The discourse around AI and employment has become predictably binary. On one side, breathless predictions of mass unemployment. On the other, techno-optimists insisting everything will be fine. Both camps are likely wrong, but not in the way you might expect.

A Brief History of Efficiency Panics

In 1865, economist William Stanley Jevons observed something counter-intuitive about coal consumption in England. As steam engines became more efficient, requiring less coal per unit of work, total coal consumption didn’t fall. It exploded. Cheaper energy meant more applications became economically viable, driving demand far beyond what efficiency gains had saved.

This pattern, now known as Jevons’ Paradox, has repeated throughout technological history. ATMs didn’t eliminate bank tellers. They made branches cheaper to operate, so banks opened more of them. Spreadsheets didn’t eliminate accountants. They created entirely new categories of financial analysis that hadn’t previously existed.

The scaremongers predicting AI will destroy software engineering jobs are making the same mistake Jevons identified 160 years ago. They’re looking at the numerator (labour per unit of output) while ignoring the denominator (total output demanded).

The Coming Explosion in Software Demand

Consider what happens when AI reduces the cost of producing software by an order of magnitude. Suddenly, custom applications become viable for problems that never justified the investment. Internal tools that companies tolerated being broken get rebuilt. Features that sat in backlogs for years become quick wins. Entire new categories of software, previously inconceivable due to development costs, emerge.

We won’t need fewer software engineers. We’ll need them to do different things at vastly greater scale. The demand curve for software is nowhere near saturated.

But Here’s What Everyone’s Missing

The optimistic reading of Jevons’ Paradox assumes the benefits of increased demand distribute evenly across the existing workforce. This assumption deserves scrutiny.

What’s actually emerging is a bifurcated industry structure. A relatively small cohort of engineers are building the AI systems, the foundational models, the inference infrastructure, the tooling that multiplies everyone else’s productivity. Call them the inner ring.

Then there’s everyone else. Engineers whose primary interaction with AI is as consumers of these systems. They’re dramatically more productive than their pre-AI counterparts. A single developer can now accomplish what previously required a team. But that’s precisely the problem.

When one person can do the work of five, companies don’t necessarily hire five times the output. They might hire one person and pocket the difference. Or they hire three people and triple their output while cutting per-engineer costs. The individual engineer is more productive but has less leverage.

The Wage Compression Thesis

This dynamic creates a specific economic pressure. The inner ring, those building the machines, command premium compensation because their skills are genuinely scarce and their work is the bottleneck on everyone else’s productivity. Demand for their labour is inelastic.

For the outer ring, supply and demand work differently. Yes, overall demand for software development will grow. But the barrier to entry drops. Someone with modest programming skills augmented by AI can now deliver work that previously required years of experience. The talent pool expands faster than demand.

The result isn’t mass unemployment. It’s wage compression and a flattening of the career ladder. Junior-level productivity catches up to senior-level productivity faster than ever before. The mid-career premium shrinks. Seniority still matters, but it matters less.

The New Scarcity

In this landscape, what becomes genuinely valuable?

First, the ability to build and improve the AI systems themselves. This requires deep technical knowledge that AI cannot easily augment because you need to understand the tools at a level beyond what the tools can explain.

Second, judgement. Knowing what to build, why to build it, and whether it should be built at all. AI can generate code quickly, but it can’t tell you whether the feature makes strategic sense or the architecture will scale with your business.

Third, system-level thinking. As individual components become easier to produce, the complexity shifts to integration, to understanding how pieces fit together, to managing emergent behaviour in large systems. The person who can hold the whole picture in their head becomes more valuable, not less.

Fourth, trust. When AI can generate plausible-looking code instantly, the ability to verify, to know when something is subtly wrong, to take responsibility for outcomes. These become premium skills. Companies will pay for engineers they trust not to ship AI-generated nonsense.

What This Means Practically

If you’re building your career in tech, the strategic implications are clear.

Don’t compete on the dimension AI compresses. Producing more code faster is a race to the bottom. You won’t beat AI at being AI.

Invest in depth over breadth. Shallow knowledge of many tools becomes less valuable when AI can provide shallow knowledge instantly. Deep expertise in specific domains, the kind that lets you spot when AI is confidently wrong, becomes the differentiator.

Move toward the inner ring if you can. Understanding how these systems work at a fundamental level is an appreciating asset. This doesn’t mean everyone needs to train LLMs, but understanding the tools deeply enough to extend and customise them creates leverage.

Focus on the parts of the stack where failure is expensive. The closer your work is to critical systems, production infrastructure, security, data integrity, the more your judgement and experience matter relative to raw output speed.

The Honest Uncertainty

None of this is certain. Jevons’ Paradox suggests demand will grow, but it doesn’t guarantee where that demand concentrates or how compensation distributes. We’re in genuinely uncharted territory.

What seems clear is that the binary narratives, mass unemployment versus business as usual, both miss the structural changes happening beneath the surface. Jobs won’t disappear, but the shape of those jobs and the rewards they command will shift in ways that advantage some skills and experiences while disadvantaging others.

The engineers who thrive will be those who understand which side of that divide they’re on and position themselves accordingly. The ones who struggle will be those who assume their current skills remain valuable simply because they always have been.

Efficiency revolutions create abundance. But abundance changes who captures value from work. That’s the lesson Jevons taught us. We’re about to learn it again.