More Work, Fewer Doors: The Jevons Paradox Nobody Is Talking About

The organisations making the clearest decisions about AI and technical talent right now are doing one thing differently: they are asking two questions at once. Not just how does AI change how much software we need? But how does AI change who we need to build it?

The answer to the first question is settling into something reassuring. Jevons’ Paradox is real, and the emerging data bears it out. When a resource becomes cheaper and easier to use, total demand tends to increase rather than decrease. Software development is following the pattern. AI coding tools cut the time to produce a given piece of code; the result is not fewer projects, but more. The addressable surface area of what organisations can build is expanding.

The answer to the second question is where the strategic picture gets less comfortable.

The Paradox Working at Two Speeds

Anthropic published its Labour Market Impacts of AI research this week, and the headline finding is being widely reported as broadly reassuring: no systematic increase in unemployment in AI-exposed occupations. That framing is accurate. It is also incomplete.

What the research actually shows is a two-speed market. At aggregate level, AI-exposed professions show no meaningful employment disruption. But among workers aged 22-25, the job-finding rate in AI-exposed occupations has dropped 14% in the post-ChatGPT period compared to 2022. The Anthropic researchers are careful: this finding is only marginally statistically significant, and alternative explanations exist. Still, it maps cleanly onto a pattern visible in the UK labour market independently of the US data.

Adzuna data cited by techUK shows a 30% fall in UK entry-level job postings since ChatGPT’s launch. Harvey Nash’s December 2025 UK software market analysis is blunt about the mechanism: senior developers working effectively with AI tools can now often deliver the same output that previously required additional junior support.

This is not a staffing problem. It is a pipeline problem with a delayed fuse.

What Jevons Actually Says

William Stanley Jevons observed in 1865 that more efficient steam engines did not reduce coal consumption; they increased it. Lower operating costs unlocked demand that had previously been suppressed by price. The paradox describes an efficiency gain at one level of the system generating volume growth at another.

Applied to AI and software development, the mechanism works like this: AI tools lower the effective cost of building software, which unlocks a backlog of projects that were previously unaffordable, which increases total demand for software development capability. Top VCs including Thrive Capital report no engineering layoffs in their portfolios, consistent with the Jevons prediction: as coding becomes cheaper, organisations build more, not the same amount with fewer people.

Microsoft CEO Satya Nadella drew attention to this pattern publicly in early 2025 and it has since become a standard framing in tech leadership circles. It is not wrong. The paradox is real and the directional logic is sound at the aggregate level.

Where the Paradox Becomes Asymmetric

The part that does not get enough attention is what happens to the bottom of the capability pyramid while the top expands.

Jevons’ Paradox describes a system-level effect. It does not guarantee that efficiency gains are distributed evenly across experience levels. In software development, the evidence suggests they are not.

The Anthropic research shows that AI coverage of computer and mathematical tasks sits at 33% in real-world observed use, against a theoretical ceiling of 94%. The gap between those numbers is revealing: the tasks currently being automated are the more routine, repeatable, and standardised ones. Which is precisely the territory that junior developers have historically occupied on their way to becoming senior ones.

Harvey Nash is specific about the structural consequence: “Senior developers who can work effectively with AI tools can often deliver the same output that previously required additional junior support. This has led to fewer opportunities at the graduate and junior levels, creating concerns about the future talent pipeline.”

In other words: the efficiency gain is being captured at the senior level. The volume growth that Jevons predicts is real. But the jobs that would historically have served as the training ground for future senior developers are the first to be rationalised.

The Strategic Question Not Being Asked Loudly Enough

PwC found that workers with AI capabilities commanded a 56% salary premium over those without in 2025, up sharply from 25% the previous year, according to techUK. That data point is typically read as a workforce development signal: upskill your people. The more useful reading for a senior leader is structural.

Where is the next generation of AI-capable senior developers going to come from?

Senior developers became senior by doing junior work first. Much of it was exactly the repetitive, constrained, well-defined-problem work that AI now handles competently. If that entry point narrows significantly, the pipeline to senior capability narrows with it. Not immediately; the existing cohort of experienced developers does not disappear overnight. But in five to seven years, when you need to replace departing senior staff, the pool you are drawing from will have been shaped by the current entry-level contraction.

This is not a problem unique to technology companies. Any organisation with internal technical capability faces the same arithmetic: publishers building digital and data teams; higher education institutions managing their own platforms; professional services firms running internal transformation programmes. The talent pool is shared, and it is being quietly thinned at the base.

What the Organisations Getting Ahead of This Are Doing

The Head of Technology at a professional services firm navigating this well is doing three things. They are preserving structured junior intake even when the short-term economics argue against it, because they understand the pipeline mathematics. They are redesigning what that intake looks like: less “do this repetitive task” and more “work alongside AI to deliver this more complex outcome, then explain the reasoning behind the decisions made”. And they are building AI fluency into career architecture explicitly, so the investment signal is visible to the people they want to retain.

This is not altruistic. It is supply chain thinking applied to human capability. The distinction that matters is AI as tutor and amplifier for a developing engineer versus AI as replacement for a developing engineer. Both configurations are technically possible. Only one of them compounds over time.

The other thing organisations getting this right are doing: they are not conflating the efficiency decision with the pipeline decision. These are separate choices with separate time horizons. Capturing efficiency gains from AI tools is a decision measured in months. Pipeline investment is measured in years. Running them through the same budget review tends to produce the wrong answer on the longer one.

Six Months from Now

The organisations that look clearest on this will have separated two decisions that are currently being conflated. The efficiency gains from AI tools are real and should be pursued. The pipeline investment is the strategic bet that those gains will compound rather than deplete over time.

The Jevons Paradox says total demand for software development capability will increase as AI makes it cheaper to produce. It does not say the supply of people capable of meeting that demand will automatically follow. That gap is the one worth planning for now, before the market forces your hand on a five-year lag you cannot see yet.

Worth considering whether your current junior intake decisions are being made by the same person responsible for your technical capability three years from now.

Related reading: The Creativity Paradox applies the same Jevons mechanism to creative professionals. The Mentoring Gap examines what happens when AI replaces the development relationship between managers and junior staff. Building Your Future in the Age of AI offers practical guidance for young professionals navigating this landscape. Ideas Were Never the Bottleneck is the other side of the same shift: who can now build, even as the junior rungs erode.