Sleeping Giants

Ten years ago, I observed that greed undermines software craftsmanship. Today, that’s truer than ever.

It’s been said that software craftsmanship is a dying art. That LLMs are steering software development toward languages that are easily generated. Languages not chosen for their elegance or fitness, but for their compatibility with a model’s training set.

Friends of mine feel it too. Their teams are ditching purpose-built languages like Erlang in favor of generic options like Java. Not because they’re better, but because LLMs can churn them out on command.

Others point out that coding isn’t the bottleneck in product development. I agree. But here’s the uncomfortable truth: businesses are ravenous for software, and even LLMs won’t be enough to feed them.

While the world gluts itself on code, the craftsman waits.

LLMs will flood the industry with net-new code. They’ll subvert human-crafted systems with generative runtimes — inferred, not designed. Sure, we’ll get longer context windows, tighter quantization, and smarter training. Tools will improve. Access will expand.

The trouble is, we treat LLMs like scalpels, but they land like hammers.

In a few years, we’ll be knee-deep in bloated, broken code and deranged prompts praying for salvation. For most, it will be a nightmare to maintain.

As our culture of credit collapses under the weight of this generative mess, the craftsman will still be there, quietly sharpening his tools. Sooner or later, the note will come due on the trade that swapped CPU precision for GPU-powered approximation. And when it does, the craftsman will be prepared to refactor from first principles.