battery and memory limits	Swift, Kotlin

Build in the productive language, measure, then rewrite only the proven hot spot. In the age of AI that last step is cheap.

A platform is processor + operating system

Processor families: x86 (Intel, AMD) and ARM (phones, Apple silicon, new servers)
Operating systems: Windows, macOS, Linux, iOS, Android
Machine code for one platform is gibberish to another
Every extra platform is another version to build, test, and support

Three ways to reach three platforms

Three ways to reach three platforms. Compiled languages produce a separate program per platform. Java-style languages compile once to an intermediate form that a virtual machine runs anywhere. Interpreted languages ship the source itself. Each step down trades speed for portability.

Recompile, virtual machine, or interpret

Recompile per platform (C, C++, Rust): fastest; one build per platform. Apple's 2020 move to ARM forced every Mac app to be rebuilt
Compile once to bytecode (Java, Kotlin, C#): "write once, run anywhere" on a virtual machine
Interpret the source (Python, JavaScript): ship the file itself. JavaScript's interpreter is in every browser, so the web is the most portable platform ever

Phones are the special case

Apple prefers Swift; Google prefers Kotlin. Both compiled for the phone
The best app on both platforms means writing it twice
Cross-platform frameworks (React Native, Flutter) produce both from one codebase, with some loss of polish
Electron does the same for desktop (Slack, VS Code, Discord)

In the age of AI: porting code between platforms is cheaper. Testing on every real device is not.

How each factor shifts

Factor	Before AI	With AI assistance
Fit for purpose	Decisive	Unchanged
Popularity and maturity	High	Higher: AI quality tracks public code
Ecosystem	High	Unchanged to higher; verify packages
Speed and memory	Decisive in a few domains	Same domains; elsewhere lower
Platforms	High (duplicated work)	Lower for writing; same for testing
Your own familiarity	Often decisive	Lower, but you still review the code

Three conclusions

Choose mainstream. Python, JavaScript/TypeScript, SQL; Java or C# when the enterprise requires; Swift and Kotlin for native mobile
Optimise for readability. When AI writes the code, the human's job is reading it
The language is rarely the risk. Products fail on requirements, data models, and security, not Python versus Go

This module uses HTML, CSS, JavaScript in the browser; Python on the server; SQL in the database.