Quick answer
Not quite. In one national ADP payroll index, software developers aged 22-25 are about 19% below their Nov 2022 level and every age band from 31 up is above it, so it is not only the youngest. The youngest bands had risen faster just before that base month, and the data cannot say why.
Key takeaways
- What was measured: an employment index (Nov 2022 = 100) of US software developers by age band, built from one payroll processor's clients. It is national, not a count of workers, and it does not measure AI.
- Ages 22-25 averaged 81 in the 12 months to Aug 2026 (-19%, monthly range 78 to 84); ages 26-30 averaged 97 (-3%); ages 31 and up were all above 100.
- The shape is a hump: 41-49 is highest (+19%), 50 and over is lower (+8%). That does not support 'only juniors affected, seniors untouched'.
- Age is not seniority, the Nov 2022 base date was the source authors' choice, the young bands had risen slightly faster just before it, and nothing here identifies an explanation.
Is AI taking junior developer jobs while senior developers are unaffected?
Not quite. In one national payroll index, software developers aged 22-25 are about 19% below their Nov 2022 level, while every age band from 31 up is above it, so the pattern is not only about the youngest, and the data cannot say why. The 26-30 band is slightly below its base, and the highest band is 41-49, not the oldest. Every gap is measured from a base month the source chose, and the youngest bands had risen slightly faster just before it.
The limits come first. What we measured is an employment index (Nov 2022 = 100) of software developers on the payrolls of one payroll processor's clients, national, split into six age bands. It is not a count of workers, not California, and not a measure of AI or agents; no public dataset measures "AI agents" taking jobs. Age is not seniority.
What did we measure, and how?
We read a published monthly index by age band and compared the last 12 readings (Sep 2025 to Aug 2026) with the Nov 2022 base. The index comes from ADP payroll records, aggregated by Stanford Digital Economy Lab researchers for their "Canaries in the Coal Mine" dashboard (release vintage 2026-09-17). Each of the six series has 60 monthly readings, from Sep 2021 to Aug 2026. The authors set Nov 2022 equal to 100.
The source publishes no margin of error, so small differences between bands should not be read into. We started from the assumption that "AI agents are only affecting junior programming roles; senior profiles have not been affected." We saw the direction (young bands down, older bands up) before we fixed our method, so this is a check in form only, not a blind test. The article reports what the data showed, including where it contradicts that belief.
What does the index show for each age band?
Employment of developers aged 22-25 averaged 81 in the 12 months to Aug 2026, about 19% below its Nov 2022 level, while every band from 31 up was above it. "Index" is the average of the last 12 monthly readings, where 100 is the Nov 2022 level; "range" is the lowest and highest single month, not a confidence interval.
| Age band | Index, Sep 2025 to Aug 2026 | Versus Nov 2022 | Range of monthly readings | Aug 2026 reading |
|---|---|---|---|---|
| 22-25 | 81 | -19% | 78 to 84 | 80.5 |
| 26-30 | 97 | -3% | 94 to 99 | 93.9 |
| 31-34 | 108 | +8% | 106 to 110 | 105.5 |
| 35-40 | 112 | +12% | 111 to 113 | 111.7 |
| 41-49 | 119 | +19% | 117 to 120 | 119.5 |
| 50 and over | 108 | +8% | 108 to 109 | 107.9 |
The 22-25 band's index was about 19% below its Nov 2022 level in the 12 months to Aug 2026 (range of monthly readings 78 to 84). It never regained 100 after that month: its highest reading from Dec 2022 to Aug 2026 was 98. Nov 2022 is the source's base month, and in the 14 months before it this band had risen slightly faster than the oldest one (see the limits below).
Technical detail: the "versus" column is the percent difference between each band's 12-month mean and its Nov 2022 level of 100.
Is it a clean line from young to old?
No. The pattern is a hump, not a ladder: the 41-49 band is highest at +19%, and the 50-and-over band is lower at +8%, the same as 31-34. Listed plainly: 22-25 is at -19%, 26-30 at -3%, 35-40 at +12%, and every band from 31 up is above its base. The youngest band had also risen faster just before the base month; but "older is better" would be the wrong summary, since the oldest band does less well than the one before it.
Technical detail: a rank correlation of rho = 0.83 between band order and change rests on six bands only and hides the hump, so it is not a straight line.
Does this support "only juniors are affected"?
No, not "only". The youngest band is down, the older bands are up, the 26-30 band is slightly down, and the peak is at 41-49, not at the oldest band. In the table, every band from 31 up is above its base. That is not an "only juniors" pattern, and the data hold ages, not job levels. It also shows nothing about why any band moved, so whether AI agents play any part is a question these data cannot reach.
What can the data not show?
They cannot show why any band moved; other developments of the same period are possible context. Read these limits together, because each one changes how far the table can be pushed.
- Age is not seniority. A 45-year-old can be a junior developer, and the index carries no job titles. Calling a band "juniors" or "seniors" is the reader's shorthand, not something measured.
- The base date is the source's choice, and the young bands had risen faster just before it. The authors set Nov 2022 equal to 100, and every figure is relative to it. From Sep 2021 to Nov 2022 the 22-25 index rose from 88 to 100, about +13%, against about +9% for the 50-and-over band (91 to 100). A small part of the later gap is that earlier rise reversing.
- No explanation is identified. Over the same period the Federal Reserve raised its policy interest rate from March 2022 to July 2023, and from tax years beginning after December 31, 2021, US law required software-development costs to be treated as research expenses (the Section 174 change). The index itself had risen through 2021-22 before the base month, so some of what follows may be that rise unwinding. We name them as context only; nothing here tests whether they, AI tools or anything else relate to any band's index.
- One payroll processor, national. The index covers firms that use ADP. It leaves out the self-employed, workers under 22 and every company on another provider. It says nothing about California or any one state, and it is not a count of workers.
- People age into and out of bands, and the readings are few. A 30-year-old moves to the next band a year later, and the data cannot separate that from real changes in employment. Months in a series also move together, so six series of 60 months carry far less information than they look, which is why no confidence intervals or p-values are printed.
How can you check it yourself?
Open the Stanford Digital Economy Lab dashboard and download the software developers file from the release. The dashboard is at the source below. We do not republish the data: the page states no licence, so this article prints figures only and gives no data file.
The file is the software developers index, vintage 2026-09-17, with one column per age band and one row per month from Sep 2021 to Aug 2026. Take the last 12 rows, average each column and compare with 100. The Census-based piece on how many small businesses use AI takes the same cautious approach to a different question.
What does this mean if you hire or work in software?
Treat it as one payroll index, not as a measure of how easy it is to hire a developer or of anyone's skill. It says nothing about an individual applicant, whatever their age. If you want to see what AI agents can and cannot do in a small business, our guide to what AI agents can and cannot do for a small business covers what is sold as hands-off and what still needs a human check.
Analysis by Deivy Hernandez, from public data files.
FAQ
Does this show AI is taking junior developer jobs?
No. The data show that an employment index for software developers aged 22-25 is well below its Nov 2022 level. They do not say why. Other things happened over the same period, such as higher interest rates and a change in how US companies deduct software-development costs, and the data cannot separate them from any other explanation.
Is it harder to get hired as a junior developer now?
These data cannot say. They measure employment levels in one payroll index, not job openings, hiring or applications. A lower index for ages 22-25 is not the same as fewer offers or more rejections.
Are developers aged 41-49 safe, then?
The data cannot say that about any person or company. The 41-49 band's index was higher than its Nov 2022 level as a group, in one payroll processor's client base. It says nothing about your own job or any one employer.
Why use age bands instead of junior and senior?
Payroll records carry age, not job level. A 45-year-old can be a junior developer and a 27-year-old can lead a team, so an age band is only a rough stand-in for experience.
Where can I see the data?
On the Stanford Digital Economy Lab's Canaries in the Coal Mine dashboard page, listed in the sources below. The software developers file is in the release downloads. We print figures only and do not republish the file.