Is the Decline in Young Programmers Entirely Due to AI?

Don’t jump to conclusions just yet.

“AI stealing young programmers’ jobs” already sounds like a foregone conclusion: the number of employed software developers aged 22 to 25 has fallen by nearly 20% over two years, employment changes in software developer and customer service roles have been described as a generational divide, and relevant research from 2025 has also been used to support this observation.

But in the same document, there is another, even more counterintuitive set of data: from 2022 to early 2025, the unemployment rate for jobs least affected by AI rose by 0.94 percentage points; conversely, for jobs most affected by AI, the increase was only 0.30 percentage points. If higher AI exposure is supposed to entail greater unemployment pressure, this ordering simply does not add up.

Where does the popular narrative make any sense?

Attributing the cause directly to AI is not without reason. The clues in the materials do indeed piece together a coherent explanation: generative AI capabilities are advancing rapidly, digital jobs such as software development have high exposure, and younger workers have less experience and are more easily replaceable, meaning entry-level positions are the first to be squeezed.

Model performance data also reinforces this intuition. A set of high-level results such as 78.1%, 91.3%, 92.7%, 94.3%, and 95.7%, as well as the presentation of GPT-3 175B set against a timeline since 2011, all provide context for the progression of model capabilities over time. For the general reader, this is enough to form a readily acceptable explanation: tools have grown more powerful, and entry-level jobs have dwindled.

A coherent narrative does not equate to established causality. A decline in employment can coincide with AI adoption and be amplified by it, but it must also align with other labor market indicators. Analyzing changes in unemployment grouped by AI exposure serves as a stress test.

Where is the anomalous number stuck?

If AI is the primary explanation, the most intuitive expectation would be: the more a job can be impacted by AI, the more its unemployment rate rises; the less susceptible a job is to AI, the less its unemployment rate rises. However, the data from 2022 to early 2025 does not show such a clear-cut trend: the unemployment rate increase for jobs with the lowest AI exposure was 0.94 percentage points, while for jobs with the highest AI exposure, it was 0.30 percentage points.

This does not mean AI has no impact. However, relying solely on “AI replacing junior programmers” to explain the decline in young programmers overlooks a broader context: during the same period, the labor market may also have been weathering the combined shocks of macroeconomic cycles, the interest rate environment, and the contraction of the tech industry. These alternative explanations may be underestimated.

Therefore, the more pertinent question is: regarding the decline in young programmers, is AI the primary cause, an accelerator, or simply a prominent concurrent factor? When the increase in unemployment is even higher in jobs with the lowest exposure, attributing the entire change to AI seems premature.

Two sets of signals from the same material

Employment data for software developers and customer service roles is described as exhibiting generational patterns, while other charts illustrate different facets of AI investment, model capabilities, and employment pressure. For instance, capital expenditure by hyperscale enterprises has more than doubled since the launch of ChatGPT, indicating that investment in AI infrastructure is indeed expanding rapidly.

A set of percentages—44%, 24%, 23%, 21%, 20%, and 20%—shows that the adoption of, exposure to, or impact of AI across different tasks or domains is not uniform. This unevenness affects judgment: if different job roles, age groups, and industry cycles are changing simultaneously, relying solely on the most conspicuous occupational group to make causal judgments can easily compress multiple forces into a single answer.

This is precisely what makes this angle so counterintuitive. The decline in young software developers and the advancement of AI capabilities can indeed be woven into a seamless narrative; however, the unemployment rate breakdowns in the very same report do not support jumping straight from “seemingly correlated” to “primarily caused by it.”

More Stable Reading Method

A more robust conclusion should be narrower in scope: employment among young software developers has seen a noticeable decline, and AI may be linked to this generational disparity; however, the assertion that “AI alone explains the significant decline in young programmers” is not yet conclusive.

This distinction is highly practical. If we only look at the nearly 20% decline among software developers aged 22 to 25 over two years, AI is easily assumed to be the sole cause; but if we then examine unemployment rates from 2022 to early 2025 grouped by AI exposure, the lowest-exposure group actually experienced a larger increase, meaning causal judgments must be drawn more slowly.

This material does not cover personal job search advice, nor does it offer country-specific prescriptions. What it does support is a boundary for judgment: the employment pressure on young programmers is a real issue, and AI is one of the key variables, but attributing all the causes to AI is not entirely commensurate with the material’s internal data.

---

Source institutions：Stanford Institute for Human-Centered Artificial Intelligence

This content is for reading and understanding research reports. It does not constitute investment advice or trading signals.