CPS data indicates that US software-developer employment has increased since 2022

2026-10-06

Released under the CC-BY license

Note: I did not use generative-AI tools for any part of this project, including research; coding and debugging; visualizations; and this blog post.

Introduction and methodology

In order to assess whether software-developer employment has been increasing or decreasing since ChatGPT was released in 2022, I downloaded Current Population Survey data from IPUMS, then analyzed it in Python using the svy library. (The CPS was also the survey of choice for a similar analysis conducted by James Bessen at Boston University; his work helped inspire my own analysis project. My analysis focuses only on the 'Software developers, applications and system software' occupation category, and not on other related fields, such as web developers and computer programmers.

The strata, PSUs, and weights that I used for my survey sample were based on guidance from Ivan Strahof at IPUMS, whose recommendations were in turn based in part on a 2007 research paper by Michael Davern et al. (I may tweak my strata a bit in the future in order to divide Delaware results by county; this tweak, however, should not change the substance of my findings.)

To see visualizations of my employment-total calculations not covered in this blog post, visit this interactive dashboard. Also, if you'd like to review and/or repurpose the code that I used for this analysis, you can find it within this GitHub project (which I've released under the MIT license).

Findings

Yearly employment trends

My analyses indicate that software-developer employment has increased from around 2 million in 2022 to over 2.5 million in 2026. (2026 data is current through August.)

US software-developer employment by year
_{Error bars represent 95% confidence intervals}

Source: Current Population Survey data from IPUMS.
This analysis is based only on the "Software Developers, Applications and
Systems Software" category for IPUMS USA's occ2010 variable.
2026 data includes January through August only.
By Ken Burchfiel. Released under the CC-BY license.
GitHub project: https://github.com/kburchfiel/employment_trends

This chart displays 95% confidence intervals together with point estimates. Note that the intervals for 2022 and 2026 do not overlap, meaning we can be confident that the increase in employment over that time is statistically significant.

Note: As with my interactive dashboard, you can can hover over points on the charts within this blog post in order to view actual confidence-interval and point-estimate values.)

Monthly employment trends

Since the CPS is normally administered each month, we can also assess trends on a monthly level. These should be interpreted with caution, as monthly confidence intervals will be larger than yearly ones. However, the upper bound of the confidence interval for November 2022 (2,289,355) within the following chart is still lower than the lower bound of the August 2026 confidence interval (2,419,564).

As a result, we can conclude that software-developer employment has increased from ChatGPT's release on November 30, 2022 to August 2026.

US software-developer employment by year and month
_{Error bars represent 95% confidence intervals}

It's challenging to use CPS data to draw conclusions about employment trends for a given occupation and age range, as the number of respondents for a given month who fall into a particular age range/occupation group may be quite small. Nonetheless, I thought it would be interesting to compare CPS estimates by age in order to see whether employment growth has been particularly large or small for certain groups.

Here's a look at the data. (The overlapping error bars make this chart quite hard to read, but you can select a single age range by double-clicking on its legend entry. In addition, clicking on an active entry will deselect it.)

US software-developer employment by year and age range
_{Error bars represent 95% confidence intervals}

For two age ranges (30-34 and 40-44), the lower confidence interval of the 2026 estimate is higher than the upper confidence interval of the 2022 estimate, which indicates that these two age ranges have seen a significant increase in software-developer employment since 2022. For all other age ranges, we can't conclude that a significant increase or decrease has occurred.

Here's a copy of the same chart that does not include error bars:

US software-developer employment by year and age range

Assessing changes in the percentage of Americans who are employed as software developers

Another way to evaluate software-development's growth is to estimate the percentage of Americans who are employed as a developer. This approach adjusts for growth in the US population; in addition, it allows for logistic-regression tests that allow us to assess whether two differences in proportions are statistically significant.

The average percentage of Americans employed as software developers for January through August 2026 is 1.33%. This is significantly higher than the percentage employed in each year from 2022 (when it was 1.14%) through 2025.

% of Americans aged 22-65, by year,
who are employed as software developers
_{Error bars represent 95% confidence intervals}

The following chart compares estimated employment percentages by age range. In order to test these changes for statistical significance, I performed a logistic regression in which the 2026 data was compared to that for years 2022 onward. I found that, for four age ranges (30-34, 35-39, 40-44, and 55-59), the percenatge of Americans employed as software developers has increased significantly relative to 2022. In contrast, while the share of those aged 22-24 who are employed as software developers grew markedly from 2010 to 2022, it has been relatively flat since then, suggesting that the job market has indeed been more challenging for recent graduates.

% of Americans, by year and age range,
who are employed as software developers
_{Error bars represent 95% confidence intervals}

The following chart is a copy of the previous one with the error bars removed.

% of Americans, by year and age range,
who are employed as software developers

What about unemployment rates?

Even if software-developer employment is growing, it's possible that unemployment rates are growing as well. This would be the case if the supply of potential software developers—i.e., due to increased numbers of computer/information-sciences graduates—grows even faster than the demand for developers

It's difficult to calculate reliable unemployment rates for software developers using CPS data, as both the percentage of respondents who are (or had been) employed developers, and the percentage of such respondents who are unemployed, are quite small. Nevertheless, here's what these unemployment-rate estimates look like:

Estimated percentage of software developers who are unemployed
_{Error bars represent 95% confidence intervals}

This chart indicates that unemployment for software developers is higher now than in the pre-LLM era. Indeed, I found via a logistic regression that the 2026 unemployment rate is significantly higher than that for 2015 through 2019; 2021; and 2022. However, it's also significantly lower than it had been in 2010, when the country was still emerging from the 2008 financial crisis.

It's important to consider these unemployment statistics in the context of other occupations, though. The estimated unemployment rate for 2026 to date (around 3%) currently compares favorably to that for all civilians (4.1% as of August 2026).

Conclusion

These analyses suggest that the overall trend in software-developer employment since 2022 has been positive, particularly for those ages 30-34 and 40-44. This growth in employment, however, has coincided with an increase in the unemployment rate for developers. Thus, while we don't have evidence that LLMs have caused the total market for developers to shrink, this doesn't mean that finding a job as a developer is necessarily easy.

It's also worth noting that the fact that the number of developers has increased in recent years does not prove that the introduction of LLM tools hasn't negatively impacted developer employment. It's quite possible, that, without the availability of these tools, employment may have increased even more quickly. (See this Stanford Digital Economy Lab page for more details on this hypothesis.)

These analyses should be interpreted with caution, given that the CPS's sample size is somewhat limited. In the future, I hope to compare these results to those generated via American Community Survey data. The ACS has a much larger sample size, and thus can allow for more precise estimates than those we can obtain through the CPS. However, as of October 2026, the latest data for which we have ACS data is 2024--which, given the changes in AI adoption since then, is too far in the past to be a useful comparison. Once ACS data for more recent years becomes available, I certainly hope to perform a similar analysis using those survey results.

References

IPUMS CPS citation:

Sarah Flood, Miriam King, Renae Rodgers, Steven Ruggles, J. Robert Warren, Daniel Backman, Etienne Breton, Grace Cooper, Julia A. Rivera Drew, Stephanie Richards, David Van Riper. Integrated Public Use Microdata Series, Current Population Survey: Version 13.0 [dataset]. Minneapolis, MN: IPUMS, 2025. https://doi.org/10.18128/D030.V13.0