Quick answer
Not detectably. This is the age mix of people who hold developer jobs, not a count of jobs. In Census data, the share of California software developers aged 22-27 went from 15.2% (2019, 2021-22) to 14.3% (2023-24), -0.9 points: no detectable change. A fall of 1 to 2 points cannot be ruled out. Data end in 2024.
Key takeaways
- What was measured: the age mix of employed Californians aged 22-64 who report an occupation coded software developers (SOC 15-1252), in Census ACS survey files. No public dataset measures AI agents as such, and this one does not measure AI.
- The share of those developers aged 22-27 shows no detectable change between 2019 and 2021-22 and 2023-24 (figures and caveat in the first section).
- The share of employed Californians aged 35-49 and 50-64 who are developers did not fall. In every age band the share who are developers was higher in 2023-24, though for ages 22-27 the difference is within sampling noise.
- The data end in 2024, count only people who have a job, and say nothing about hiring, layoffs, seniority or why anything moved.
Are there fewer young software developers in California since AI coding tools arrived?
Not detectably, and this is the age mix of people who hold developer jobs, not a count of jobs: the share of California software developers (SOC 15-1252) aged 22-27 went from 15.2% in 2019 and 2021-22 to 14.3% in 2023-24, a change of -0.9 percentage points, which is no detectable change.
A survey is a sample, so each change comes with a range of values it could plausibly produce; this one runs from -2.3 to 0.4 points and includes zero. The survey could detect a change of about 1.9 points, so a fall of 1 to 2 points cannot be ruled out. This is not proof that younger developers were unaffected by anything.
A person who loses a developer job leaves the developer count, so the age mix can stay flat while the number of jobs changes. What was measured: of employed Californians aged 22-64 who report working as software developers, what share are in each age band. No public dataset measures "AI agents" as such, and this one does not measure AI at all.
What does the age mix of California developers look like?
About 15% of California software developers were aged 22-27 in the earlier years and about 14% in 2023-24; the other bands look much the same. Shares are rounded. By this measure California's developer workforce did not get measurably older through 2024.
| Age band | Share of California software developers, 2019, 2021 and 2022 together | Share, 2023-24 |
|---|---|---|
| 22-27 | 15.2% | 14.3% |
| 28-34 | 29% | 30% |
| 35-49 | 40% | 40% |
| 50-64 | 16% | 16% |
Read from the other side, the share of all employed Californians in each band who are developers was higher in 2023-24 in every band, including ages 22-27 (2.3% in the earlier years and 2.5% in 2023-24), though for 22-27 the difference is small enough that sampling noise could produce it (its range includes no difference). We did not rank the bands, because the differences between them are not distinguishable from each other.
Did the share of older Californians who are developers fall?
No: the share of employed Californians aged 35-49 and 50-64 who are developers did not fall. For ages 35-49 it was 2.3% before and 2.5% in 2023-24; for ages 50-64 it was 1.1% before and 1.3% in 2023-24. That is a different statement from "older developers were unaffected", which these data cannot support. It says nothing about hiring, layoffs or any role for AI.
Does this confirm or contradict the national payroll study?
It neither confirms nor contradicts our earlier article on payroll jobs for developers by age, because the two measure different things. That article follows an index of payroll employment levels for the whole country, month by month to August 2026, for ages 22-25. This one counts who holds a developer job in California, by age, among employed people who answer a survey, year by year to 2024, for ages 22-27.
The age mix of those still employed as developers can stay flat while the number of jobs changes, as the first section explains. We print no figures from the two studies side by side because they do not share a measure, a place, an age band or a period.
What did we measure, and how?
We tabulated Census Bureau survey records, comparing 2019, 2021 and 2022 together with 2023 and 2024. The source is the American Community Survey 1-year Public Use Microdata Sample for California, covering employed Californians aged 22-64. Records are weighted to represent the state.
We expected fewer young developers and no fall among older ones, and ran the tests to find out. The direction of three of our six tests was visible to us before we froze the method, because we had already seen a national payroll index and a sample count, so these were not blind tests. Comparison groups are all employed Californians aged 22-64 and "other computer occupations" (SOC 15-1xxx except 15-1252).
One check was exploratory: it used only 2019 as the earlier period. For software developers, the share aged 22-27 changed by -2.0 percentage points (range -4.0 to 0.0). The comparison groups moved by about as much: other computer occupations changed by -2.3 points and all employed Californians by -1.2 points. 2019 (16.3%) was the highest of the three earlier years, above 2021 (13.9%) and 2022 (15.5%), so year-to-year noise is as large as the change, and this check does not show a fall specific to developers.
Technical detail (skippable): the data hold n = 786,954 records of employed Californians aged 22-64, of which n = 19,384 are coded software developers (n = 10,989 in 2019, 2021 and 2022; n = 8,395 in 2023 and 2024). For the -0.9 point change, the 95% CI was -2.3 to 0.4 (p = 0.167 unadjusted, Holm-adjusted p = 0.501 across six tests); the design effect was 1.9, meaning the survey's design makes the estimate noisier than a simple random sample. The change for developers was 0.4 points lower than the change for all employed Californians (-0.5 percentage points, 95% CI -0.7 to -0.4): the gap was -0.4 percentage points (95% CI -1.7 to 0.9; Holm p = 0.875), not distinguishable from zero. The odds of being a developer, 2023-24 against the earlier years: ages 22-27 odds ratio 1.10 (95% CI 0.99 to 1.22, which includes 1; Holm p = 0.298), ages 35-49 odds ratio 1.09 (95% CI 1.03 to 1.15) and ages 50-64 odds ratio 1.18 (95% CI 1.09 to 1.29). For the two older bands we tested whether the odds ratio was above a margin we set in advance with the direction visible; both passed (Holm p < 0.001 each). The band differences are not distinguishable from each other.
What can this data not show?
It cannot show hiring, layoffs, seniority or any role for AI, and it stops before much of the period your question is about.
- The window ends in 2024. The data stop before much of 2025 and 2026: the 2025 file was not available when we ran this analysis (October 2026). The Bureau did not release standard 2020 1-year estimates, citing the pandemic's disruption of data collection, so there is no standard 2020 file. 2021 was collected in the same pandemic period; we cannot say whether that affected it, which is one reason we ran the 2019-only check.
- Only people with a job are counted. Unemployed people and those out of the labor force are invisible: a young developer who lost a job and has not found another is in neither the developer count nor the employed total.
- Age is not seniority. A 25-year-old can lead a team and a 45-year-old can be new to the field.
- Occupation is self-reported and coded by Census. "Software developers (SOC 15-1252)" means what respondents described and the Bureau coded, which can differ from an employer's job title.
- California only, ages 22-64, civilian employed. Nothing here applies elsewhere.
- It is a before/after association. Many other things changed over the same years, and this comparison cannot separate any of them from AI. Nothing in the data identifies AI or agents.
How can you check it yourself?
Download the California person files from the Census Bureau and tabulate them. The 1-year ACS PUMS files for 2019, 2021, 2022, 2023 and 2024 are in each year's "1-Year" folder of the Census PUMS directory, linked in the sources. The 2024 data dictionary, also linked, shows that occupation code 151252 is software developers.
Steps: keep employed people (ESR codes 1 and 2) aged 22 to 64; split them into the four age bands; use each year's person weight (PWGTP) to compute the share of software developers in each band; and pool 2019, 2021 and 2022 against 2023 and 2024. The Census "Accuracy of the Data" statement describes the replicate weights used for the margin of error. We do not offer a data file with this article. For a different Census-based look at who uses AI, see AI use by business size and industry.
What can you do with this?
Before you repeat a headline about "fewer young developers", ask four things: is it a count of jobs or a share of workers, which ages does it use, does the period reach 2025-26, and does it measure AI or only the timing?
To see what is on offer in your own area now, look at current local job postings and the BLS occupation tables for your metro. Software job postings on Indeed show openings on one job site, and Sales and marketing occupations vs software occupations uses federal occupation tables by metro. This article's figures end in 2024 and describe an age mix, so they do not tell you whether candidates in their twenties are harder or easier to find.
Analysis by Deivy Hernandez, from public data files.
FAQ
Does this show AI is not affecting young software developers?
No. The data have no measure of AI, so they cannot show an effect or the absence of one. They show that the age mix of people employed as software developers in California had no detectable change by 2023-24 compared with earlier years.
Can I see 2025 or 2026 figures?
Not from this analysis. The 2025 file was not available when we ran it (October 2026), and there is no standard 2020 file at all. Check the Census PUMS page for newer 1-year files before relying on this article's end date.
What counts as a software developer here?
A person who describes their job in the survey in a way the Census Bureau codes as software developers, occupation code SOC 15-1252. It is the respondent's own description, so it can differ from a job title on an employer's payroll.
Can I re-run the analysis myself?
Yes, from the public Census files listed in the sources. We do not offer a data file with this article; the Census files hold the person-level records, and the method is described in the method section.