🤖 Jevons and the automated developer

I built a working app using AI and it made me think about the future of software

In 1865, William Jevons noticed something strange. When engineers made coal-powered steam engines more efficient, coal consumption didn’t drop - it exploded. This counterintuitive pattern, now known as the Jevons paradox, shows up throughout history: make something easier or cheaper and people find exponentially more ways to use it.

William Stanley Jevons. Source: Wikipedia

Now, we have to ask… Could something similar happen with software?

If we slash the cost and effort of developing software—thanks to AI-driven tools—will that reduce or increase the amount of software (and the number of developers) we need?

In this essay, I argue that making software development faster and easier will amplify overall demand and lead to more—rather than fewer—opportunities for developers.

Specifically, I will explore five trends that suggest an ever-increasing appetite for software:

1. Accelerated demand for both software and developers.

2. A booming collaborative ecosystem fueled by open-source and lower barriers to entry.

3. Democratisation & untapped talent as AI makes coding more accessible to everyone.

4. The ongoing need for human expertise to handle complex architecture and regulations.

5. The rise of ‘just-in-time’ solutions, where apps are spun up on-the-fly.

Building my dream app

I have been coding since I was ten, writing my first programmes in 1982. I still have my first coding manual at home. But I am no professional coder. By 2011, my engineering team asked me to stop, saying my code was… basically… shit. My CTO at the time gently suggested I should simply explain what I wanted instead and his team would build it for me.

Now, AI has ridden to the rescue. Services like Replit let you use ordinary language prompts to build working software. Many others, like Cursor and Devin, let amateur developers (or anyone, really) create working apps.

My first computer and programming book.

Rabbits, foxes and Malthus

My interest in ecosystem dynamics began when my mother introduced me to Thomas Malthus’s theories about population instability. In the mid-1980s, I brought these concepts to life by programming ecological simulations on my BBC Micro computer. I still find these models fascinating because they reveal complex feedback loops and show how systems change over time.

While BBC Basic was an accessible programming language in its time, it’s now obsolete. Today’s programming landscape presents a different challenge. Even with generous estimates, only about 100 million people worldwide can create applications independently, including roughly 28 million professional developers. This represents a tiny fraction of the global population, highlighting an often overlooked truth that Replit founder Amjad Masad astutely (and casually) observed:

Normies have higher expectations of what computers could do. If you ever taught programming, many people think they can just talk to it. Like they think the comments are talking to the machine. So ChatGPT is kind of a “well duh” moment.

I certainly was not talking to my computer when I was eight. But I am now. Starting with an almost incoherent voice note, I was able to create a working simulator of ecosystem dynamics in under ten minutes.

The TL;DR is that you can play with my simulator here. Let me walk you through how I built it and discuss what it made me realise about AI’s impact on developers.

My predator-play simulation

Claude as my co-developer

I was feeling pretty lazy so I enlisted Claude, a large-language model created by Anthropic, to define the scope of the software. Claude has a voice interface so I jabbered my thoughts as I jogged across the garden to answer the front door. Read my prompt to Claude carefully below. I really didn’t give much thought to it. Between jumbled ideas and poor transcription, I sound a bit boisterous and drunk. (I was neither).

Claude want to use Replit Agent, which is an LLM powered soft web developing tool to develop a really nice quick app. You use a prompt to do this. So I would like it to create an application that shows a Simple Malthusian system of rabbits That Feed off. Rabbits that have babies and then foxes. Foxes eat the rabbits. And you imagine how that feedback loop works, and it should all be charted could you please, write a prompt for me, which I can post into Redcliffe. Do it very detailed.

Claude produced a staggeringly good specification. Honestly, it is a clear professional requirements doc, the kind a good product manager might have produced. An extract is below. The full response is here.

Create a web application that simulates and visualizes a Lotka-Volterra predator-prey system featuring rabbits and foxes. The application should:
Core Features:

Implement a mathematical model where:

Rabbits reproduce at a constant rate
Rabbits die when eaten by foxes
Foxes reproduce proportionally to how many rabbits they eat
Foxes die at a constant rate without food
Use standard Lotka-Volterra differential equations

Visualization Requirements:

Create a main chart showing:

Population of rabbits over time (blue line)
Population of foxes over time (red line)
X-axis: Time in months
Y-axis: Population counts
Interactive tooltip showing exact values

Superspeed developer

I took the prompt and dumped it into Replit.

With a couple of minutes, I had a full running app. It was buggy. Graphs weren’t visible. It took three iterations for Replit to correct this and five-and-a-half-minutes later the app was running correctly. A couple more minutes and it was deployed to the cloud.

The paradox of progress

So what does this mean? Circling back to the Jevons paradox… Essentially, the Jevons paradox highlights a counterintuitive economic pattern: when something becomes cheaper or easier to access, our consumption tends to increase dramatically rather than decrease or level off. This is an example of positive price elasticity of demand—as prices fall, usage rises.