Disposable Software

In 1750, a shirt was estimated to cost around $3500 adjusted for inflation. Making one took time and skill - nearly 500 hours to spin the thread, sew the cloth, and hem the shirt.

Today, you can order a t-shirt on Shein for under $5.

The industrialisation of manufacturing processes has brought tremendous gains in productivity and prices, along with massive labour disruption and environmental impact. Software looks to be following the same trend.

The historical analogy is useful here. Steam power and the Spinning Mule transformed textile production and altered the economics of production. Cotton became cheaper than wool and the distribution of labour shifted from being 80% weaving the yarn to to 80% sewing and tailoring.

Over the next 200 years, mass industrialisation moved along the value-added chain - from raw inputs to assembly and now distribution. Today, the biggest line item from many fashion brands is marketing and advertising.

The costs and benefits of mass industrialisation are a complex mix of tradeoffs. It’s never been cheaper and easier for consumers to buy clothing but the quality is commensurate with the price. Clothes used to literally last a lifetime - now, fast-fashion tees barely make it past a few washes.

Once, each jacket or pair of trousers was unique and individually tailored for a perfect fit. Now oversizing ensures that just a handful of cloth patterns fit everyone badly. And while nobody wants to go back to paying thousands for a shirt, today’s cheap prices hide significant environmental and social damage that will need to be paid for eventually.

Just like fast fashion, we now have disposable software. GenAI is the new Spinning Jenny that is transforming software development from a craft to an industrial process. Like industrialisation, it’s making the generation of code cheaper and more accessible but it’s doing so by reducing quality, increasing homogenity, and plundering the environment.

Like oversized hoodies, disposable software is homogenous. LLMs are pattern guns that produce the most statistically likely output given some input. Prompt it to “build me a SAAS app that does Tinder for horses” and you’ll almost certainly get a slight variation of the same React-Tail-Strap app used by everyone else’s Next Big Idea. It works well enough for simple cases, but the moment you need to customize it or add complexity, that generic foundation becomes an impossible to maintain AI slop of spaghetti code.

But at least it will be cheap. At $20 in tokens, you can throw it away and try again.

This isn’t a moral argument - although I do have strong moral opinions on the subject. It’s an observation, a statement of fact. And like the industrialisation of textiles, this transformation creates both opportunities and risks for everyone involved. Understanding how to navigate this new landscape - what to embrace and what to avoid - is something I’ll explore in Part 2.