Software Developer Employment in BLS Data: 1.69M Up, Programmers Down 54%

The Bureau of Labor Statistics counted 1,687,890 software developers in May 2025 and 92,230 computer programmers. Six years earlier there were 199,540 programmers, so that occupation has lost 54% of its jobs. Software developer employment in BLS data rose over the same years. The split is real, but most of the programmer decline was already in the data before ChatGPT launched.

What the BLS OEWS count measures, and what it does not

The Occupational Employment and Wage Statistics program, OEWS, is the BLS survey that counts jobs by occupation. It asks employers how many workers they have in each detailed occupation and what those workers earn. It is the only federal source that publishes a national headcount for software developers, computer programmers and QA testers as separate lines.

That makes it the right place to start if you want a real number rather than a survey of developer sentiment. Vendor reports on developer jobs usually sample their own users. OEWS samples employers across every industry, and its May 2025 estimates were published on 15 May 2026 in news release USDL-26-0725.

Six panels, three years

The detail that changes how you read the series is the panel design. According to the OEWS technical note, each annual estimate combines six semiannual survey panels collected over three years, one in May and one in November. Wages from older panels are adjusted forward with the Employment Cost Index.

So the May 2025 figure is not a snapshot of May 2025. It blends survey responses from November 2022 through May 2025. The May 2022 figure blends responses from November 2019 through May 2022. That second window closes six months before ChatGPT launched on 30 November 2022, which is what makes the timing argument later in this post testable rather than rhetorical.

BLS itself warns about this. Its OEWS FAQ says the three-year method "limits the usefulness of OEWS data for time series analysis". I agree with the warning, and I still think a seven-year chart is useful, provided you read multi-year direction and ignore single-year wiggles.

Programmer, developer, tester: the definitions that decide the count

OEWS uses the 2018 Standard Occupational Classification, the federal list of occupation codes. The software family sits under the broad group "Software and web developers, programmers, and testers". Three of its detailed codes matter here, and their official definitions explain most of the chart before you look at it.

Quoted definitions are from the BLS OEWS May 2023 occupation pages. The QA row paraphrases the longer official text and is not a quotation. The 15-1256 row reflects the BLS note that the combined code "includes the 2018 SOC occupations 15-1252 Software Developers and 15-1253 Software Quality Assurance Analysts and Testers".

Read the first two definitions side by side. A programmer writes and tests code that someone else designed. A developer researches, designs and builds the software. In practice nearly every working engineer does both, so the code an employer picks depends on the job title in its HR system, not on the work.

That is the first reason I treat the programmer line as a classification story before an AI story. When employers retitle "programmer analyst" roles as "software engineer," the work stays and the code changes. The OEWS count follows the code.

Why 2019 and 2020 have one software number

BLS moved OEWS from the 2010 to the 2018 classification in stages. For May 2019 and May 2020 it published developers and testers together under the temporary code 15-1256, because the older survey data could not split them. The May 2019 combined table shows 1,406,870 jobs. Separate developer and tester counts start in May 2021.

Any chart claiming to show "software developers since 2019" as one clean line has either combined these codes or quietly mixed two definitions. This post uses the combined developer plus tester figure for the full seven years and the separate lines only from 2021.

Software developer employment in BLS data, 2019 to 2025

Here is every national OEWS employment figure for the three occupations. The 2019 to 2023 figures come from the BLS occupation tables, and the 2024 and 2025 figures from the news release tables. No figure in this table is estimated or interpolated.

Sources: BLS OEWS occupation pages for SOC 15-1251, 15-1252, 15-1253 and 15-1256, May 2019 to May 2023; OEWS news release Table 1 for May 2024 (USDL-25-0451, 2 April 2025) and May 2025 (USDL-26-0725, 15 May 2026). The 2021 to 2025 combined column is the sum of the two separate columns. Retrieved 1 November 2026.

Across the full window, programmers lost 107,310 jobs and developers plus testers gained 467,760. That is about 4.4 developer and tester jobs added for every programmer job removed. The three occupations together grew from 1,606,410 jobs to 1,966,860, a rise of 22.4%.

So the headline that coding jobs collapsed is wrong on this measure. The narrow programmer code collapsed, and the much larger family around it grew. The more useful question is when each line changed direction.

Computer programmer jobs fell 54%, and most of it predates ChatGPT

The computer programmer jobs decline is the most quoted number in this data, usually as evidence that AI is replacing coders. The timing does not support that reading for most of the drop.

The decline was steady before 2022

Between May 2019 and May 2022, OEWS programmer employment fell from 199,540 to 132,740. That is 66,800 jobs, or 62% of the full six-year decline. The annual drops were 10.7% in 2020, 14.3% in 2021 and 13.0% in 2022, all before any generative coding tool reached mass adoption.

Because of the panel design, the May 2022 estimate contains no survey responses from after May 2022. Whatever drove those 66,800 lost jobs, it was not ChatGPT. Offshoring of routine coding, the retitling described above, and the long shift from "programmer" toward "engineer" in job ads are all better candidates. The data cannot separate them.

The decline is also much older than this window. Press coverage of Current Population Survey data, a different household survey, put programmer employment at its lowest level since 1980, Fortune reported in March 2025. That CPS series peaked above 700,000 during the dot-com boom. A two-decade slide is not a two-year AI event.

2025 was the sharpest single-year drop

The case for an AI effect is not zero, and the most recent year is where it lives. Programmer employment fell 8.7% in 2024 and then 16.1% in 2025, from 109,870 to 92,230. That 2025 drop is the largest in the series.

The May 2025 estimate is also the first OEWS release built almost entirely from panels collected after ChatGPT launched. If AI coding tools are compressing the narrowest "write the code to spec" roles, this is where you would expect the first trace to appear. I count it as a real but early signal, one year deep, and I would want May 2026 data before calling it a trend.

The AI-era signal is the developer plateau, not the programmer decline

If you only remember one thing from this data, make it the shape of the software developer line. It is a far larger occupation, and its change in direction after 2023 says more about AI software engineer jobs than the shrinking programmer code does.

Developers rose 21.5%, then went flat

OEWS software developer employment rose from 1,364,180 in May 2021 to 1,656,880 in May 2023. That is 292,700 jobs in two years, a 21.5% gain built on survey panels from the 2020 to 2022 hiring boom. Then it stopped. The May 2024 count was 1,654,440, and May 2025 was 1,687,890.

A gain of 1.9% over two years, in an occupation that had been adding roughly 10% a year, is a regime change. The cause is harder to pin down. Interest rates rose sharply in 2022, many technology employers cut staff in 2023, and AI coding tools spread through the same window. All three push the same way, which is why this flat line is consistent with an AI effect without proving one.

The plateau also hides who is affected. A flat total can contain fewer entry-level hires and more senior ones, and OEWS does not split by experience. The payroll and job-posting evidence on that question is covered in the analysis of the junior developer pipeline. The payroll studies behind the AI job loss headlines are examined in what payroll data says about AI job losses.

QA testers fell two years running

The software QA analyst and tester count peaked at 203,040 in May 2023. It fell to 199,800 in 2024 and 186,740 in 2025, a cumulative 8.0% drop. Developers grew in both of those years, so this is not just a tech downturn.

I think testing is the more plausible place to look for automation pressure than programming. Generating test cases is one of the tasks coding assistants do best, though whether those tests catch real defects is a separate question, covered in the mutation score evidence on AI-written tests. A shrinking tester line with a stable developer line fits a world where developers absorb more of the testing work.

What wages say about the programmer and developer split

Employment counts tell you where jobs are. Wages tell you which jobs employers value. On that measure, the gap between the two codes has widened sharply.

Nominal dollars, not adjusted for inflation. Mean annual wages from OEWS occupation pages (2021 to 2023) and release Table 1 (2024 and 2025). Gap is a Zan Digital calculation.

The mean wage gap between developers and programmers grew from $24,340 in 2021 to $42,930 in 2025, an increase of 76%. Programmer mean pay actually fell between 2023 and 2024, from $107,750 to $103,640, while developer pay kept rising.

Medians tell the same story. The BLS Occupational Outlook Handbook reports a May 2025 median annual wage of $135,980 for software developers and $104,300 for QA analysts and testers. The equivalent programmer median was $100,390.

A falling headcount with flat or falling pay is what you see when the remaining roles in a code are the more routine ones. That supports the retitling reading: the work that commands design and judgement migrated into the developer code, and the programmer code kept the remainder. How the premium is spreading inside engineering itself is the subject of the AI engineer salary spread analysis.

Where this reading of the OEWS data is weakest

I have argued that the programmer decline is mostly classification and that the developer plateau is the more important signal. Here is where that argument is most exposed.

The rolling panel smooths and lags

Every OEWS estimate averages three years of responses. A sharp change in 2025 hiring shows up only partly in the May 2025 number and keeps feeding into the next two releases. The developer plateau could therefore be milder than reality, or the start of a deeper fall that the method is still averaging away. The same smoothing means the 2021 to 2023 boom was partly a 2020 to 2022 boom.

BLS also notes that a federal funding lapse from 1 October to 12 November 2025 added collection and processing time for the May 2025 panel. I found no sign that it biased the counts, but it is one more reason not to over-read a single year.

Coding work moved, it did not stop being counted

OEWS counts wage and salary jobs by occupation code. It does not count the self-employed, and it does not count code written by people whose title sits in another occupation. Data scientists, web developers, analysts and product managers all write code that never touches the programmer or developer lines.

So the strongest objection to my "classification, not collapse" view is that I cannot prove the retitling happened. The employer-level reclassification records that would show it are not public. A competing reading, that offshoring removed a large share of these roles and they never came back under any code, fits the same aggregate numbers. Separately, the CPS household survey and OEWS disagree on levels because they ask different people different questions, a problem examined in why entry-level hiring data conflicts.

What the BLS software developer count means for your hiring plan

If you run an engineering team or a services business, you are not trying to settle the AI jobs debate. You are deciding whom to hire next year, and the OEWS data gives you three practical signals.

First, job titles are a weak proxy for skill. The programmer code is shrinking because employers stopped using the word, not because writing code stopped mattering. When you benchmark compensation, match on responsibilities, not titles, or the $42,930 mean wage gap will mislead you in both directions.

Second, the developer market has stopped growing. After two years of near-zero net growth, you are hiring from a pool that is no longer expanding by 10% a year. Mid-career candidates who design systems are not getting cheaper, and the wage line shows it.

Third, decide deliberately who owns testing. The tester decline suggests many teams have already shifted testing onto developers, intentionally or not. If your team did the same, put a measure on test quality before the gap shows up in production incidents. A useful scorecard for that is in the developer productivity measurement guide.

For career development, software engineers should read the same chart differently. The safe position in this data is the developer code's job description, research and design, not its title. Roles defined only as "write code to a spec" have been shrinking for six years, with or without AI.

Frequently asked questions

Where to start

Pull your own team's job titles and map each one to the three SOC codes above. If your HR system still calls people "programmer analysts" while you pay them as engineers, your compensation benchmarks are comparing against the wrong line.

Then put a reminder in your calendar for the May 2026 OEWS release, due next spring. Check one number first: whether software developer employment breaks out of its two-year plateau or turns down. That single line will tell you more about AI and engineering headcount than any vendor survey published between now and then.

Related reading