United States: AI Slows Growth of Programming Jobs

Programmers in the United States are facing a radical transformation in their sector, largely influenced by the rise of generative AI. A recent study by the Federal Reserve Board highlights the notable repercussions of this technology on employment in the programming field.

The study relies on monthly employment data from the United States, cross-referenced with a database from the Department of Labor, to analyze professions heavily focused on programming. These professions represent approximately 3.7% of the American workforce.

Decline in Programmer Job Growth

Before the arrival of ChatGPT in November 2022, programming jobs were growing at an annual rate of less than 5%, outpacing the overall labor market. Since then, this growth has notably slowed, particularly in IT services and software development, where it has stabilized.

To distinguish the impact of AI from other economic factors, researchers developed a counterfactual employment curve. This shows that even after adjustment, programmer employment continues to decline by about three percentage points per year. Companies appear to be deliberately reducing the proportion of programmers in their workforce.

Impact on the Job Market

Over three years, this trend represents approximately 500,000 jobs that have not been created. However, this does not necessarily mean job losses. Many potential programmers have likely shifted to related fields, and AI could redistribute tasks among different job categories.

Programmer salaries have not experienced a significant decline. Data from job sites indicate that job postings for software developers have remained stable since 2024, following a drop of more than half in 2022 and 2023.

An Industry in Transition

About 40% of American programmers work for IT service providers rather than large tech companies. This is where the slowdown is most pronounced.

The gap between actual employment and expected employment widened starting mid-2024, about 1.5 years after the launch of ChatGPT. Companies have taken time to adjust their workforce in line with the productivity gains expected from AI.

Measurement Challenges and Outlook

The study emphasizes the difficulty of measuring the impact of AI on professions. There is no standard method for identifying the professions most affected by generative AI. However, programmers are universally recognized as among the most impacted.

Researchers believe their work is just a first step in understanding how AI could transform the labor market in the long term. Further studies are needed to assess whether AI will open new markets or redefine other professions.

Programmers are among the workers whose daily lives have changed the most due to generative AI. A new study from the Federal Reserve Board shows that this change is also reflected in employment figures.

The study analyzes monthly employment data in the U.S. from a large household survey and cross-references it with a professional database from the Department of Labor that classifies jobs based on their requirements and skills. This allows researchers to identify professions with a high share of programming work. This group represents about 3.7% of all American workers.

Growth Rate Almost Halved

Before the launch of ChatGPT in November 2022, programming-focused jobs in the U.S. were growing at a rate of less than 5% per year, well above the overall labor market. Since then, this rate has dropped significantly. In sectors with a particularly high share of programmers, such as IT services and software development, growth has essentially stabilized.

The obvious argument: was it not simply due to the pressure faced by the entire tech sector in 2022? Rising interest rates, the end of the Covid-related online services boom, and the collapse of cryptocurrencies all hit the industry at the same time.

To eliminate these sectoral effects, researchers constructed a counterfactual employment curve. It shows how many programmers there would have been if their share within each industry had remained constant and only the size of the industries had changed.

Even after this adjustment, programmer employment continues to decline by about three percentage points per year. It seems that companies are deliberately reducing the share of programmers in their workforce. A control test using professions barely affected by AI shows no comparable decline.

Fewer than 500,000 Jobs, but Not Necessarily Job Losses

Spread over three years, the gap represents about 500,000 jobs that would likely have existed without the rise of language models. However, the authors caution against interpreting this figure as a simple count of lost jobs.

Many potential programmers have likely found work in adjacent fields. AI could redistribute tasks among job categories, with programming skills now appearing more in other roles. The study also does not capture broader macroeconomic feedback effects. If AI sufficiently increases productivity, the total demand for labor could even grow in the long term.

The researchers also found no clear decline in salaries. So far, the effect has mainly manifested in the number of jobs filled rather than in compensation. Data from job sites show that job postings for software developers have remained largely stable since 2024 and have slightly increased in recent months. Before that, they had dropped by more than half in 2022 and 2023.

In the long term, software-related sectors and other programming-rich sectors have experienced faster growth than the overall labor market. The information sector as a whole only partially reflects this trend.

The Typical Programmer is Not in Silicon Valley

One finding concerns the structure of the industry: about 40% of all American programmers do not work for large tech companies or startups, but for IT service providers that develop software on a contract basis. This contract sector is by far the largest employer of programmers, and this is where the slowdown is most pronounced.

According to the study, the gap between actual employment and expected employment only opened up starting mid-2024—about 1.5 years after the launch of ChatGPT. If AI is behind the change in hiring plans, companies seem to have needed time to see the patterns improve before reducing their workforce. The data do not specify whether real productivity gains or simply the expectation of them motivated this decision.

AI may not be the only factor at play. A provision of the Tax Cuts and Jobs Act of 2017, which took effect in 2022, requires companies to amortize research expenses over several years instead of deducting them immediately.

Since software development is considered research for tax purposes, this could have dampened hiring. However, existing research on its actual impact is mixed, and the study's results hold even in sectors where the tax change should matter less.

The authors acknowledge that it is difficult to determine a causal effect with so many factors at play. They validate their approach using past disruptions: for tellers, the method correctly identifies ATMs as a profession-specific shock, even though the banking industry as a whole has grown. For seamstresses, it correctly identifies an industry-wide shock due to the relocation of textile manufacturing abroad.

Researchers Disagree on the Professions Most Affected by AI

The study also highlights a fundamental measurement problem in this area. There is still no standard method for determining which professions are most affected by generative AI. Regarding the identification of the most affected group, only about half of the common measurement approaches align with each other.

Programmers are an exception: over 98% of them fall into the most affected group according to every measurement method. This corresponds to data from the Anthropic Economic Index, which shows that programming-related queries account for more than a third of all interactions with the chatbot Claude.

Whether the trend will reverse over time as cheaper programming services open new markets, whether offshore workers will be more significantly impacted, or whether AI will ultimately reshape other professions entirely remains uncertain. The authors describe their work as "only a first step" toward answering these questions.

A recent study from Carnegie Mellon and Stanford revealed that the development of AI agents is almost entirely focused on programming tasks, while economically more significant areas are barely considered.